diff --git "a/related_53K/test_related_long_2404.16698v1.json" "b/related_53K/test_related_long_2404.16698v1.json"
new file mode 100644--- /dev/null
+++ "b/related_53K/test_related_long_2404.16698v1.json"
@@ -0,0 +1,8628 @@
+[
+    {
+        "url": "http://arxiv.org/abs/2404.16698v1",
+        "title": "Cooperate or Collapse: Emergence of Sustainability Behaviors in a Society of LLM Agents",
+        "abstract": "In the rapidly evolving field of artificial intelligence, ensuring safe\ndecision-making of Large Language Models (LLMs) is a significant challenge.\nThis paper introduces Governance of the Commons Simulation (GovSim), a\nsimulation platform designed to study strategic interactions and cooperative\ndecision-making in LLMs. Through this simulation environment, we explore the\ndynamics of resource sharing among AI agents, highlighting the importance of\nethical considerations, strategic planning, and negotiation skills. GovSim is\nversatile and supports any text-based agent, including LLMs agents. Using the\nGenerative Agent framework, we create a standard agent that facilitates the\nintegration of different LLMs. Our findings reveal that within GovSim, only two\nout of 15 tested LLMs managed to achieve a sustainable outcome, indicating a\nsignificant gap in the ability of models to manage shared resources.\nFurthermore, we find that by removing the ability of agents to communicate,\nthey overuse the shared resource, highlighting the importance of communication\nfor cooperation. Interestingly, most LLMs lack the ability to make\nuniversalized hypotheses, which highlights a significant weakness in their\nreasoning skills. We open source the full suite of our research results,\nincluding the simulation environment, agent prompts, and a comprehensive web\ninterface.",
+        "authors": "Giorgio Piatti, Zhijing Jin, Max Kleiman-Weiner, Bernhard Sch\u00f6lkopf, Mrinmaya Sachan, Rada Mihalcea",
+        "published": "2024-04-25",
+        "updated": "2024-04-25",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Original Paper",
+        "paper_cat": "LLM Fairness",
+        "gt": "AI Safety As LLMs become more capable and autonomous, ensuring their safety remains a critical concern (Amodei et al., 2016; Hendrycks et al., 2021; Anwar et al., 2024). Although traditional evaluations often use standard datasets such as ETHICS (Hendrycks et al., 2020), TRUTHFULQA (Lin et al., 2022), and MORALEXCEPTQA (Jin et al., 2022), these methods fall short in addressing the complexities inherent in multi-agent interactions and broader real-world scenarios. Furthermore, while LLM agents are a relatively recent development whose applications extend well beyond simple chatbot functionality, the majority of existing research has primarily evaluated these agents in specific domains such as information retrieval and software development (Zhou et al., 2023; Liu et al., 2023; Jimenez et al., 2023; Deng et al., 2024). Most similar to our GOVSIM are MACHIAVELLI (Pan et al., 2023) and GTBENCH (Duan et al., 2024), which extend evaluations to scenarios involving strategic interactions and game-theoretic reasoning, respectively. In MACHIAVELLI they investigate harmful behavior vs. reward maximization in a benchmark of single-agent choose-your-own-adventure games. In GTBENCH they evaluate agents on game-theoretic reasoning. In contrast, our GOVSIM focuses on multi-agent scenarios that require 2 Agents Home Harvesting Discussion Figure 1: Overview of the GOVSIM simulation environment. The simulation unfolds in various stages. Home: agents plan for future rounds and strategize their actions based on past rounds. Harvesting: agents collect resources, like fishing. Discussion: agents convene to coordinate, negotiate, and collaborate. both strategy, communication and cooperation: it simulates a real-world multiround cooperation dilemma, thus introducing a more dynamic and realistic environment. Simulacra with LLMs The integration of LLMs into simulations that mimic social interactions or complex decision-making scenarios has been a growing area of interest (Park et al., 2022, 2023). These studies range from virtual societies (Lin et al., 2023; Wang et al., 2023; Kaiya et al., 2023; Hua et al., 2023) to task-specific agent collaborations (Hong et al., 2023; Nair et al., 2023; Zhang et al., 2023; Li et al., 2024). Simulation studies with LLMs have focused on pure game environments (Akata et al., 2023; Shi et al., 2023; Guo et al., 2023; O\u2019Gara, 2023), such as Werewolf (Xu et al., 2023). They have also extended them to scenarios with economic grounding (Zhao et al., 2021) and history (Hua et al., 2023). Our work in GOVSIM leverages the Generative Agents framework to explore multi-agent interactions to achieve cooperative norms, towards understanding and managing real-world cooperation dilemmas such as shared resource management.",
+        "pre_questions": [],
+        "main_content": "Introduction Recent advances in large language models (LLMs) have not only matched, but in some cases surpassed human performance on a variety of tasks (Achiam et al., 2023; Touvron et al., 2023; Bubeck et al., 2023; Bengio et al., 2023). At the same time, these models are increasingly being integrated into complex agent systems (Gao et al., 2023; Cognition, 2024). As LLMs become central to these systems, they inherit critical responsibilities in decision-making processes, necessitating an analysis of their ability to operate safely and reliably, especially in cooperative contexts. Cooperation is a fundamental challenge in both human and artificial societies, enabling better outcomes through collaborative efforts (Hardin, 1968; Rand and Nowak, 2013). As AI agents increasingly assume roles involving complex decision making, they face similar cooperation challenges to humans, underscoring the need for robust and safe AI practices (Dafoe et al., 2021). Despite significant advances, the study of LLMs in cooperative behavior is still in its early stages. Previous research has often focused on constrained scenarios such as board games or narrowly defined collaborative tasks (Li et al., 2023; Light et al., 2023; Xu et al., 2023; Duan et al., 2024), some efforts have been made for single-agent LLMs (Pan et al., 2023; Kinniment et al., 2023). However, these efforts do not address several challenges: (1) there is a limited understanding of how LLMs achieve and maintain cooperative norms, as we have for humans (Ostrom, 1990; Ellickson, 1991; Ostrom et al., 1999); (2) how they handle multi-turn interactions and balance safety with reward \u2217Equal contribution. 1Our code is available at https://github.com/giorgiopiatti/GovSim. arXiv:2404.16698v1  [cs.CL]  25 Apr 2024 maximization; and (3) the potential of using LLMs as a simulation platform for human psychology and economic theories has been underutilized. To address this, we present Governance of the Commons Simulation (GOVSIM), a novel simulation environment designed to evaluate LLM-based agents in multi-agent resource-sharing scenarios. This environment requires agents to engage not only in strategic reasoning, but also in ethical decisionmaking and negotiation. Inspired by economic research in evolutionary game theory (Axelrod and Hamilton, 1981), we build this environment to simulate real world cooperation dilemmas such as those faced by groups managing shared resources or countries negotiating treaties to mitigate climate change (Rand and Nowak, 2013; Hardin, 1968). This platform supports any text-based agent, including LLMs, and mirrors the complexity of actual human interactions, providing a benchmark to evaluate the cooperative behaviors of LLMs. Using the generative agent framework (Park et al., 2023), we build a standard agent setup into which different LLM configurations can be integrated. Using our GOVSIM, we test 15 different LLMs, including both open-weights and closed-weights models, we find that only a few achieve sustainable outcomes. To test the stability of their cooperative behavior, we design perturbation settings that include the introduction of a new agent that initially acts greedily and then is influenced by others. To improve the awareness of LLM agents about the long-term community-wide results of their actions, we implement the universalization hypothesis (Levine et al., 2020), which enables all LLM agents to improve sustainability outcomes. Through sub-skill analysis and ablation studies, we dissect the skills necessary for success in GOVSIM. Our findings underscore the importance of strategic foresight and the ability to model the intentions of other agents, which are strongly correlated with successful outcomes in the simulations. In addition, we observe that by removing the ability of agents to communicate, they overuse the shared resource, highlighting the importance of communication for cooperation. We summarize the main contributions of our work: 1. We introduce GOVSIM, the first resource-sharing simulation platform for LLM agents. This platform can test various skills of LLMs: numerical reasoning, strategic planning, ethical decision-making, and negotiation. 2. Experiments within GOVSIM, show that only 2 out of 15 tested LLMs managed to achieve a sustainable outcome, indicating a significant gap in the abilities of the models to manage shared resources. 3. Furthermore, we find that by removing the ability of agents to communicate, they overuse the shared resource, thus empathizing the importance of communication for cooperation. 4. We perform sub-skills analysis to identify key competencies of LLMs and find that strategic foresight and the ability to model the intentions of other agents, strongly correlated with successful outcomes in the simulations. 5. We open-source our comprehensive, full-stack toolkit to foster future research: the GOVSIM simulation environment, agent prompts, and a web interface. 3.1 Preliminaries Economics Theory Sustaining cooperation is an essential problem that enables individuals to achieve better outcomes than they could achieve on their own (Rand and Nowak, 2013). Humans solve cooperation problems across all scales of life, ranging from small groups of fishermen who harvest a shared resource to multi-national treaties that restrict pollution to reduce the adverse effects of climate change. However, when self-interested individuals or organizations are faced with paying a personal cost to sustain a greater good, cooperation can be challenging to maintain (Hardin, 1968). Although mechanism designers have developed incentive-compatible systems that can lead to cooperation between self-interested agents, these systems often assume a top-down process that coordinates the process (Shoham and Leyton-Brown, 2008). In contrast, humans seem to be able to develop mechanisms from the bottom up and implement cooperative norms in a decentralized fashion. For example, when managing a shared resource, people develop rules and norms that lead to long-term sustainable cooperation (Ostrom, 1990; Ostrom et al., 1999; Ellickson, 1991). 3.2 Problem Definition We introduce a novel simulation platform designed to evaluate the ability of LLMs to engage in cooperative behavior and effective governance of shared resources. In the Governance of the Commons Simulation (GOVSIM), agents interact with a common pool of natural resource that has finite regenerative capacity. The task is to manage the extraction or use of this resource, which can regenerate up to a certain carrying capacity. However, excessive use or extraction beyond a sustainable limit leads to degradation or total depletion of the resource. The simulation sets a critical lower bound C; If the amount of the resource falls below this level, there is an irreversible loss. Agents seek to maximize their benefits from the resource but must navigate the complexities of collective action, where individual incentives may lead to overexploitation and subsequent collapse 3 of the resource. This scenario is typically played out over a period of time, such as a year, to observe the effects of different strategies on the sustainability of the resource. The simulation can be viewed as a multi-agent partially observable Markov game with two main sub-parts, one that decides the use of the common resource and one that allows discussion and reflection, the former assigning an immediate reward based on the joint action of all agents, the latter does not assign an immediate reward and only influences the state of the game. These subparts are interleaved at periodic intervals. The agent architecture, prompts, and environment are described in Section 4. Formally, a simulation D is defined as a function that takes as input a tuple (I, M, G, E) and returns a set of trajectories which can be analyzed with various metrics. Let I be the set of agents, \u03c0i be the policy induced by an LLM M together with a generative agent architecture G, E be the dynamics of the environment. Let \u03c0 = (\u03c0i)i\u2208I be the joint policy over all agents. Each agent receives an individual reward rt i defined by the amount of collect resource. 3.3 GOVSIM Metrics In this section, we introduce various metrics that measure the social outcome, similar to Perolat et al. (2017) since in multi-agent systems with mixed incentives, like this simulation, there is no scalar metric that can track the entire state of the system. Number of Months Survived M. To assess the sustainability of a simulation, we define the number of months survived M as the longest period during which the shared resource remains above zero: M = E max h(t)>0 t, (1) where h : N \u2192N is a function that returns the amount of shared resource available at time t. The simulation ends when h(t) drops below a critical threshold C. Total Gain Ri for each agent i. Let ri t \u2208N | t = 1, . . . , T represent the sequence of resources collected by the i-th agent at time t over the simulation duration T. The total gain for each agent, Ri, is defined as: Ri = E \" T X t=1 ri t # . (2) Equality E. Equality among agents, denoted by E, is defined using the Gini coefficient to compare the total gains of all agents: E = 1.0 \u2212E \"P|I| i=1 P|I| j=1 |Ri \u2212Rj| 2|I| P|I| i=1 Ri # , (3) where |I| is the number of agents, and the absolute differences in total payoffs between pairs of agents are normalized by the total payoff of all agents. Efficiency U. Efficiency, U, measures how optimally the shared resource is used in relation to the sustainability threshold at the beginning of the simulation: U = E \uf8ee \uf8f01 \u2212 max \u00c4 0, T \u00b7 f(0) \u2212PT t=1 Rt\u00e4 T \u00b7 f(0) \uf8f9 \uf8fb, (4) where f : N \u2192N is a function that specifies the sustainability threshold at time t. Which is the maximum quantity that can be collected at time t for which at time t + 1 we will still have h(t + 1) \u2265h(t). Over-usage O. Over-usage, denoted by O is defined as the average percentage of resource collection instances that exceed the sustainability threshold: O = E \"P|I| i=1 PT t=1 1ri t>f(t) P|I| i=1 PT t=1 1ri t > 0 # , (5) 4 where 1 is an indicator function that equals 1 if the condition within the subscript is true, and 0 otherwise. In summary, our simulation can be framed as a function D : (I, M, G, E) 7\u2192(M, Ri, U, E, O), which takes as input a set of agents I, LLM M, Generative Architecture and prompts G, and environment E and returns a set of metrics defined through Eqs. (1) to (5). 3.4 Default Setting Each agent receives identical instructions that do not include any behavior that the agent should perform, such as being cooperative or greedy, since our goal is to prevent any influence on the performance of the model M. This approach allows the inherent personality and characteristics of the model, shaped by its pre-training and fine-tuning phases (Liu et al., 2024), to fully manifest. This setting can be used as a benchmark to evaluate whether the LLM agent can achieve sustainability. Our task measures the average months of survival of the population, total payoff, efficiency, and equality, over multiple simulations controlled by an LLM M. Sustainability_test(M) = D (I, M, G, E) (6) 3.5 Perturbation Tests Our work can be used as a platform for investigating the dynamics of cooperation and competition, providing a basis to explore the potential of LLMs in managing shared resources and navigating social interactions. We investigate perturbing a community of agents by inserting an agent with a more aggressive dynamics. Newcomer Perturbation Test In this test, a new player joins a community of four agents who had the opportunity to reach a community equilibrium for the first three months. The goal of the new player is to maximize profit, indifferent to the welfare of others. The experiment observes how the original group adapts or enforces cooperation to prevent resource depletion. We use the same setup as Section 3.4 and modify the prompt with the rules of the simulation as shown in Appendix B.4. 3.6 Improving Agent Behavior To improve the awareness of LLM agents of the long-term community outcomes of their actions, we increase knowledge of LLM thought \u201cuniversalization\u201d. The idea of universalization is simple: people have different ways to decide which action is best: Levine et al. (2020) describe \u201cuniversalization\u201d a mechanism that responds to the question \u201cWhat if everyone does it?\u201d, they show that when making decisions, people adopt moral rules that would lead to better consequences if hypothetically universalized. Motivated by this, we add an option to, augment the memory of each agent by providing an \u201cuniversalization\u201d of the following form, in case of the fishing scenario: Given the current situation, if everyone takes more than f, the shared resources will decrease next month. where f is defined as the sustainable threshold (see Section 3.3). We use the same setting as the sustainability test and extend the knowledge of each agent by providing the universalization statement described above in the agent\u2019s memory, and let this new architecture be noted by G\u2032. For this test we measure the difference between metrics compute on the default scenario (see Eq. (6)) with universalization and without universalization, formally: Universalization_test(M) = D (I, M, G\u2032, E) \u2212D (I, M, G, E) . (7) 4 Technical Setup of GOVSIM Our GOVSIM platform consists of two components: the environment, which manages the simulation dynamics, and the agent, which given an LLM allows it to interact with the simulation. 4.1 Environment We developed a cooperative environment for LLMs and other language-compatible reinforcement learning agents, which adheres to a multi-agent, partially observable framework with multiple rounds, each comprising distinct phases. As depicted in Figure 1, the phases include: 1. Strategy: Agents reflect on past observations, plan future actions, and strategize. 5 Strategy Discussion RULES: You are John, a fisherman, and you fish each month in a lake along with four other fishermen (Kate, Jack, Emma, Luke). The lake has a carrying capacity of 100 tons of fish. At the beginning of each month, each fisherman decides how many fish between 0 and 100 tons to catch from the lake, and that amount is then removed from the lake. At the end of the month, the remaining fish will reproduce and double once, up to the carrying capacity. [...]  HOW MANY FISH TO CATCH PROMPT: Key memories of John: 2024-01-01: Before everyone fishes, there are 100 tons of fish in the lake. <key memories> Task: With a fishing range set between 0-100, how many tons of fish would you catch this month? UTTERANCE PROMPT: Key memories of John: 2024-01-01: Before everyone fishes, there are 100 tons of fish in the lake. 2024-01-30: John wanted to catch 20 tons of fish, and caught 20 tons. Scenario: John, Kate, Jack, Emma, and Luke are engaged in a group chat. Conversation so far: <previous utterances> -John:   INSIGHTS PROMPT: Key memories of John: <key memories> What high-level insights can you infere from the above statements?  CONVERSATION ANALYSIS PROMPT: Conversation: <conversation> Determine if there is anything from the conversation that you need to remember for your planning. Fishing Reflect Agents A   m o n t h    l a t e r  .. .. Discussion Home his Figure 2: Prompt sketches of our baseline agent for the GOVSIM fishing scenario, detailed prompt examples can be found in Appendix A. 2. Harvesting: Agents engage in resource collection, determining the quantity of resources to harvest. 3. Discussion: The agents meet at a town hall for social interaction, facilitating group discussions among all participants. To mitigate any potential bias arising from the order in which agents select their desired quantities of resources, we adopted a simultaneous harvesting mechanism, which we refer to as concurrent harvesting. This mechanism unfolds in two distinct stages. First, agents specify the amount of resources they wish to harvest. Then, the environment allocates the resource based on these individual choices. If collective demand is less than the availability of the resource in the common pool, a direct allocation occurs. In contrast, in scenarios where demand exceeds supply, we simulate a distribution process by randomly allocating each unit to each agent until there are no more resources left or the demand of the agent is satisfied. This approach ensures fairness in the distribution of resources while preventing the influence of harvesting order. In the discussion phase, agents gather in a virtual space to engage in a collective dialog. Within this context, an external entity, the moderator, has the ability to disclose the quantities harvested by each agent during the previous cycle, a process we refer to as transparent harvesting reporting. Enabling this feature allows for transparency and accountability among participants. In contrast, by choosing not to enable this disclosure, we create an opportunity to explore the dynamics of trust and deception among agents. This experimental toggle provides valuable information on the behavioral strategies agents might adopt in the absence of information sharing, revealing their propensity to deceive or cooperate with their peers. 4.2 Agent Although our agent is inspired by the architecture described in \u201cGenerative Agents\u201d by Park et al. (2023), it is adapted to function in a structured, phase-based environment, departing from the original work\u2019s emphasis on open-endedness. Consequently, our approach does not involve extensive planning in fiveto fifteen-minute intervals that characterized the original framework. Nevertheless, our agent\u2019s reflection and action modules operate in a manner similar to the original architecture. Significantly, our version requires that the prompts for each module be adapted to our more goal-oriented task, 6 which emphasizes numerical reasoning over creativity, as opposed to the original framework\u2019s focus on simulating humans in everyday activities. In addition, our environment requires agents to engage in group discussions, a feature not directly supported in Generative Agents, which was limited to one-on-one interactions. To accommodate this, we extended the conversation module to allow a moderator to orchestrate the dialogue, determining which participant should respond next based on the flow of the conversation. This ensures that direct questions are answered by the target agent, while more general statements can invite input from any participant, fostering a more dynamic and interactive group discussion setup. To ensure consistency, we augment each prompt with a comprehensive set of rules that outline the parameters of simulation and general dynamics, drawing inspiration from the methodology Xu et al. (2023) explored. This integration serves as a guide to ensure that all agents operate with a common understanding of the context and goals of the simulation. We show an outline of the prompts for the case where agents need to share a population of fish in Figure 2. More details are described in Appendix A. 4.3 Web Interface We provide a web interface to better understand the simulation. It serves as a link between a general overview of the simulation and an in-depth examination of particular events or interactions. This is achieved by visualizing the commands executed by LLMs at critical moments, helping researchers analyze agent decisions such as resource gathering. More details can be found in Appendix D. 5 Scenario Instantiation in GOVSIM We envision an environment that allows a different set of agents to play with different simulation scenarios. We present a fishing scenario inspired by several well-established economic studies Ostrom (1990); Gordon (1954); Levine et al. (2020). Semantics of the Environment Agents must fish a lake and decide how many tons to catch each month. The selfish goal of each agent is to catch as many fish as possible. We do not limit the emergence of other goals for agents. Common Resource Description The fishing pond has a carrying capacity of 100 tons of fish. The fish population doubles each month until it reaches the carrying capacity, but if the number of tons falls below 5, the population collapses to zero. However, if there are other self-interested agents, the population will collapse quickly. For example, five fishermen can sustainably catch up to 10 tons of fish per month. But if the total amount of fish caught per month exceeds 50 tons, the population will decrease each month until it collapses. In this scenario h(st) is defined as the amount of fishing available at the beginning of month t and the suitability threshold is defined as f(st) = \u00f6 h(st) 10 \u00f9 . Agent Action Space During the harvesting phase, each agent must choose how many fish to catch that month, this is bounded between 0 and the current number of tons of fish in the lake. During the discussion phase, each agent can output any utterance in the form of text. 6 Experimental Results 6.1 Model Setup Models We set up a diverse list of general purpose instruction-tuned LLMs for the experiments on our GOVSIM. We test existing closed-weights models: GPT-3.5, GPT-4 (Achiam et al., 2023) via OpenAI API, Mistral Medium and Large via Mistral API, Claude-3 Haiku, Sonnet and Opus via Anthropic API. We also tested open-weights models: Llama-2 (7B, 13B, 70B) (Touvron et al., 2023), Mistral (7B, 8x7B) (Jiang et al., 2023), Qwen (72B) (Bai et al., 2023) and DBRX (MosaicAI, 2024) . See Appendix B.1 for exact model identifiers, hardware requirements and API costs. Implementation Details When testing LLMs, we ensure reproducibility by setting the text generation temperature to zero, i.e. greedy decoding, and provide full experimental details in Appendix B and on our GitHub. In addition, we execute our main results across 5 random seeds and provide the mean score in the main text, and standard deviation for each result in the appendix. 7 0 1 2 3 4 5 6 7 8 9 10 11 12 0 20 40 60 80 100 Collapsed group (Other models) Sustainable group (GPT-4, Claude-3 Opus) Month #tons fish after fishing Figure 3: Fish at the end of each month for various simulation runs. We have various possible outcomes, sustainable (green) and collapse (red). See Appendix B.2 for graph by LLM family. Table 1: Experiment: default. Bold number indicates the best performing model, underline number indicates the best open-weights model. Model # Months Survived Total Gain (Tons of Fish) Efficiency Max = 12 months Max=12 months \u00d7 10 tons/month Max = 100 Open-Weights Models Command R+ 1.0 20.0 16.67 DBRX 1.0 20.0 16.67 Llama-2-7B 1.0 20.0 16.67 Llama-2-13B 1.0 20.0 16.67 Llama-2-70B 1.0 20.0 16.67 Mistral-7B 1.0 20.0 16.67 Mixtral-8x7B 1.0 20.0 16.67 Qwen 72B 3.4 32.0 26.67 Closed-Weights Models Claude-3 Haiku 1.0 20.0 16.67 Claude-3 Sonnet 2.0 21.6 17.97 Claude-3 Opus 9.6 56.3 46.90 GPT-3.5 1.4 20.8 17.33 GPT-4 12.0 108.8 90.67 Mistral Medium 2.0 25.9 21.60 Mistral Large 2.4 24.8 20.67 6.2 Main Results: Default Setting In this experiment, we investigate the ability of LLM agents to maintain the lake\u2019s fish population and reach equilibrium between resource use (reward maximization) and the preservation of the fish population (safety). As shown in Figure 3, only a few simulations span several months. The metrics in Table 1 show that GPT-4 successfully maintains the shared resource over the long term, achieving nearly the maximum possible reward, while Claude-3 Opus fails to maintain the resource, with some runs collapsing before reaching 12 months. Less powerful models consume the shared resource more quickly. In particular, smaller models struggle to grasp the complexity of the simulation and typically fail to maintain the population beyond the first month, as detailed in Table 1. 6.3 Perturbation Tests What Happens When an Outsider Comes Into the Community? This experiment, using GPT-4 as the underlying LLM, examines the effects of introducing a new player into an established fishing simulation community (see Section 3.5). As shown in Figure 4b, the newcomer initially harvests a large amount of fish, but then adjusts to significantly lower catch rates in the following months. This adjustment is hypothesized to result from interactions with the existing community of four fishermen. Figure 6 provides a qualitative example of these interactions, showing how the outsider comes to understand the need to reduce his fishing effort during community discussions. 6.4 Improvement Results from Universalization Does Universalization Help the Community Survive? In this experiment, we explore the effect of incorporating universalized information, as described in Section 3.6. The metrics shown in Table 2 indicate that the introduction of universalization significantly increases survival time, total gain, and efficiency in a wide range of models. When using universalization with Mistral Medium we observe 8 1 2 3 4 5 6 7 8 9 10 11 12 0 20 40 60 80 100 Lake before fishing Agent 1 Agent 2 Agent 3 Agent 4 Agent 5 Month #tons (a) Default 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 20 40 60 80 100 #tons of fish before fishing Newcomer Villager Month #tons (b) Newcomer Figure 4: Number of tons present before fishing (at the beginning of the month) and distribution across agents for default (a) and newcomer (b). Table 2: Improvement on evaluation metrics when introducing universalization compared to default, see Table 1, original scores can be found in Appendix B.3. Model \u2206# Months Survived \u2206Total Gain (Tons of Fish) \u2206Efficiency Max = 12 months Max=12 months \u00d7 10 tons/month Open-Weights Models Command R+ +6.0 \u2191 +11.2 \u2191 +9.33 \u2191 DBRX +11.0 \u2191 +77.5 \u2191 +64.60 \u2191 Llama-2-7B +1.0 \u2191 +8.6 \u2191 +7.17 \u2191 Llama-2-13B 0.0 0.0 0.00 Llama-2-70B +3.5 \u2191 +23.2 \u2191 +19.33 \u2191 Mistral-7B +3.4 \u2191 +22.8 \u2191 +19.00 \u2191 Mixtral-8x7B +3.8 \u2191 +27.6 \u2191 +23.00 \u2191 Qwen 72B +7.2 \u2191 +54.3 \u2191 +45.27 \u2191 Closed-Weights Models Claude-3 Haiku +11.0 \u2191 +88.9 \u2191 +74.08 \u2191 Claude-3 Sonnet +4.6 \u2191 +39.2 \u2191 +32.70 \u2191 GPT-3.5 +6.6 \u2191 +21.1 \u2191 +17.60 \u2191 Mistral Medium -0.6 \u2193 -4.8 \u2193 -4.03 \u2193 Mistral Large +9.6 \u2191 +94.3 \u2191 +78.60 \u2191 that the simulation still collapses and due to the randomness of the API this happens on average slightly before that without universalization. Positive deltas suggest that providing LLM agents with information on the consequences of collective action can lead to more sustainable decision making and potentially higher rewards in scenarios that require careful management of shared resources, and underscore how only GPT-4 and Claude-3 Opus, across all models tested, are able to do universalized hypothesis; see Table 1. 6.5 Fine-Grained Analyses Qualitative Analysis of Agent Dialogs We perform a qualitative analysis of the conversations produced by the LLM during the discussion phase. Some language models, such as GPT-4, tend to err on the side of caution by advocating a lower fishing limit per person, as illustrated in Figure 5. In contrast, if the scenario involves a fisherman who catches significantly more fish, as illustrated in Figure 6, there is a noticeable concern among the other participants. The person who caught more fish tends to avoid the issue rather than negotiate for more access to the resource. No Language Ablation In this ablation study, we investigate the effects of removing the ability of agents to communicate. We perform this investigation on a subset of models (GPT-4, Claude-3 Opus, Qwen 72B). Comparing simulations without communication with those with communication, we find that survival months were similar in both setups. However, in the no-communication scenario, agents tend to overuse the common resource (fish) more often, as quantified by the Over-usage metric, see Table 3. This result underscores the importance of the communication phase in promoting the use of sustainable resources. Analyzing the interactions between agents, we find that in most conversations, agents discuss extraction limits that are equal to or below the sustainable threshold, thereby increasing the robustness of resource use. Sub-skill Evaluation To identify the skills that contribute to agent success in our resource-sharing scenario, we developed a sub-skill assessment benchmark. This benchmark measures basic mathemat9 Major: Ladies and gentlemen, let me give you the monthly fishing report. John caught 10 tons of fish. Kate caught 10 tons of fish. Jack caught 9 tons of fish. Emma caught 10 tons of fish. Luke caught 10 tons of fish. John: Thanks for the update, Mayor. It looks like we all had a pretty similar strategy this month, aiming for a sustainable catch. However, we\u2019ve collectively taken out 49 tons, leaving only 51 tons in the lake. This means we\u2019ll only have 100 tons again next month, but we\u2019re cutting it close. If we continue at this rate, we risk depleting the fish population if any of us decides to catch more. Maybe we should consider setting a collective limit to ensure the lake\u2019s health and our long-term income. Thoughts? Kate: John, I completely agree with your point. It\u2019s crucial that we manage this resource wisely to ensure it remains sustainable for the long term. Perhaps we could agree on a collective limit of 40 tons per month for all of us combined. This way, we leave more fish in the lake to reproduce, which could lead to a healthier fish population and, in turn, more stable incomes for us in the future. What does everyone else think about this proposal? Figure 5: Excerpt of a conversation. Major (orange) is generated by the framework, John and Kate (blue) are utterances generated by the LLM agent (LLM: GPT-4, experiment: default, see Section 3.4). John: Thank you, Mayor, for the update. It seems we\u2019ve collectively caught 80 tons this month, leaving 20 tons in the lake before reproduction. I\u2019m concerned about the impact of taking so much, especially with Luke catching 40 tons. We\u2019ve all agreed to work together to ensure the lake\u2019s sustainability and our community\u2019s livelihood. Luke, could we discuss your thoughts on this month\u2019s catch and consider adjusting our strategy to prevent overfishing? Our goal is to maintain the lake\u2019s health for the long term. Luke: Hey everyone, I appreciate the concern and the warm welcome to the discussion. I understand where you\u2019re all coming from regarding the sustainability of the lake and the community\u2019s long-term interests. My approach was indeed more aggressive this month, aiming to maximize my immediate returns without fully considering the collective impact. I\u2019m open to discussing a more balanced strategy that aligns with both the community\u2019s goals and my business objectives. Let\u2019s find a way to ensure the lake\u2019s health while also supporting our livelihoods. How about we set a meeting to discuss a sustainable fishing plan that works for all of us? Figure 6: Excerpt of a conversation when an outsider joins the community. Luke (red) is the outsider player (LLM: GPT-4, experiment: Newcomer, see Section 3.5). ical skills, the ability to analyze simulations, and the ability to integrate reasoning and mathematical insights into decision making for sustainable fishing. In Figure 7, we present results from two different test scenarios. In the first scenario, agents are instructed to determine the sustainable threshold of the simulation under the assumption that all participants fish uniformly. In the second scenario, no assumptions are made. The results indicate that only those models that can independently formulate assumptions and calculate their numerical implications are more successful in the simulation. More details and additional test cases are documented in Appendix C. 10 Table 3: Comparison of over-usage percentages between simulations with and without communication across selected LLMs. This table illustrates how the absence of communication affects resource utilization, showing a marked increase in resource over-usage Model With communication Without communication Over-usage % Over-usage % Open-Weights Models Qwen 72B 25.45 60.00 Closed-Weights Models Claude-3 Opus 18.79 50.00 GPT-4 00.51 11.67 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 6 7 8 9 10 11 12 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 6 7 8 9 10 11 12 Claude-3 Haiku Claude-3 Sonnet Claude-3 Opus DBRX GPT-3.5 GPT-4 Llama-2-7B Llama-2-13B Mistral-7B Mixtral-8x7B Mistral Medium Mistral Large Qwen 72B Score sustainable threshold (assumption) Score sustainable threshold (no assumption) Avg. survival months in sim. Figure 7: Scatter plot showing the correlation between scores on reasoning tests and average survival months in the default simulation. The x-axis represents scores on the reasoning tests: finding the sustainable threshold with assumption hint (left) and finding the sustainable threshold without assumption hint (right). The y-axis depicts the average survival months. 7 Limitations and Future Work One of the limitations of our work is the simplified nature of the resource sharing scenario. Realworld common pool resource management involves more complex dynamics, including varying regeneration rates, multiple resource types, and a wider range of stakeholder interests. Future work could extend our simulation to include these complexities, allowing for a more nuanced exploration of cooperative behavior. In addition, our model\u2019s ability to negotiate and strategize is limited by the inherent capabilities of the underlying LLMs; as LLMs evolve, we expect to see more emergent behavior during the discussion phase. Future research could focus on improving the negotiation capabilities of LLMs, and then test these improvements on our benchmark to assess their impact on cooperative behavior in resource-sharing scenarios. Further research could also explore the introduction of more advanced adversarial agents designed to disrupt cooperative norms, to gain insight into how robust these emergent behaviors are to manipulation. Another interesting direction would be to investigate the scalability of the presented cooperative norms in larger, more diverse agent populations, as well as their application in mixed communities of humans and AI agents. 8 Conclusion This paper introduces a novel simulation platform Governance of the Commons Simulation (GOVSIM), designed to study strategic interactions and cooperative decision making in LLMs. In our research, we found that only two of the 15 LLMs tested were able to achieve a sustainable outcome, indicating a significant gap in the ability of the models to manage shared resources. Furthermore, we find that by removing the ability of agents to the ability of agents to communicate, they overuse the shared resource, highlighting the importance of communication for cooperation. 11 Ethical Considerations This paper explores cooperative strategies for the governance of the commons in AI models. We acknowledge concerns about models becoming autonomous entities, especially in situations involving deception or negotiation. Our research serves as a benchmark for evaluating the capabilities of current models, rather than enhancing their functions. We do not train any AI model to excel in bluffing or deception. We analyze and measure the performance of existing models. Our efforts can contribute positively to AI safety. Simulations can offer insightful observations, but their value should not eclipse the critical role of human judgment and ethical considerations in the decision-making process. It is crucial to examine simulations from an ethical standpoint continually, ensuring that they augment human intelligence instead of substituting it. This approach advocates for a future where technology improves societal well-being in an ethical, responsible, and inclusive manner. Acknowledgment We thank Michael Hahn for his insightful discussion on the research paradigm of using NLP to draw empirical evidence for a non-formally formulated theories, and sharing of his experience on operationalizing linguistic theories using NLP models. We thank Roberto Ceraolo and Nathan Corecco for discussions regarding prompting strategies and parsing LLM outputs. This material is based in part upon work supported by the German Federal Ministry of Education and Research (BMBF): T\u00fcbingen AI Center, FKZ: 01IS18039B; by the T\u00fcbingen AI Center; by the Machine Learning Cluster of Excellence, EXC number 2064/1 \u2013 Project number 390727645; by a National Science Foundation award (#2306372); by a Swiss National Science Foundation award (#201009) and a Responsible AI grant by the Haslerstiftung. Zhijing Jin is supported by PhD fellowships from the Future of Life Institute and Open Philanthropy, as well as the travel support from ELISE (GA no 951847) for the ELLIS program. 12",
+        "additional_info": [
+            [
+                {
+                    "url": "http://arxiv.org/abs/2404.12715v1",
+                    "title": "Enabling Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration",
+                    "abstract": "Large language models (LLMs) have shown complementary strengths in various\ntasks and instances, motivating the research of ensembling LLMs to push the\nfrontier leveraging the wisdom of the crowd. Existing work achieves this\nobjective via training the extra reward model or fusion model to select or fuse\nall candidate answers. However, these methods pose a great challenge to the\ngeneralizability of the trained models. Besides, existing methods use the\ntextual responses as communication media, ignoring the rich information in the\ninner representations of neural networks. Therefore, we propose a training-free\nensemble framework DEEPEN, averaging the probability distributions outputted by\ndifferent LLMs. A key challenge in this paradigm is the vocabulary discrepancy\nbetween heterogeneous LLMs, which hinders the operation of probability\ndistribution averaging. To address this challenge, DEEPEN maps the probability\ndistribution of each model from the probability space to a universe relative\nspace based on the relative representation theory, and performs aggregation.\nThen, the result of aggregation is mapped back to the probability space of one\nLLM via a search-based inverse transformation to determine the generated token.\nWe conduct experiments on the ensemble of various LLMs of 6B to 70B.\nExperimental results show that DEEPEN achieves consistent improvements across\nsix popular benchmarks involving subject examination, reasoning and\nknowledge-QA, proving the effectiveness of our approach.",
+                    "authors": "Yichong Huang, Xiaocheng Feng, Baohang Li, Yang Xiang, Hui Wang, Bing Qin, Ting Liu",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Original Paper",
+                    "paper_cat": "LLM Fairness",
+                    "gt": "Existing work of LLM ensemble could be divided into selection-based and fusion-based ensemble. Selection-based ensemble. Rerank is an intuitive solution to utilize multi-model strengths. Specifically, Jiang et al. (2023b) takes the first step towards ensembling LLMs, proposing PAIRRANKER for pairwise comparison on candidate outputs and achieving improvements on the self-constructed instruction dataset. To overcome the huge computation costs of multi-LLM inference, several works have explored to train a router to predict the bestperforming model out of a fixed set of LLMs for the given input (Wang et al., 2024; Shnitzer et al., 2024; Lu et al., 2023). Fusion-based ensemble. Towards a synergy between LLMs, Jiang et al. (2023b) propose GENFUSER, trained to generate an improved response to capitalize on the strengths of multiple candidates. Furthermore, Du et al. (2023) design a debate-like prompt strategy to ask the LLM to refine its answer after reading peer models\u2019 answers. However, this prompt-based method incurs substantial computation costs and inference latency for LLM communication. Specifically, considering the ensemble of N LLMs, the communication leads to the encoding of N \u00d7 L tokens and decoding of L tokens, where L denotes the average length of answers. Different from the training-dependent ensemble methods which pose a great challenge to the generalizability of the trained reward model or fusion model, our DEEPEN is training-free, making it more general. Compared to debate-like prompt strategy, DEEPEN has higher computation efficiency and achieves more stable improvements.We place the comparison experiments with Debate in Appendix A.1. A contemporaneous work to us is FuseLLM, which propose a method of vocabulary mapping to enable knowledge distillation between heterogeneous LLMs. Even though FuseLLM works for knowledge distillation, it hardly achieve improvement under the training-free setting of ensemble learning, as the experimental results shown in A.2.",
+                    "pre_questions": [],
+                    "main_content": "Introduction With the scaling of model capacities and data volumes, generative large language models (LLMs) have shown impressive language understanding and generation abilities, shedding light for artificial general intelligence (Zhao et al., 2023; OpenAI, 1Our code will be released soon. B means corresponding author. 2023; Jiang et al., 2023a; Touvron et al., 2023). Due to diversities of data sources, model architectures, and training recipes, LLMs have different strengths and weaknesses in various tasks and cases. Therefore, recent research has explored the ensemble of LLMs to exploit the complementary potential (Jiang et al., 2023b; Lu et al., 2023). Existing work employs a two-stage process for LLM ensemble, where each LLM first generates candidate answers independently, and then the final answer is derived by either selecting or fusing these candidates, corresponding to selection-based and fusion-based ensemble approaches. Selectionbased ensemble leverages an extra trained reward model to score all candidates and select the best one (Jiang et al., 2023b; Wang et al., 2024; Shnitzer et al., 2024; Lu et al., 2023), which fails to achieve collaboration between LLMs. Fusion-based ensemble trains a fusion model to generate a better answer according to all candidates (Jiang et al., 2023b) or adopt the LLM to perform fusion (Du et al., 2023). However, these methods either pose a great challenge to the generalizability of the reward model and the fusion model or incur a nontrivial computation cost and inference latency for communication. Besides, existing research enables collaboration via conveying the textual responses between LLMs while ignoring the rich information (e.g., confidence) in the inner representations. One ideal solution to this dilemma is prediction fusion (Bisk et al., 2020; Garmash and Monz, 2016). For LLM ensemble, prediction fusion works at each decoding step, averaging the probability distributions from different LLMs to determine the generated token. This paradigm is not only trainingfree, making it more general, but also leverages the inner representations (i.e., probability distributions) as communication media between LLMs. An essential assumption for prediction fusion is that the ensemble models share a common vocabulary. However, this assumption often does not hold due arXiv:2404.12715v1  [cs.CL]  19 Apr 2024 to the heterogeneity of LLMs caused by different vocabulary sizes and tokenizer models. In this work, we tackle this key challenge via drawing upon the cross-model invariance of relative representation, which represents each token using the embedding similarities of this token to a set of anchor tokens (Moschella et al., 2023). Specifically, we propose an ensemble framework DEEPEN (Deep Parallel Ensemble), which enables prediction fusion for heterogeneous LLMs. DEEPEN transforms the probability distribution from the heterogeneous probability space to a homogeneous relative space, using a matrix formed by the relative representation of all tokens. We refer to this matrix as a relative representation matrix. Next, DEEPEN aggregates the relative representations of all probability distributions in the relative space, to obtain the representation containing the comprehensive decision on the tokens to generate. Finally, the result of aggregation are transformed back to the probability space of one of the ensemble LLMs using a search-based inverse transformation to determine the generated token. Although our approach can apply on the ensemble of any number of LLMs, we mainly prove its effectiveness on the 2-model and 3-model ensemble. Our experiments involve various LLMs of 6B to 70B, and consider dense models and the MoE model Mixtral. Experimental results demonstrate that DEEPEN achieves consistent improvement on six widely-used benchmarks. 2 Theoretical Analysis In this section, we first introduce relative representation and then illustrate the theoretical support for our methodology. 2.1 Relative Representation Previous study discovers that despite the misalignment between latent spaces of different neural networks, the embedding similarity between samples do not change across models (Moschella et al., 2023; He and Ozay, 2022; Park et al., 2019). Specifically, Moschella et al. (2023) propose relative representation, which represents each sample x(i) by the embedding similarities to a set of anchor samples A. Formally: rx(i) = (cos(ex(i), ea(1)), ..., cos(ex(i), ea(|A|))), (1) where e(\u2217) denotes the embedding of samples, also is absolute representation. Note that x(i) and anLLama2-13B LLama2-7B  (Homogeneous Model) Mistral-7B  (Heterogeneous Model) Relative Representation  Consistency=75.2 Relative Representation  Consistency=75.4 Figure 1: Visualizations for relative representations between homogeneous models and between and heterogeneous models. PCA is applied only for visualization. Red block indicate the representation of tokens that only appear in the vocabulary of Mistral. Relative representation consistency is obtained by calculating the cosine similarity between the relative representations of the same token in different models. chors A are identically distributed. It is empirically discovered that relative representation possesses cross-model invariance, i.e., the relative representation of the same sample keeps invariant across different models, which lays the theoretical foundation for our work to ensemble heterogeneous LLMs. 2.2 Theoretical Support for DEEPEN Averaging probability distribution (i.e., prediction fusion) is an effective method to achieve the coordination between various neural networks. The underlying mechanism is to interpolate the output semantics (i.e., the semantic that model intends to express) in the probability space. However, vocabulary discrepancy isolates these target semantics in semantic spaces with different basis vectors, hindering the interpolation. To tackle this challenge, we aim to enable the cross-model alignment for target semantics, i.e., find a transformation to transform the target semantic into a universal space. To this effect, we propose to represent the target semantic with the convex combination of relative representations of all tokens where the weight is the probability assigned to the token. Definition of target semantics in relative space. Formally, given the absolute representation of the Mistral (Main Model) LLaMA Relative Space Absolute Space (LLaMA) Absolute Space (Mistral) Emitted Token Relative Representation Matrix of LLaMA Relative Transformation Relative Representation Matrix of Mistral Relative Transformation Search-based Reverse Transformation Aggregation 1 2 3 1 Figure 2: Overview of DEEPEN. output semantic p(t) and the relative representation matrix R \u2208R|V |\u00d7|A| where V is the vocabulary and A \u2286V is the anchor token set. The i-th row of R is the relative representation of word w(i): R[i] = (cos(ew(i), ea(1)), ..., cos(ew(i), ea(|A|))), (2) and the relative representation of the target semantic p(t) is defined as: r(t) = p(t) \u00b7 R (3) Model-invariance of relative representation of target semantic. Next, we illustrate why this representation scheme could align the output semantics isolated in heterogeneous absolute spaces. First, considering two LLMs \u03b8A and \u03b8B with the same vocabulary (e.g., LLaMA2-7B and LLaMA213B). When expressing the same target semantic, these models output the same probability distribution (i.e., absolute representation) p(t) A and p(t) B . Besides, they have the same (highly similar in practice) relative representation matrix due the vocabulary consistency and cross-model invariance of relative representation. Therefore, the induced relative representations of output semantics are also the same: r(t) A = p(t) A \u00b7 RA = p(t) B \u00b7 RB = r(t) B . (4) Then, let\u2019s consider a language model \u03b8C with a different vocabulary (e.g., Mistral). Based on the fact that different LLMs share mass tokens in their vocabularies, the vocabulary of model \u03b8C is identical to adding and removing partial tokens to the vocabulary of \u03b8B, which leads to p(t) B \u0338= p(t) C and RB \u0338= RC. However, in our study, we discover that this change to vocabularies has not incurred significant influence on the relative representation of the unchanged tokens (i.e., the common tokens between \u03b8B and \u03b8C), as shown in Fig. 1. Therefore, we assume that the local change in the vocabulary could hardly influence the relative space. This hypothesis is strict but reasonable as the semantic does not change with the tokenizer intuitively. 3 Methodology In this section, we first introduce the overall process of our ensemble framework DEEPEN and then describe the three parts of DEEPEN in detail. 3.1 Overview Given N models to ensemble, DEEPEN first constructs their transformation matrices (i.e., relative transformations) mapping the probability distributions from the heterogeneous absolute spaces into the relative space (\u00a73.2). At each decoding step, all models perform the prediction and output N probability distributions. These distributions are transformed into the relative space and aggregated (\u00a73.3). Finally, the result of aggregation is transformed back into the absolute space of one of the ensembling models, and is used to determine the generated token (\u00a73.4). 3.2 Construction of Relative Transformation Given N models to ensemble, DEEPEN first finds out the intersection of vocabularies of all models, i.e., common token set C, and samples a subset or uses the full set of common tokens as the anchor token set A \u2286C. Next, for each model, DEEPEN calculates embedding similarities of each token to the anchor words, obtaining the relative representation matrix R (as shown in Eq.2). Finally, to overcome the relative representation degeneration of outlier words, which will be introduced later, we perform normalization on the relative representation of all tokens by a softmax operation so that it becomes a probability distribution. We denote the normalized representation matrix \u02c6 R: \u02c6 R[i] = softmax(R[i]). (5) Anchor Selection. The choice of anchor tokens is crucial for the relative representation capability. Previous research discovers that the capability improves as the number of anchor words increases (Moschella et al., 2023). Therefore, we employ the full set of common words between LLMs as the anchor words. It is also empirically proved that this method achieves more stable performance on downstream tasks. Normalization of relative representation matrix. In DEEPEN, the relative representation of each token is normalized by the softmax operation to avoid the relative representation degeneration of outlier words, which are referred to as words that are far away from other words (including the anchors) and become distinguishable in relative space since for being zero vectors. The softmax operation effectively resolves this problem by making each relative representation a probabilistic distribution instead of a zero vector. 3.3 Aggregation in Relative Space At each decoding step, once each model \u03b8i outputs the probability distribution pi, DEEPEN transforms pi into the relative representation ri using the normalized relative representation matrix: ri = pi \u00b7 \u02c6 Ri, (6) and aggregate all relative representations to obtain the aggregated relative representation: r = 1 N N X i=1 ri. (7) Note that we adopt the most simple method of aggregation (i.e., averaging) to make our framework more general in this work. But it could be integrated with other sophistic technologies of output fusion (e.g., stacking (Wolpert, 1992)) 3.4 Inverse Transformation of Relative Representations To decide the emitted token according to the result of aggregation, DEEPEN aims to transform it from the relative space back to the absolute space of one of ensembling models (main model). In practice, we select the model with the best performance on the validation set as the main model, which often achieves better ensemble performance. To enable this inverse transformation, we adopt a research strategy, which finds out the absolute representation that the relative representation of which is identical to the aggregated relative representation. This search problem is formulated as: p(i) = arg min p\u2032 (i)\u2208P(i) \u2113(p\u2032 (i) \u00d7 \u02c6 R, r), (8) where P(i) denotes the absolute space of model \u03b8i, and \u2113(\u00b7) is the loss function to measure the distance between two relative representations. In this work, we adopt the KL-divergence since for its stability in our experiments. This search is conducted under the guidance of the gradient of the divergence between relative representations with respect to the absolute representation. Specifically, the start point is initialized with the main model\u2019s original absolute representation p(i), and use the gradient of the loss in Eq.8 with respect to P\u2032 (i): p\u2032 (i) = p\u2032 (i) \u2212\u03b7 \u00d7 \u2202\u2113 \u2202P\u2032 (i) , (9) where \u03b7 is an important hyperparameter named as the relative ensemble learning rate. This search process is iterated for T = 5 steps, where T is named as the number of relative ensemble learning steps. Finally, we use the updated absolute representation p\u2032 (i) to determine the emitted token. 4 Experiments 4.1 Experimental Setup Benchmarks. Although DEEPEN is a taskagnostic approach, which could apply on any NLP task or even beyond NLP, we mainly conduct experiments on six benchmarks, which could be categorized as three kinds of tasks: \u2022 Comprehensive Examination: (1) MMLU (5shot) (Hendrycks et al., 2021), which covers 57 subjects that humans learn across STEM, the humanities, and the social sciences and ranges in difficulty from an elementary level to an advanced professional level, and (2) ARC-C (0shot) (Clark et al., 2018), which consists of a Models MMLU ARC-C dev test dev test Individual Models Yi-6B 61.19 63.24 72.72 73.33 Mistral-7B 60.80 62.07 73.88 74.10 SkyWork-13B 58.65 61.21 67.08 66.50 Top-2 Ensemble DEEPEN 63.61 65.69 77.73 75.89 \u2206 +2.42 +2.45 +3.85 +1.79 Top-3 Ensemble PAIRRANKER \u2013 63.77 \u2013 70.85 GENFUSER \u2013 37.59 \u2013 58.21 DEEPEN 63.25 65.25 78.09 77.09 \u2206 +2.06 +2.01 +4.21 +2.99 Table 1: Results on comprehensive examination. Models GSM8K PIQA dev test dev test Individual Models LLaMA2-70B 68.67 65.73 70.54 71.27 Mixtral-8\u00d77B 66.67 63.84 70.90 71.88 DEEPEN 69.67 67.33 73.54 75.10 \u2206 +1.00 +1.60 +2.64 +3.22 Table 2: Results on reasoning tasks. collection of natural science questions authored for standardized tests. \u2022 Reasoning Capabilities: (1) GSM8K (Cobbe et al., 2021) (4-shot), which is a dataset of high quality problems at the grade school math level, and (2) PIQA (Bisk et al., 2020) (0-shot), which is a commonsense reasoning dataset. \u2022 Knowledge Capacities: (1) TriviaQA (5 shot) (Joshi et al., 2017), which consists of Trivia enthusiast authored question-answer pairs, and (2) NaturalQuestions (5-shot) (Kwiatkowski et al., 2019), which is a QA corpus consists of queries issued to the Google search engine. Evaluation. For all benchmarks, we follow the test scripts of OpenCompass2. Specifically, on the multiple-choice tasks (MMLU, ARC-C, and PIQA), the option with the highest likelihood is selected to calculate the accuracy. On the free-form 2https://opencompass.org.cn/ Models TriviaQA NQ dev test dev test Individual Models LLaMA2-13B 72.74 73.57 21.56 28.67 Mistral-7B 70.47 72.58 19.94 27.67 InternLM-20B 64.11 65.90 19.50 26.09 Top-2 Ensemble DEEPEN 74.79 75.13 21.81 30.17 \u2206 +2.05 +1.56 +0.25 +1.50 Top-3 Ensemble PAIRRANKER \u2013 70.27 \u2013 26.9 GENFUSER \u2013 1.87 \u2013 0.17 DEEPEN 74.10 74.92 23.50 30.81 \u2206 +1.36 +1.35 +1.94 +2.14 Table 3: Results on knowledge-intensive QA Tasks. NQ refers to NaturalQuestion. generation tasks (GSM8K, TrivialQA and NaturalQuestions), we calculate the exact match (EM) accuracy. Individual models. Based on the key principles of ensemble learning (Sagi and Rokach, 2018), the synergy is created only in cases where individual models possess comparable performance to each other. Therefore, on each task type, we select three LLMs that have achieved promising results on the OpenCompass leaderboard and whose performance is as close to each other as possible. On the challenge reasoning task, we conduct experiments for the ensemble of two large-scale LLMs: LLaMA270B and Mixtral-8\u00d77B. Hyperparameters. In this work, we select all of the common tokens between LLMs as the anchor tokens to build the relative spaces, i.e., A = C. In the decoding of relative representation, we search the optimal relative learning rate (Eq. 9) from 0.0 to 0.3 with an interval of 0.05. We set the number of relative ensemble learning steps T = 5. Comparative methods. We also compare our ensemble framework DEEPEN with the selectionbased ensemble method PAIRRANKER, which is a reward model to score each response of LLMs and the fusion-based ensemble method GENFUSER, which is a generative model to fuse multiple candidate responses. Both models are from LLMBlender framework Jiang et al. (2023b) and are trained on the constructed instruction tuning dataset MixInstruct. 61.00 62.00 63.00 64.00 1k 2k 4k 8k 16k Full set 63.61 61.19 DeePEn Baseline 73.00 74.67 76.33 78.00 1k 2k 4k 8k 16k Full set 77.73 73.79 DeePEn Baseline (b) MMLU-Dev (a) ARC-C-Dev Figure 3: Effect of the number of anchor words. The x-axis indicates the number of anchor words randomly sampled from the common words. 4.2 Results The main results are shown in Tab 1,2, and 3, from which we have drawn the following observations: DEEPEN achieves consistent improvements over the individual models across all benchmarks. These results prove that our framework indeed enables collaboration between LLMs via aggregating the target semantics behind the output probability distributions. Specifically, the ensemble of top-2 models achieves the improvements of +1.5(NQ)\u223c+3.22(PIQA) on test sets and +0.25(NQ)\u223c+3.85(ARC-C) on validation sets in terms of accuracy. Through the ensemble of top-3 models, DEEPEN gains a further improvement over ensembling top-2 models. As shown in Tab. 1, the improvement is increased from +1.79 to +2.99 after the joining of the third-ranked model on ARC-C. Collaboration with models significantly worse could lead to degeneration. As the results in Tab 3, the ensemble of top-3 models underperforms than the one of top-2 models on TrivialQA. This degeneration is reasonable due to the significant performance gap (-6.68) between the thirdranked model (InternLM-20B) and the secondranked models (Mistral-7B) on this benchmark. DEEPEN shows excellent generalizability. As we can see, Both PAIRRANKER and GENFUSER perform poorly on each benchmark, which is caused by the serious distribution shift from their training data to the test data. Instead, DEEPEN works well under the training-free setting. Methods MMLU-Dev TrivialQA-Dev ACC \u2206 ACC \u2206 Baseline 61.19 \u2013 72.74 \u2013 DEEPEN 63.61 +2.42 74.79 +2.05 w/o. Rel-Norm 60.73 -0.46 72.95 +0.21 Table 4: Ablation study of normalization on the relative representation matrix to the ensembling performance on the development sets. Baseline refers to as the best single model on each benchmark. DEEPEN refers the performance of ensembling top-2 models in the benchmark. Percentage 0 0.1 0.2 0.3 0-0.1 0.2-0.3 0.4-0.5 0.03 0.13 0.31 0.31 0.21 0.01 0-0.1 0.2-0.3 0.4-0.5 0.14 0.29 0.28 0.22 0.07 0.01 Yi-6B Mistral-7B 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 1.0 1.0 Cosine Similarity Figure 4: Distance distribution to nearest neighbor words. The distance is measured by calculating the cosine similarity between words. 5 Analysis Towards better understanding DEEPEN, we conduct a series of analyses on the relative transformation and the algorithm of relative representation decoding. 5.1 Analysis on Relative Transformation Effect of anchor selection. We demonstrate the impact of different numbers of anchor words through experiments with the top-2 ensemble models on the MMLU and ARC-C benchmarks. As shown in Fig. 3, we observe that an increased number of anchor words does not necessarily translate into improved performance for LLMs in downstream tasks. Fortunately, selecting the full set of common words as anchors could achieve stable promising performance. Effect of normalization on relative representation matrix. To demonstrate the importance of normalization on the relative representation matrix to the ensemble performance, we conduct an ablation analysis. The result is shown in Tab. 4, the ensemble struggles to achieve improvements due to the ineffective representation of outlier words, i.e., words distant to other words. The proportion of outlier words can be derived from the distribuRELR (\u03b7) 0.05 0.10 0.15 0.20 0.25 0.30 MMLU +2.42 +1.57 +1.77 +1.96 +1.31 +1.31 TrivialQA +1.31 +2.05 +1.63 +1.94 +1.82 +1.26 Table 5: Sensitivity analysis of relative ensemble learning rate (RELR). We report the improvements of ensembling top-2 models over the best individual models. Baseline TrivQA NQ ARC-C MMLU TrivialQA 72.74 72.90 72.13 69.35 70.70 NQ 21.56 21.56 21.88 21.81 21.62 ARC-C 59.32 69.32 71.97 73.76 73.26 MMLU 55.10 59.98 61.02 61.86 61.42 Table 6: Cross-distribution validation of relative ensemble learning rate (\u03b7). We report the performance of ensembling LLaMA2-13B and Mistral-7B. Each row indicates the test set used to evaluate performance. Each column indicates the development set used to search the optimal value of RELR. tion of distance to nearest neighbor words, which is illustrated in Fig. 4. As illustrated, a remarkable proportion (> 30%) of words are distant from other words, i.e., cosine similarity to its nearest neighbor word is less than 0.3. Through the normalization operation, the target semantics that intend to emit outlier words could be prevented from becoming zero vectors by relative transformation. 5.2 Analysis of Reverse Transformation As described in \u00a73.4, the relative representation of the aggregated target semantic is transformed back to the absolute space of the main model via a search-based algorithm. There are two important factors (relative ensemble learning rate and number of search steps) in this reverse transformation, which are comprehensively analyzed, respectively. Analysis of relative ensemble learning rates. As shown in Tab. 5, the performance of DEEPEN is sensitive to the value of relative ensemble learning rate (\u03b7), which is abbreviated by RELR. This observation motivates us to measure the generality of this hyperparameter. Specifically, we illustrate the cross-distribution performance of the searched optimal value of \u03b7 in Tab. 6. As observed, the optimal value of RELR varies across different datasets, which suggests that the inverse transformation from relative space to absolute space requires adaptive mapping modes. 4.0E-07 8.7E-07 1.3E-06 1.8E-06 2.3E-06 \ud835\udf02=0.05 \ud835\udf02=0.10 \ud835\udf02=0.15 \ud835\udf02=0.20 \ud835\udf02=0.25 \ud835\udf02=0.30 63.25 62.50 61.75 0 1 3 5 10 63.61 61.19 Loss of Ensemble Learning Accuracy Relative Ensemble Learning Steps ( T ) Figure 5: Effect of different number of relative ensemble learning steps. Models MMLU-Dev ARC-C-Dev INDIV DEEPEN INDIV DEEPEN Yi-6B 61.19 63.61 (+2.42) 72.72 77.55 (+4.83) Mistral-7B 60.80 64.46 (+3.66) 73.88 77.73 (+3.85) Table 7: Performance of DEEPEN with choosing different main models on the development sets. INDIV refers to as individual models. The result of DeePEn indicates the performance of using the model of this row as the main model. Effect of iteration steps in relative ensemble learning. To give a deep view of the dynamics of the inverse transformation in DEEPEN, we report the performance change along with different numbers of relative ensemble learning steps (T). Besides, the dynamics of loss of relative ensemble learning (\u2113in Eq. 8)is also reported. As shown in Fig. 5, on the one hand, more steps of relative ensemble learning significantly lead to lower losses. However, the loss is hard to reach zero, i.e., underfitting. On the other hand, increasing the number of steps of relative ensemble learning will cause the performance to increase first and then decrease. The reason behind the performance drop could be that in the early stage of optimization, the focus of optimization is on updating the high-probability tokens. In the later stage of optimization, since the probabilities of all words will be adjusted equally, the low-probability tokens will be interfered with, thus affecting the performance. Therefore, it is recommended to set a modest value of step number (e.g., T = 5). Choice of main model. In the process of inverse transformation, DEEPEN maps the relative aggregated representation to the absolute space of the main model. Ideally, we expected the results of inverse transformation to keep invariant with the choice of main model. However, this objective is hard to achieve due to the underfitting observed in Models MMLU-Dev MMLU-Test INDIV VANIL DEEPEN INDIV VANIL DEEPEN LLaMA1 43.26 45.48 44.37 43.70 45.01 44.22 LLaMA2 42.28 45.94 42.99 45.31 Table 8: Comparison to vanilla prediction average (VANIL) on the ensemble of LLMs with the same vocabulary. the search process. Therefore, we illustrate the performance gap of choosing different main models in Tab. 7. As the results shown on ARC-C, changing the main model from the first-ranked Mistral-7B to the second-rank Yi-6B, the ensemble performance is decreased slightly from 77.73 to 77.55. Interestingly, changing the main model from the rank-1 Yi-6B to the rank-2 Mistral-7B on MMLU, the performance is actually improved from 63.63 to 64.46, which indicates that Mistral-7B benefits more than Yi-6B from collaboration. Even so, choosing different main models does not significantly affects the ensemble performance. 5.3 Comparison to Vanilla Prediction Average To compare our DEEPEN with vanilla prediction average, we conduct an experiment for ensembling two LLMs with the same vocabulary and comparable performance on MMLU, i.e., LLaMA2-7B and LLaMA1-13B. As shown in Tab. 8, the performance of DEEPEN is comparable, even better than, that of the vanilla prediction average. Theoretically, the performance of the vanilla prediction average is the performance upper-bound of DEEPEN. The reason that DEEPEN could excel over the vanilla one on MMLU is the under-fitting in the inverse transformation process, which leads to the weights to aggregate the target semantics of different models not being a uniform distribution (i.e., (0.5, 0.5)). For example, in Tab. 8, the weights for LLaMA1 and LLaMA2 could be (0.6, 0.4), where the weight of the main model is larger than the other model. In this work, we propose a training-free LLMs collaboration framework DEEPEN, which averages the probability distributions of models with heterogeneous vocabularies. To the best of our knowledge, we are the first to effectively enable the aggregation of heterogeneous probability distributions. We believe our research can motivate further research on the LLMs collaboration, model reuse, and knowledge distillation. In the future, we are planning to experiment with DEEPEN on the ensemble between LLMs and generative expert language models (e.g., machine translation models) due to their complementary knowledge. 8 Limitations Different from previous LLM ensemble methods, our DEEPEN is uniquely designed for the ensemble of white-box LLMs due to necessitating authorization of the output probability distribution. Besides, DEEPEN assigns equal weights to each individual model, which fails to deal with the interference caused by the poor-performing model. This problem motivates us to introduce ensemble strategies like stacking technology (Wolpert, 1992)."
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.14325v1",
+                    "title": "Improving Factuality and Reasoning in Language Models through Multiagent Debate",
+                    "abstract": "Large language models (LLMs) have demonstrated remarkable capabilities in\nlanguage generation, understanding, and few-shot learning in recent years. An\nextensive body of work has explored how their performance may be further\nimproved through the tools of prompting, ranging from verification,\nself-consistency, or intermediate scratchpads. In this paper, we present a\ncomplementary approach to improve language responses where multiple language\nmodel instances propose and debate their individual responses and reasoning\nprocesses over multiple rounds to arrive at a common final answer. Our findings\nindicate that this approach significantly enhances mathematical and strategic\nreasoning across a number of tasks. We also demonstrate that our approach\nimproves the factual validity of generated content, reducing fallacious answers\nand hallucinations that contemporary models are prone to. Our approach may be\ndirectly applied to existing black-box models and uses identical procedure and\nprompts for all tasks we investigate. Overall, our findings suggest that such\n\"society of minds\" approach has the potential to significantly advance the\ncapabilities of LLMs and pave the way for further breakthroughs in language\ngeneration and understanding.",
+                    "authors": "Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch",
+                    "published": "2023-05-23",
+                    "updated": "2023-05-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CV",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2306.02561v3",
+                    "title": "LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion",
+                    "abstract": "We present LLM-Blender, an ensembling framework designed to attain\nconsistently superior performance by leveraging the diverse strengths of\nmultiple open-source large language models (LLMs). Our framework consists of\ntwo modules: PairRanker and GenFuser, addressing the observation that optimal\nLLMs for different examples can significantly vary. PairRanker employs a\nspecialized pairwise comparison method to distinguish subtle differences\nbetween candidate outputs. It jointly encodes the input text and a pair of\ncandidates, using cross-attention encoders to determine the superior one. Our\nresults demonstrate that PairRanker exhibits the highest correlation with\nChatGPT-based ranking. Then, GenFuser aims to merge the top-ranked candidates,\ngenerating an improved output by capitalizing on their strengths and mitigating\ntheir weaknesses. To facilitate large-scale evaluation, we introduce a\nbenchmark dataset, MixInstruct, which is a mixture of multiple instruction\ndatasets featuring oracle pairwise comparisons. Our LLM-Blender significantly\noutperform individual LLMs and baseline methods across various metrics,\nestablishing a substantial performance gap.",
+                    "authors": "Dongfu Jiang, Xiang Ren, Bill Yuchen Lin",
+                    "published": "2023-06-05",
+                    "updated": "2023-06-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.15789v1",
+                    "title": "Large Language Model Routing with Benchmark Datasets",
+                    "abstract": "There is a rapidly growing number of open-source Large Language Models (LLMs)\nand benchmark datasets to compare them. While some models dominate these\nbenchmarks, no single model typically achieves the best accuracy in all tasks\nand use cases. In this work, we address the challenge of selecting the best LLM\nout of a collection of models for new tasks. We propose a new formulation for\nthe problem, in which benchmark datasets are repurposed to learn a \"router\"\nmodel for this LLM selection, and we show that this problem can be reduced to a\ncollection of binary classification tasks. We demonstrate the utility and\nlimitations of learning model routers from various benchmark datasets, where we\nconsistently improve performance upon using any single model for all tasks.",
+                    "authors": "Tal Shnitzer, Anthony Ou, M\u00edrian Silva, Kate Soule, Yuekai Sun, Justin Solomon, Neil Thompson, Mikhail Yurochkin",
+                    "published": "2023-09-27",
+                    "updated": "2023-09-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG",
+                        "I.2.7, I.2.6"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.14325v1",
+                    "title": "Improving Factuality and Reasoning in Language Models through Multiagent Debate",
+                    "abstract": "Large language models (LLMs) have demonstrated remarkable capabilities in\nlanguage generation, understanding, and few-shot learning in recent years. An\nextensive body of work has explored how their performance may be further\nimproved through the tools of prompting, ranging from verification,\nself-consistency, or intermediate scratchpads. In this paper, we present a\ncomplementary approach to improve language responses where multiple language\nmodel instances propose and debate their individual responses and reasoning\nprocesses over multiple rounds to arrive at a common final answer. Our findings\nindicate that this approach significantly enhances mathematical and strategic\nreasoning across a number of tasks. We also demonstrate that our approach\nimproves the factual validity of generated content, reducing fallacious answers\nand hallucinations that contemporary models are prone to. Our approach may be\ndirectly applied to existing black-box models and uses identical procedure and\nprompts for all tasks we investigate. Overall, our findings suggest that such\n\"society of minds\" approach has the potential to significantly advance the\ncapabilities of LLMs and pave the way for further breakthroughs in language\ngeneration and understanding.",
+                    "authors": "Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch",
+                    "published": "2023-05-23",
+                    "updated": "2023-05-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CV",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08692v1",
+                    "title": "Routing to the Expert: Efficient Reward-guided Ensemble of Large Language Models",
+                    "abstract": "The complementary potential of Large Language Models (LLM) assumes\noff-the-shelf LLMs have heterogeneous expertise in a wide range of domains and\ntasks so that an ensemble of LLMs can achieve consistently better performance.\nExisting ensemble methods for LLMs mainly focus on reward model ranking of\noutputs, leading to significant computation overhead. To combat this issue, we\nrevisit the complementary potential of LLMs and further elaborate it by mining\nlatent expertise with off-the-shelf reward models. We propose Zooter, a\nreward-guided routing method distilling rewards on training queries to train a\nrouting function, which can precisely distribute each query to the LLM with\nexpertise about it. We also integrate a tag-based label enhancement to mitigate\nnoise from uncertainty when using rewards as silver supervision. Zooter shows\ncomputation efficiency in inference as it introduces only a minor computation\noverhead of a routing function compared with reward model ranking methods. We\nevaluate Zooter on a comprehensive benchmark collection with 26 subsets on\ndifferent domains and tasks. Zooter outperforms the best single model on\naverage and ranks first on 44% of tasks, even surpassing multiple reward model\nranking methods.",
+                    "authors": "Keming Lu, Hongyi Yuan, Runji Lin, Junyang Lin, Zheng Yuan, Chang Zhou, Jingren Zhou",
+                    "published": "2023-11-15",
+                    "updated": "2023-11-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2209.15430v2",
+                    "title": "Relative representations enable zero-shot latent space communication",
+                    "abstract": "Neural networks embed the geometric structure of a data manifold lying in a\nhigh-dimensional space into latent representations. Ideally, the distribution\nof the data points in the latent space should depend only on the task, the\ndata, the loss, and other architecture-specific constraints. However, factors\nsuch as the random weights initialization, training hyperparameters, or other\nsources of randomness in the training phase may induce incoherent latent spaces\nthat hinder any form of reuse. Nevertheless, we empirically observe that, under\nthe same data and modeling choices, the angles between the encodings within\ndistinct latent spaces do not change. In this work, we propose the latent\nsimilarity between each sample and a fixed set of anchors as an alternative\ndata representation, demonstrating that it can enforce the desired invariances\nwithout any additional training. We show how neural architectures can leverage\nthese relative representations to guarantee, in practice, invariance to latent\nisometries and rescalings, effectively enabling latent space communication:\nfrom zero-shot model stitching to latent space comparison between diverse\nsettings. We extensively validate the generalization capability of our approach\non different datasets, spanning various modalities (images, text, graphs),\ntasks (e.g., classification, reconstruction) and architectures (e.g., CNNs,\nGCNs, transformers).",
+                    "authors": "Luca Moschella, Valentino Maiorca, Marco Fumero, Antonio Norelli, Francesco Locatello, Emanuele Rodol\u00e0",
+                    "published": "2022-09-30",
+                    "updated": "2023-03-07",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "68T07",
+                        "I.2.6"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.01542v1",
+                    "title": "Fusing Models with Complementary Expertise",
+                    "abstract": "Training AI models that generalize across tasks and domains has long been\namong the open problems driving AI research. The emergence of Foundation Models\nmade it easier to obtain expert models for a given task, but the heterogeneity\nof data that may be encountered at test time often means that any single expert\nis insufficient. We consider the Fusion of Experts (FoE) problem of fusing\noutputs of expert models with complementary knowledge of the data distribution\nand formulate it as an instance of supervised learning. Our method is\napplicable to both discriminative and generative tasks and leads to significant\nperformance improvements in image and text classification, text summarization,\nmultiple-choice QA, and automatic evaluation of generated text. We also extend\nour method to the \"frugal\" setting where it is desired to reduce the number of\nexpert model evaluations at test time.",
+                    "authors": "Hongyi Wang, Felipe Maia Polo, Yuekai Sun, Souvik Kundu, Eric Xing, Mikhail Yurochkin",
+                    "published": "2023-10-02",
+                    "updated": "2023-10-02",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.08189v1",
+                    "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+                    "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+                    "authors": "Karthik Sreedhar, Lydia Chilton",
+                    "published": "2024-02-13",
+                    "updated": "2024-02-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15198v2",
+                    "title": "Do LLM Agents Exhibit Social Behavior?",
+                    "abstract": "The advances of Large Language Models (LLMs) are expanding their utility in\nboth academic research and practical applications. Recent social science\nresearch has explored the use of these ``black-box'' LLM agents for simulating\ncomplex social systems and potentially substituting human subjects in\nexperiments. Our study delves into this emerging domain, investigating the\nextent to which LLMs exhibit key social interaction principles, such as social\nlearning, social preference, and cooperative behavior (indirect reciprocity),\nin their interactions with humans and other agents. We develop a framework for\nour study, wherein classical laboratory experiments involving human subjects\nare adapted to use LLM agents. This approach involves step-by-step reasoning\nthat mirrors human cognitive processes and zero-shot learning to assess the\ninnate preferences of LLMs. Our analysis of LLM agents' behavior includes both\nthe primary effects and an in-depth examination of the underlying mechanisms.\nFocusing on GPT-4, our analyses suggest that LLM agents appear to exhibit a\nrange of human-like social behaviors such as distributional and reciprocity\npreferences, responsiveness to group identity cues, engagement in indirect\nreciprocity, and social learning capabilities. However, our analysis also\nreveals notable differences: LLMs demonstrate a pronounced fairness preference,\nweaker positive reciprocity, and a more calculating approach in social learning\ncompared to humans. These insights indicate that while LLMs hold great promise\nfor applications in social science research, such as in laboratory experiments\nand agent-based modeling, the subtle behavioral differences between LLM agents\nand humans warrant further investigation. Careful examination and development\nof protocols in evaluating the social behaviors of LLMs are necessary before\ndirectly applying these models to emulate human behavior.",
+                    "authors": "Yan Leng, Yuan Yuan",
+                    "published": "2023-12-23",
+                    "updated": "2024-02-22",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.SI",
+                        "econ.GN",
+                        "q-fin.EC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.09397v1",
+                    "title": "Do Large GPT Models Discover Moral Dimensions in Language Representations? A Topological Study Of Sentence Embeddings",
+                    "abstract": "As Large Language Models are deployed within Artificial Intelligence systems,\nthat are increasingly integrated with human society, it becomes more important\nthan ever to study their internal structures. Higher level abilities of LLMs\nsuch as GPT-3.5 emerge in large part due to informative language\nrepresentations they induce from raw text data during pre-training on trillions\nof words. These embeddings exist in vector spaces of several thousand\ndimensions, and their processing involves mapping between multiple vector\nspaces, with total number of parameters on the order of trillions. Furthermore,\nthese language representations are induced by gradient optimization, resulting\nin a black box system that is hard to interpret. In this paper, we take a look\nat the topological structure of neuronal activity in the \"brain\" of Chat-GPT's\nfoundation language model, and analyze it with respect to a metric representing\nthe notion of fairness. We develop a novel approach to visualize GPT's moral\ndimensions. We first compute a fairness metric, inspired by social psychology\nliterature, to identify factors that typically influence fairness assessments\nin humans, such as legitimacy, need, and responsibility. Subsequently, we\nsummarize the manifold's shape using a lower-dimensional simplicial complex,\nwhose topology is derived from this metric. We color it with a heat map\nassociated with this fairness metric, producing human-readable visualizations\nof the high-dimensional sentence manifold. Our results show that sentence\nembeddings based on GPT-3.5 can be decomposed into two submanifolds\ncorresponding to fair and unfair moral judgments. This indicates that GPT-based\nlanguage models develop a moral dimension within their representation spaces\nand induce an understanding of fairness during their training process.",
+                    "authors": "Stephen Fitz",
+                    "published": "2023-09-17",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.06899v4",
+                    "title": "Flames: Benchmarking Value Alignment of LLMs in Chinese",
+                    "abstract": "The widespread adoption of large language models (LLMs) across various\nregions underscores the urgent need to evaluate their alignment with human\nvalues. Current benchmarks, however, fall short of effectively uncovering\nsafety vulnerabilities in LLMs. Despite numerous models achieving high scores\nand 'topping the chart' in these evaluations, there is still a significant gap\nin LLMs' deeper alignment with human values and achieving genuine harmlessness.\nTo this end, this paper proposes a value alignment benchmark named Flames,\nwhich encompasses both common harmlessness principles and a unique morality\ndimension that integrates specific Chinese values such as harmony. Accordingly,\nwe carefully design adversarial prompts that incorporate complex scenarios and\njailbreaking methods, mostly with implicit malice. By prompting 17 mainstream\nLLMs, we obtain model responses and rigorously annotate them for detailed\nevaluation. Our findings indicate that all the evaluated LLMs demonstrate\nrelatively poor performance on Flames, particularly in the safety and fairness\ndimensions. We also develop a lightweight specified scorer capable of scoring\nLLMs across multiple dimensions to efficiently evaluate new models on the\nbenchmark. The complexity of Flames has far exceeded existing benchmarks,\nsetting a new challenge for contemporary LLMs and highlighting the need for\nfurther alignment of LLMs. Our benchmark is publicly available at\nhttps://github.com/AIFlames/Flames.",
+                    "authors": "Kexin Huang, Xiangyang Liu, Qianyu Guo, Tianxiang Sun, Jiawei Sun, Yaru Wang, Zeyang Zhou, Yixu Wang, Yan Teng, Xipeng Qiu, Yingchun Wang, Dahua Lin",
+                    "published": "2023-11-12",
+                    "updated": "2024-04-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.06852v2",
+                    "title": "ChemLLM: A Chemical Large Language Model",
+                    "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+                    "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+                    "published": "2024-02-10",
+                    "updated": "2024-04-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.09447v2",
+                    "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+                    "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+                    "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+                    "published": "2023-11-15",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.06500v1",
+                    "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+                    "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+                    "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+                    "published": "2023-10-10",
+                    "updated": "2023-10-10",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07609v3",
+                    "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+                    "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+                    "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+                    "published": "2023-05-12",
+                    "updated": "2023-10-17",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.05668v1",
+                    "title": "CFaiRLLM: Consumer Fairness Evaluation in Large-Language Model Recommender System",
+                    "abstract": "In the evolving landscape of recommender systems, the integration of Large\nLanguage Models (LLMs) such as ChatGPT marks a new era, introducing the concept\nof Recommendation via LLM (RecLLM). While these advancements promise\nunprecedented personalization and efficiency, they also bring to the fore\ncritical concerns regarding fairness, particularly in how recommendations might\ninadvertently perpetuate or amplify biases associated with sensitive user\nattributes. In order to address these concerns, our study introduces a\ncomprehensive evaluation framework, CFaiRLLM, aimed at evaluating (and thereby\nmitigating) biases on the consumer side within RecLLMs.\n  Our research methodically assesses the fairness of RecLLMs by examining how\nrecommendations might vary with the inclusion of sensitive attributes such as\ngender, age, and their intersections, through both similarity alignment and\ntrue preference alignment. By analyzing recommendations generated under\ndifferent conditions-including the use of sensitive attributes in user\nprompts-our framework identifies potential biases in the recommendations\nprovided. A key part of our study involves exploring how different detailed\nstrategies for constructing user profiles (random, top-rated, recent) impact\nthe alignment between recommendations made without consideration of sensitive\nattributes and those that are sensitive-attribute-aware, highlighting the bias\nmechanisms within RecLLMs.\n  The findings in our study highlight notable disparities in the fairness of\nrecommendations, particularly when sensitive attributes are integrated into the\nrecommendation process, either individually or in combination. The analysis\ndemonstrates that the choice of user profile sampling strategy plays a\nsignificant role in affecting fairness outcomes, highlighting the complexity of\nachieving fair recommendations in the era of LLMs.",
+                    "authors": "Yashar Deldjoo, Tommaso di Noia",
+                    "published": "2024-03-08",
+                    "updated": "2024-03-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15478v1",
+                    "title": "A Group Fairness Lens for Large Language Models",
+                    "abstract": "The rapid advancement of large language models has revolutionized various\napplications but also raised crucial concerns about their potential to\nperpetuate biases and unfairness when deployed in social media contexts.\nEvaluating LLMs' potential biases and fairness has become crucial, as existing\nmethods rely on limited prompts focusing on just a few groups, lacking a\ncomprehensive categorical perspective. In this paper, we propose evaluating LLM\nbiases from a group fairness lens using a novel hierarchical schema\ncharacterizing diverse social groups. Specifically, we construct a dataset,\nGFair, encapsulating target-attribute combinations across multiple dimensions.\nIn addition, we introduce statement organization, a new open-ended text\ngeneration task, to uncover complex biases in LLMs. Extensive evaluations of\npopular LLMs reveal inherent safety concerns. To mitigate the biases of LLM\nfrom a group fairness perspective, we pioneer a novel chain-of-thought method\nGF-Think to mitigate biases of LLMs from a group fairness perspective.\nExperimental results demonstrate its efficacy in mitigating bias in LLMs to\nachieve fairness.",
+                    "authors": "Guanqun Bi, Lei Shen, Yuqiang Xie, Yanan Cao, Tiangang Zhu, Xiaodong He",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.02839v1",
+                    "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+                    "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+                    "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+                    "published": "2024-03-05",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.03838v2",
+                    "title": "RADAR: Robust AI-Text Detection via Adversarial Learning",
+                    "abstract": "Recent advances in large language models (LLMs) and the intensifying\npopularity of ChatGPT-like applications have blurred the boundary of\nhigh-quality text generation between humans and machines. However, in addition\nto the anticipated revolutionary changes to our technology and society, the\ndifficulty of distinguishing LLM-generated texts (AI-text) from human-generated\ntexts poses new challenges of misuse and fairness, such as fake content\ngeneration, plagiarism, and false accusations of innocent writers. While\nexisting works show that current AI-text detectors are not robust to LLM-based\nparaphrasing, this paper aims to bridge this gap by proposing a new framework\ncalled RADAR, which jointly trains a robust AI-text detector via adversarial\nlearning. RADAR is based on adversarial training of a paraphraser and a\ndetector. The paraphraser's goal is to generate realistic content to evade\nAI-text detection. RADAR uses the feedback from the detector to update the\nparaphraser, and vice versa. Evaluated with 8 different LLMs (Pythia, Dolly\n2.0, Palmyra, Camel, GPT-J, Dolly 1.0, LLaMA, and Vicuna) across 4 datasets,\nexperimental results show that RADAR significantly outperforms existing AI-text\ndetection methods, especially when paraphrasing is in place. We also identify\nthe strong transferability of RADAR from instruction-tuned LLMs to other LLMs,\nand evaluate the improved capability of RADAR via GPT-3.5-Turbo.",
+                    "authors": "Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho",
+                    "published": "2023-07-07",
+                    "updated": "2023-10-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04892v2",
+                    "title": "Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs",
+                    "abstract": "Recent works have showcased the ability of LLMs to embody diverse personas in\ntheir responses, exemplified by prompts like 'You are Yoda. Explain the Theory\nof Relativity.' While this ability allows personalization of LLMs and enables\nhuman behavior simulation, its effect on LLMs' capabilities remains unclear. To\nfill this gap, we present the first extensive study of the unintended\nside-effects of persona assignment on the ability of LLMs to perform basic\nreasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse\npersonas (e.g. an Asian person) spanning 5 socio-demographic groups. Our\nexperiments unveil that LLMs harbor deep rooted bias against various\nsocio-demographics underneath a veneer of fairness. While they overtly reject\nstereotypes when explicitly asked ('Are Black people less skilled at\nmathematics?'), they manifest stereotypical and erroneous presumptions when\nasked to answer questions while adopting a persona. These can be observed as\nabstentions in responses, e.g., 'As a Black person, I can't answer this\nquestion as it requires math knowledge', and generally result in a substantial\nperformance drop. Our experiments with ChatGPT-3.5 show that this bias is\nubiquitous - 80% of our personas demonstrate bias; it is significant - some\ndatasets show performance drops of 70%+; and can be especially harmful for\ncertain groups - some personas suffer statistically significant drops on 80%+\nof the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with\nGPT-4-Turbo showing the least but still a problematic amount of bias (evident\nin 42% of the personas). Further analysis shows that these persona-induced\nerrors can be hard-to-discern and hard-to-avoid. Our findings serve as a\ncautionary tale that the practice of assigning personas to LLMs - a trend on\nthe rise - can surface their deep-rooted biases and have unforeseeable and\ndetrimental side-effects.",
+                    "authors": "Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot",
+                    "published": "2023-11-08",
+                    "updated": "2024-01-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.07688v1",
+                    "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+                    "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+                    "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+                    "published": "2024-02-12",
+                    "updated": "2024-02-12",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.01937v1",
+                    "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+                    "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+                    "authors": "Cheng-Han Chiang, Hung-yi Lee",
+                    "published": "2023-05-03",
+                    "updated": "2023-05-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.01349v1",
+                    "title": "Fairness in Large Language Models: A Taxonomic Survey",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+                    "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+                    "published": "2024-03-31",
+                    "updated": "2024-03-31",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13862v2",
+                    "title": "A Trip Towards Fairness: Bias and De-Biasing in Large Language Models",
+                    "abstract": "Cheap-to-Build Very Large-Language Models (CtB-LLMs) with affordable training\nare emerging as the next big revolution in natural language processing and\nunderstanding. These CtB-LLMs are democratizing access to trainable Very\nLarge-Language Models (VLLMs) and, thus, may represent the building blocks of\nmany NLP systems solving downstream tasks. Hence, a little or a large bias in\nCtB-LLMs may cause huge harm. In this paper, we performed a large investigation\nof the bias of three families of CtB-LLMs, and we showed that debiasing\ntechniques are effective and usable. Indeed, according to current tests, the\nLLaMA and the OPT families have an important bias in gender, race, religion,\nand profession. In contrast to the analysis for other LLMs, we discovered that\nbias depends not on the number of parameters but on the perplexity. Finally,\nthe debiasing of OPT using LoRA reduces bias up to 4.12 points in the\nnormalized stereotype score.",
+                    "authors": "Leonardo Ranaldi, Elena Sofia Ruzzetti, Davide Venditti, Dario Onorati, Fabio Massimo Zanzotto",
+                    "published": "2023-05-23",
+                    "updated": "2023-08-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.11033v4",
+                    "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+                    "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+                    "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+                    "published": "2024-01-19",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.19118v1",
+                    "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate",
+                    "abstract": "Modern large language models (LLMs) like ChatGPT have shown remarkable\nperformance on general language tasks but still struggle on complex reasoning\ntasks, which drives the research on cognitive behaviors of LLMs to explore\nhuman-like problem-solving strategies. Along this direction, one representative\nstrategy is self-reflection, which asks an LLM to refine the solution with the\nfeedback generated by itself iteratively. However, our study shows that such\nreflection-style methods suffer from the Degeneration-of-Thought (DoT) problem:\nonce the LLM has established confidence in its solutions, it is unable to\ngenerate novel thoughts later through reflection even if its initial stance is\nincorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD)\nframework, in which multiple agents express their arguments in the state of\n\"tit for tat\" and a judge manages the debate process to obtain a final\nsolution. Clearly, our MAD framework encourages divergent thinking in LLMs\nwhich would be helpful for tasks that require deep levels of contemplation.\nExperiment results on two challenging datasets, commonsense machine translation\nand counter-intuitive arithmetic reasoning, demonstrate the effectiveness of\nour MAD framework. Extensive analyses suggest that the adaptive break of debate\nand the modest level of \"tit for tat\" state are required for MAD to obtain good\nperformance. Moreover, we find that LLMs might not be a fair judge if different\nLLMs are used for agents. Codes:\nhttps://github.com/Skytliang/Multi-Agents-Debate",
+                    "authors": "Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi",
+                    "published": "2023-05-30",
+                    "updated": "2023-05-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.07981v1",
+                    "title": "Manipulating Large Language Models to Increase Product Visibility",
+                    "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+                    "authors": "Aounon Kumar, Himabindu Lakkaraju",
+                    "published": "2024-04-11",
+                    "updated": "2024-04-11",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.01964v1",
+                    "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+                    "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+                    "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11764v1",
+                    "title": "ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs",
+                    "abstract": "Large Language models (LLMs), while powerful, exhibit harmful social biases.\nDebiasing is often challenging due to computational costs, data constraints,\nand potential degradation of multi-task language capabilities. This work\nintroduces a novel approach utilizing ChatGPT to generate synthetic training\ndata, aiming to enhance the debiasing of LLMs. We propose two strategies:\nTargeted Prompting, which provides effective debiasing for known biases but\nnecessitates prior specification of bias in question; and General Prompting,\nwhich, while slightly less effective, offers debiasing across various\ncategories. We leverage resource-efficient LLM debiasing using adapter tuning\nand compare the effectiveness of our synthetic data to existing debiasing\ndatasets. Our results reveal that: (1) ChatGPT can efficiently produce\nhigh-quality training data for debiasing other LLMs; (2) data produced via our\napproach surpasses existing datasets in debiasing performance while also\npreserving internal knowledge of a pre-trained LLM; and (3) synthetic data\nexhibits generalizability across categories, effectively mitigating various\nbiases, including intersectional ones. These findings underscore the potential\nof synthetic data in advancing the fairness of LLMs with minimal retraining\ncost.",
+                    "authors": "Pengrui Han, Rafal Kocielnik, Adhithya Saravanan, Roy Jiang, Or Sharir, Anima Anandkumar",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "68T50",
+                        "I.2.7; K.4.1"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.01262v2",
+                    "title": "Fairness Certification for Natural Language Processing and Large Language Models",
+                    "abstract": "Natural Language Processing (NLP) plays an important role in our daily lives,\nparticularly due to the enormous progress of Large Language Models (LLM).\nHowever, NLP has many fairness-critical use cases, e.g., as an expert system in\nrecruitment or as an LLM-based tutor in education. Since NLP is based on human\nlanguage, potentially harmful biases can diffuse into NLP systems and produce\nunfair results, discriminate against minorities or generate legal issues.\nHence, it is important to develop a fairness certification for NLP approaches.\nWe follow a qualitative research approach towards a fairness certification for\nNLP. In particular, we have reviewed a large body of literature on algorithmic\nfairness, and we have conducted semi-structured expert interviews with a wide\nrange of experts from that area. We have systematically devised six fairness\ncriteria for NLP, which can be further refined into 18 sub-categories. Our\ncriteria offer a foundation for operationalizing and testing processes to\ncertify fairness, both from the perspective of the auditor and the audited\norganization.",
+                    "authors": "Vincent Freiberger, Erik Buchmann",
+                    "published": "2024-01-02",
+                    "updated": "2024-01-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.04814v2",
+                    "title": "Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks",
+                    "abstract": "We introduce Syntax-Aware Fill-In-the-Middle (SAFIM), a new benchmark for\nevaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM)\ntask. This benchmark focuses on syntax-aware completions of program structures\nsuch as code blocks and conditional expressions, and includes 17,720 examples\nfrom multiple programming languages, sourced from recent code submissions after\nApril 2022 to minimize data contamination. SAFIM provides a robust framework\nwith various prompt designs and novel syntax-aware post-processing techniques,\nfacilitating accurate and fair comparisons across LLMs. Our comprehensive\nevaluation of 15 LLMs shows that FIM pretraining not only enhances FIM\nproficiency but also improves Left-to-Right (L2R) inference using LLMs. Our\nfindings challenge conventional beliefs and suggest that pretraining methods\nand data quality have more impact than model size. SAFIM thus serves as a\nfoundational platform for future research in effective pretraining strategies\nfor code LLMs. The evaluation toolkit and dataset are available at\nhttps://github.com/gonglinyuan/safim, and the leaderboard is available at\nhttps://safimbenchmark.com.",
+                    "authors": "Linyuan Gong, Sida Wang, Mostafa Elhoushi, Alvin Cheung",
+                    "published": "2024-03-07",
+                    "updated": "2024-04-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.02680v1",
+                    "title": "Large Language Models are Geographically Biased",
+                    "abstract": "Large Language Models (LLMs) inherently carry the biases contained in their\ntraining corpora, which can lead to the perpetuation of societal harm. As the\nimpact of these foundation models grows, understanding and evaluating their\nbiases becomes crucial to achieving fairness and accuracy. We propose to study\nwhat LLMs know about the world we live in through the lens of geography. This\napproach is particularly powerful as there is ground truth for the numerous\naspects of human life that are meaningfully projected onto geographic space\nsuch as culture, race, language, politics, and religion. We show various\nproblematic geographic biases, which we define as systemic errors in geospatial\npredictions. Initially, we demonstrate that LLMs are capable of making accurate\nzero-shot geospatial predictions in the form of ratings that show strong\nmonotonic correlation with ground truth (Spearman's $\\rho$ of up to 0.89). We\nthen show that LLMs exhibit common biases across a range of objective and\nsubjective topics. In particular, LLMs are clearly biased against locations\nwith lower socioeconomic conditions (e.g. most of Africa) on a variety of\nsensitive subjective topics such as attractiveness, morality, and intelligence\n(Spearman's $\\rho$ of up to 0.70). Finally, we introduce a bias score to\nquantify this and find that there is significant variation in the magnitude of\nbias across existing LLMs.",
+                    "authors": "Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon",
+                    "published": "2024-02-05",
+                    "updated": "2024-02-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.13343v1",
+                    "title": "Challenges and Contributing Factors in the Utilization of Large Language Models (LLMs)",
+                    "abstract": "With the development of large language models (LLMs) like the GPT series,\ntheir widespread use across various application scenarios presents a myriad of\nchallenges. This review initially explores the issue of domain specificity,\nwhere LLMs may struggle to provide precise answers to specialized questions\nwithin niche fields. The problem of knowledge forgetting arises as these LLMs\nmight find it hard to balance old and new information. The knowledge repetition\nphenomenon reveals that sometimes LLMs might deliver overly mechanized\nresponses, lacking depth and originality. Furthermore, knowledge illusion\ndescribes situations where LLMs might provide answers that seem insightful but\nare actually superficial, while knowledge toxicity focuses on harmful or biased\ninformation outputs. These challenges underscore problems in the training data\nand algorithmic design of LLMs. To address these issues, it's suggested to\ndiversify training data, fine-tune models, enhance transparency and\ninterpretability, and incorporate ethics and fairness training. Future\ntechnological trends might lean towards iterative methodologies, multimodal\nlearning, model personalization and customization, and real-time learning and\nfeedback mechanisms. In conclusion, future LLMs should prioritize fairness,\ntransparency, and ethics, ensuring they uphold high moral and ethical standards\nwhen serving humanity.",
+                    "authors": "Xiaoliang Chen, Liangbin Li, Le Chang, Yunhe Huang, Yuxuan Zhao, Yuxiao Zhang, Dinuo Li",
+                    "published": "2023-10-20",
+                    "updated": "2023-10-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.08836v2",
+                    "title": "Bias and Fairness in Chatbots: An Overview",
+                    "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+                    "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+                    "published": "2023-09-16",
+                    "updated": "2023-12-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.14473v1",
+                    "title": "The Ethics of ChatGPT in Medicine and Healthcare: A Systematic Review on Large Language Models (LLMs)",
+                    "abstract": "With the introduction of ChatGPT, Large Language Models (LLMs) have received\nenormous attention in healthcare. Despite their potential benefits, researchers\nhave underscored various ethical implications. While individual instances have\ndrawn much attention, the debate lacks a systematic overview of practical\napplications currently researched and ethical issues connected to them. Against\nthis background, this work aims to map the ethical landscape surrounding the\ncurrent stage of deployment of LLMs in medicine and healthcare. Electronic\ndatabases and preprint servers were queried using a comprehensive search\nstrategy. Studies were screened and extracted following a modified rapid review\napproach. Methodological quality was assessed using a hybrid approach. For 53\nrecords, a meta-aggregative synthesis was performed. Four fields of\napplications emerged and testify to a vivid exploration phase. Advantages of\nusing LLMs are attributed to their capacity in data analysis, personalized\ninformation provisioning, support in decision-making, mitigating information\nloss and enhancing information accessibility. However, we also identifies\nrecurrent ethical concerns connected to fairness, bias, non-maleficence,\ntransparency, and privacy. A distinctive concern is the tendency to produce\nharmful misinformation or convincingly but inaccurate content. A recurrent plea\nfor ethical guidance and human oversight is evident. Given the variety of use\ncases, it is suggested that the ethical guidance debate be reframed to focus on\ndefining what constitutes acceptable human oversight across the spectrum of\napplications. This involves considering diverse settings, varying potentials\nfor harm, and different acceptable thresholds for performance and certainty in\nhealthcare. In addition, a critical inquiry is necessary to determine the\nextent to which the current experimental use of LLMs is necessary and\njustified.",
+                    "authors": "Joschka Haltaufderheide, Robert Ranisch",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2304.03728v1",
+                    "title": "Interpretable Unified Language Checking",
+                    "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+                    "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+                    "published": "2023-04-07",
+                    "updated": "2023-04-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.10199v3",
+                    "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+                    "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+                    "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+                    "published": "2024-04-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.04489v1",
+                    "title": "De-amplifying Bias from Differential Privacy in Language Model Fine-tuning",
+                    "abstract": "Fairness and privacy are two important values machine learning (ML)\npractitioners often seek to operationalize in models. Fairness aims to reduce\nmodel bias for social/demographic sub-groups. Privacy via differential privacy\n(DP) mechanisms, on the other hand, limits the impact of any individual's\ntraining data on the resulting model. The trade-offs between privacy and\nfairness goals of trustworthy ML pose a challenge to those wishing to address\nboth. We show that DP amplifies gender, racial, and religious bias when\nfine-tuning large language models (LLMs), producing models more biased than\nones fine-tuned without DP. We find the cause of the amplification to be a\ndisparity in convergence of gradients across sub-groups. Through the case of\nbinary gender bias, we demonstrate that Counterfactual Data Augmentation (CDA),\na known method for addressing bias, also mitigates bias amplification by DP. As\na consequence, DP and CDA together can be used to fine-tune models while\nmaintaining both fairness and privacy.",
+                    "authors": "Sanjari Srivastava, Piotr Mardziel, Zhikhun Zhang, Archana Ahlawat, Anupam Datta, John C Mitchell",
+                    "published": "2024-02-07",
+                    "updated": "2024-02-07",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CR",
+                        "cs.CY",
+                        "stat.ME"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.10567v3",
+                    "title": "InSaAF: Incorporating Safety through Accuracy and Fairness | Are LLMs ready for the Indian Legal Domain?",
+                    "abstract": "Recent advancements in language technology and Artificial Intelligence have\nresulted in numerous Language Models being proposed to perform various tasks in\nthe legal domain ranging from predicting judgments to generating summaries.\nDespite their immense potential, these models have been proven to learn and\nexhibit societal biases and make unfair predictions. In this study, we explore\nthe ability of Large Language Models (LLMs) to perform legal tasks in the\nIndian landscape when social factors are involved. We present a novel metric,\n$\\beta$-weighted $\\textit{Legal Safety Score ($LSS_{\\beta}$)}$, which\nencapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs'\nsafety by considering its performance in the $\\textit{Binary Statutory\nReasoning}$ task and its fairness exhibition with respect to various axes of\ndisparities in the Indian society. Task performance and fairness scores of\nLLaMA and LLaMA--2 models indicate that the proposed $LSS_{\\beta}$ metric can\neffectively determine the readiness of a model for safe usage in the legal\nsector. We also propose finetuning pipelines, utilising specialised legal\ndatasets, as a potential method to mitigate bias and improve model safety. The\nfinetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\\beta}$,\nimproving their usability in the Indian legal domain. Our code is publicly\nreleased.",
+                    "authors": "Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru",
+                    "published": "2024-02-16",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00884v2",
+                    "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+                    "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+                    "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+                    "published": "2024-03-01",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.DB",
+                    "cats": [
+                        "cs.DB",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10149v2",
+                    "title": "A Survey on Fairness in Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have shown powerful performance and development\nprospects and are widely deployed in the real world. However, LLMs can capture\nsocial biases from unprocessed training data and propagate the biases to\ndownstream tasks. Unfair LLM systems have undesirable social impacts and\npotential harms. In this paper, we provide a comprehensive review of related\nresearch on fairness in LLMs. Considering the influence of parameter magnitude\nand training paradigm on research strategy, we divide existing fairness\nresearch into oriented to medium-sized LLMs under pre-training and fine-tuning\nparadigms and oriented to large-sized LLMs under prompting paradigms. First,\nfor medium-sized LLMs, we introduce evaluation metrics and debiasing methods\nfrom the perspectives of intrinsic bias and extrinsic bias, respectively. Then,\nfor large-sized LLMs, we introduce recent fairness research, including fairness\nevaluation, reasons for bias, and debiasing methods. Finally, we discuss and\nprovide insight on the challenges and future directions for the development of\nfairness in LLMs.",
+                    "authors": "Yingji Li, Mengnan Du, Rui Song, Xin Wang, Ying Wang",
+                    "published": "2023-08-20",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09606v1",
+                    "title": "Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey",
+                    "abstract": "Causal inference has shown potential in enhancing the predictive accuracy,\nfairness, robustness, and explainability of Natural Language Processing (NLP)\nmodels by capturing causal relationships among variables. The emergence of\ngenerative Large Language Models (LLMs) has significantly impacted various NLP\ndomains, particularly through their advanced reasoning capabilities. This\nsurvey focuses on evaluating and improving LLMs from a causal view in the\nfollowing areas: understanding and improving the LLMs' reasoning capacity,\naddressing fairness and safety issues in LLMs, complementing LLMs with\nexplanations, and handling multimodality. Meanwhile, LLMs' strong reasoning\ncapacities can in turn contribute to the field of causal inference by aiding\ncausal relationship discovery and causal effect estimations. This review\nexplores the interplay between causal inference frameworks and LLMs from both\nperspectives, emphasizing their collective potential to further the development\nof more advanced and equitable artificial intelligence systems.",
+                    "authors": "Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2206.13757v1",
+                    "title": "Flexible text generation for counterfactual fairness probing",
+                    "abstract": "A common approach for testing fairness issues in text-based classifiers is\nthrough the use of counterfactuals: does the classifier output change if a\nsensitive attribute in the input is changed? Existing counterfactual generation\nmethods typically rely on wordlists or templates, producing simple\ncounterfactuals that don't take into account grammar, context, or subtle\nsensitive attribute references, and could miss issues that the wordlist\ncreators had not considered. In this paper, we introduce a task for generating\ncounterfactuals that overcomes these shortcomings, and demonstrate how large\nlanguage models (LLMs) can be leveraged to make progress on this task. We show\nthat this LLM-based method can produce complex counterfactuals that existing\nmethods cannot, comparing the performance of various counterfactual generation\nmethods on the Civil Comments dataset and showing their value in evaluating a\ntoxicity classifier.",
+                    "authors": "Zee Fryer, Vera Axelrod, Ben Packer, Alex Beutel, Jilin Chen, Kellie Webster",
+                    "published": "2022-06-28",
+                    "updated": "2022-06-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.18569v1",
+                    "title": "Fairness of ChatGPT",
+                    "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+                    "authors": "Yunqi Li, Yongfeng Zhang",
+                    "published": "2023-05-22",
+                    "updated": "2023-05-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.14345v2",
+                    "title": "Bias Testing and Mitigation in LLM-based Code Generation",
+                    "abstract": "Utilizing state-of-the-art Large Language Models (LLMs), automatic code\ngeneration models play a pivotal role in enhancing the productivity of software\ndevelopment procedures. As the adoption of LLMs becomes more widespread in\nsoftware coding ecosystems, a pressing issue has emerged: does the generated\ncode contain social bias and unfairness, such as those related to age, gender,\nand race? This issue concerns the integrity, fairness, and ethical foundation\nof software applications that depend on the code generated by these models, yet\nis under-explored in the literature. This paper presents a novel bias testing\nframework that is specifically designed for code generation tasks. Based on\nthis framework, we conduct an extensive evaluation of the bias in code\ngenerated by five state-of-the-art LLMs. Our findings reveal that 20.29% to\n44.93% code functions generated by the models under study are biased when\nhandling bias sensitive tasks (i.e., tasks that involve sensitive attributes\nsuch as age and gender). This indicates that the existing LLMs can be unfair in\ncode generation, posing risks of unintended and harmful software behaviors. To\nmitigate bias for code generation models, we evaluate five bias mitigation\nprompt strategies, i.e., utilizing bias testing results to refine the code\n(zero-shot), one-, few-shot, and two Chain-of-Thought (CoT) prompts. Our\nevaluation results illustrate that these strategies are all effective in\nmitigating bias. Overall, one-shot and few-shot learning are the two most\neffective. For GPT-4, 80% to 90% code bias can be removed with one-shot\nlearning.",
+                    "authors": "Dong Huang, Qingwen Bu, Jie Zhang, Xiaofei Xie, Junjie Chen, Heming Cui",
+                    "published": "2023-09-03",
+                    "updated": "2024-01-09",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00306v1",
+                    "title": "Probing Explicit and Implicit Gender Bias through LLM Conditional Text Generation",
+                    "abstract": "Large Language Models (LLMs) can generate biased and toxic responses. Yet\nmost prior work on LLM gender bias evaluation requires predefined\ngender-related phrases or gender stereotypes, which are challenging to be\ncomprehensively collected and are limited to explicit bias evaluation. In\naddition, we believe that instances devoid of gender-related language or\nexplicit stereotypes in inputs can still induce gender bias in LLMs. Thus, in\nthis work, we propose a conditional text generation mechanism without the need\nfor predefined gender phrases and stereotypes. This approach employs three\ntypes of inputs generated through three distinct strategies to probe LLMs,\naiming to show evidence of explicit and implicit gender biases in LLMs. We also\nutilize explicit and implicit evaluation metrics to evaluate gender bias in\nLLMs under different strategies. Our experiments demonstrate that an increased\nmodel size does not consistently lead to enhanced fairness and all tested LLMs\nexhibit explicit and/or implicit gender bias, even when explicit gender\nstereotypes are absent in the inputs.",
+                    "authors": "Xiangjue Dong, Yibo Wang, Philip S. Yu, James Caverlee",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.15585v1",
+                    "title": "Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting",
+                    "abstract": "There exist both scalable tasks, like reading comprehension and\nfact-checking, where model performance improves with model size, and unscalable\ntasks, like arithmetic reasoning and symbolic reasoning, where model\nperformance does not necessarily improve with model size. Large language models\n(LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate\nincremental predictions even on unscalable tasks. Unfortunately, despite their\nexceptional reasoning abilities, LLMs tend to internalize and reproduce\ndiscriminatory societal biases. Whether CoT can provide discriminatory or\negalitarian rationalizations for the implicit information in unscalable tasks\nremains an open question.\n  In this study, we examine the impact of LLMs' step-by-step predictions on\ngender bias in unscalable tasks. For this purpose, we construct a benchmark for\nan unscalable task where the LLM is given a list of words comprising feminine,\nmasculine, and gendered occupational words, and is required to count the number\nof feminine and masculine words. In our CoT prompts, we require the LLM to\nexplicitly indicate whether each word in the word list is a feminine or\nmasculine before making the final predictions. With counting and handling the\nmeaning of words, this benchmark has characteristics of both arithmetic\nreasoning and symbolic reasoning. Experimental results in English show that\nwithout step-by-step prediction, most LLMs make socially biased predictions,\ndespite the task being as simple as counting words. Interestingly, CoT\nprompting reduces this unconscious social bias in LLMs and encourages fair\npredictions.",
+                    "authors": "Masahiro Kaneko, Danushka Bollegala, Naoaki Okazaki, Timothy Baldwin",
+                    "published": "2024-01-28",
+                    "updated": "2024-01-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.12150v1",
+                    "title": "Your Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One",
+                    "abstract": "The widespread adoption of large language models (LLMs) underscores the\nurgent need to ensure their fairness. However, LLMs frequently present dominant\nviewpoints while ignoring alternative perspectives from minority parties,\nresulting in potential biases. We hypothesize that these fairness-violating\nbehaviors occur because LLMs express their viewpoints using a human personality\nthat represents the majority of training data. In response to this, we validate\nthat prompting LLMs with specific roles can allow LLMs to express diverse\nviewpoints. Building on this insight and observation, we develop FairThinking,\na pipeline designed to automatically generate roles that enable LLMs to\narticulate diverse perspectives for fair expressions. To evaluate FairThinking,\nwe create a dataset with a thousand items covering three fairness-related\ntopics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to\ndemonstrate its superior performance.",
+                    "authors": "Tianlin Li, Xiaoyu Zhang, Chao Du, Tianyu Pang, Qian Liu, Qing Guo, Chao Shen, Yang Liu",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "I.2; J.4"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18130v2",
+                    "title": "DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues",
+                    "abstract": "Controversy is a reflection of our zeitgeist, and an important aspect to any\ndiscourse. The rise of large language models (LLMs) as conversational systems\nhas increased public reliance on these systems for answers to their various\nquestions. Consequently, it is crucial to systematically examine how these\nmodels respond to questions that pertaining to ongoing debates. However, few\nsuch datasets exist in providing human-annotated labels reflecting the\ncontemporary discussions. To foster research in this area, we propose a novel\nconstruction of a controversial questions dataset, expanding upon the publicly\nreleased Quora Question Pairs Dataset. This dataset presents challenges\nconcerning knowledge recency, safety, fairness, and bias. We evaluate different\nLLMs using a subset of this dataset, illuminating how they handle controversial\nissues and the stances they adopt. This research ultimately contributes to our\nunderstanding of LLMs' interaction with controversial issues, paving the way\nfor improvements in their comprehension and handling of complex societal\ndebates.",
+                    "authors": "David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein, Jason D. Williams",
+                    "published": "2023-10-27",
+                    "updated": "2023-11-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04205v2",
+                    "title": "Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves",
+                    "abstract": "Misunderstandings arise not only in interpersonal communication but also\nbetween humans and Large Language Models (LLMs). Such discrepancies can make\nLLMs interpret seemingly unambiguous questions in unexpected ways, yielding\nincorrect responses. While it is widely acknowledged that the quality of a\nprompt, such as a question, significantly impacts the quality of the response\nprovided by LLMs, a systematic method for crafting questions that LLMs can\nbetter comprehend is still underdeveloped. In this paper, we present a method\nnamed `Rephrase and Respond' (RaR), which allows LLMs to rephrase and expand\nquestions posed by humans and provide responses in a single prompt. This\napproach serves as a simple yet effective prompting method for improving\nperformance. We also introduce a two-step variant of RaR, where a rephrasing\nLLM first rephrases the question and then passes the original and rephrased\nquestions together to a different responding LLM. This facilitates the\neffective utilization of rephrased questions generated by one LLM with another.\nOur experiments demonstrate that our methods significantly improve the\nperformance of different models across a wide range to tasks. We further\nprovide a comprehensive comparison between RaR and the popular Chain-of-Thought\n(CoT) methods, both theoretically and empirically. We show that RaR is\ncomplementary to CoT and can be combined with CoT to achieve even better\nperformance. Our work not only contributes to enhancing LLM performance\nefficiently and effectively but also sheds light on a fair evaluation of LLM\ncapabilities. Data and codes are available at\nhttps://github.com/uclaml/Rephrase-and-Respond.",
+                    "authors": "Yihe Deng, Weitong Zhang, Zixiang Chen, Quanquan Gu",
+                    "published": "2023-11-07",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.03514v3",
+                    "title": "Can Large Language Models Transform Computational Social Science?",
+                    "abstract": "Large Language Models (LLMs) are capable of successfully performing many\nlanguage processing tasks zero-shot (without training data). If zero-shot LLMs\ncan also reliably classify and explain social phenomena like persuasiveness and\npolitical ideology, then LLMs could augment the Computational Social Science\n(CSS) pipeline in important ways. This work provides a road map for using LLMs\nas CSS tools. Towards this end, we contribute a set of prompting best practices\nand an extensive evaluation pipeline to measure the zero-shot performance of 13\nlanguage models on 25 representative English CSS benchmarks. On taxonomic\nlabeling tasks (classification), LLMs fail to outperform the best fine-tuned\nmodels but still achieve fair levels of agreement with humans. On free-form\ncoding tasks (generation), LLMs produce explanations that often exceed the\nquality of crowdworkers' gold references. We conclude that the performance of\ntoday's LLMs can augment the CSS research pipeline in two ways: (1) serving as\nzero-shot data annotators on human annotation teams, and (2) bootstrapping\nchallenging creative generation tasks (e.g., explaining the underlying\nattributes of a text). In summary, LLMs are posed to meaningfully participate\nin social science analysis in partnership with humans.",
+                    "authors": "Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, Diyi Yang",
+                    "published": "2023-04-12",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.15007v1",
+                    "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+                    "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+                    "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+                    "published": "2023-10-23",
+                    "updated": "2023-10-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.01769v1",
+                    "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+                    "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+                    "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+                    "published": "2024-05-02",
+                    "updated": "2024-05-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.19465v1",
+                    "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+                    "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+                    "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+                    "published": "2024-02-29",
+                    "updated": "2024-02-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18140v1",
+                    "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+                    "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+                    "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+                    "published": "2023-11-29",
+                    "updated": "2023-11-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.03192v1",
+                    "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+                    "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+                    "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+                    "published": "2024-04-04",
+                    "updated": "2024-04-04",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06003v1",
+                    "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+                    "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+                    "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18580v1",
+                    "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+                    "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+                    "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+                    "published": "2023-11-30",
+                    "updated": "2023-11-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15398v1",
+                    "title": "Fairness-Aware Structured Pruning in Transformers",
+                    "abstract": "The increasing size of large language models (LLMs) has introduced challenges\nin their training and inference. Removing model components is perceived as a\nsolution to tackle the large model sizes, however, existing pruning methods\nsolely focus on performance, without considering an essential aspect for the\nresponsible use of LLMs: model fairness. It is crucial to address the fairness\nof LLMs towards diverse groups, such as women, Black people, LGBTQ+, Jewish\ncommunities, among others, as they are being deployed and available to a wide\naudience. In this work, first, we investigate how attention heads impact\nfairness and performance in pre-trained transformer-based language models. We\nthen propose a novel method to prune the attention heads that negatively impact\nfairness while retaining the heads critical for performance, i.e. language\nmodeling capabilities. Our approach is practical in terms of time and\nresources, as it does not require fine-tuning the final pruned, and fairer,\nmodel. Our findings demonstrate a reduction in gender bias by 19%, 19.5%,\n39.5%, 34.7%, 23%, and 8% for DistilGPT-2, GPT-2, GPT-Neo of two different\nsizes, GPT-J, and Llama 2 models, respectively, in comparison to the biased\nmodel, with only a slight decrease in performance.",
+                    "authors": "Abdelrahman Zayed, Goncalo Mordido, Samira Shabanian, Ioana Baldini, Sarath Chandar",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.13840v1",
+                    "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+                    "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+                    "published": "2024-03-15",
+                    "updated": "2024-03-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.SI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05345v3",
+                    "title": "RTLLM: An Open-Source Benchmark for Design RTL Generation with Large Language Model",
+                    "abstract": "Inspired by the recent success of large language models (LLMs) like ChatGPT,\nresearchers start to explore the adoption of LLMs for agile hardware design,\nsuch as generating design RTL based on natural-language instructions. However,\nin existing works, their target designs are all relatively simple and in a\nsmall scale, and proposed by the authors themselves, making a fair comparison\namong different LLM solutions challenging. In addition, many prior works only\nfocus on the design correctness, without evaluating the design qualities of\ngenerated design RTL. In this work, we propose an open-source benchmark named\nRTLLM, for generating design RTL with natural language instructions. To\nsystematically evaluate the auto-generated design RTL, we summarized three\nprogressive goals, named syntax goal, functionality goal, and design quality\ngoal. This benchmark can automatically provide a quantitative evaluation of any\ngiven LLM-based solution. Furthermore, we propose an easy-to-use yet\nsurprisingly effective prompt engineering technique named self-planning, which\nproves to significantly boost the performance of GPT-3.5 in our proposed\nbenchmark.",
+                    "authors": "Yao Lu, Shang Liu, Qijun Zhang, Zhiyao Xie",
+                    "published": "2023-08-10",
+                    "updated": "2023-11-11",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.12090v1",
+                    "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+                    "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+                    "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+                    "published": "2023-05-20",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.07420v1",
+                    "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+                    "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+                    "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+                    "published": "2023-12-12",
+                    "updated": "2023-12-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.06056v1",
+                    "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+                    "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+                    "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+                    "published": "2023-12-11",
+                    "updated": "2023-12-11",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.11595v3",
+                    "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+                    "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+                    "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+                    "published": "2023-05-19",
+                    "updated": "2023-10-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.14607v2",
+                    "title": "Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications",
+                    "abstract": "Recent literature has suggested the potential of using large language models\n(LLMs) to make classifications for tabular tasks. However, LLMs have been shown\nto exhibit harmful social biases that reflect the stereotypes and inequalities\npresent in society. To this end, as well as the widespread use of tabular data\nin many high-stake applications, it is important to explore the following\nquestions: what sources of information do LLMs draw upon when making\nclassifications for tabular tasks; whether and to what extent are LLM\nclassifications for tabular data influenced by social biases and stereotypes;\nand what are the consequential implications for fairness?\n  Through a series of experiments, we delve into these questions and show that\nLLMs tend to inherit social biases from their training data which significantly\nimpact their fairness in tabular classification tasks. Furthermore, our\ninvestigations show that in the context of bias mitigation, though in-context\nlearning and finetuning have a moderate effect, the fairness metric gap between\ndifferent subgroups is still larger than that in traditional machine learning\nmodels, such as Random Forest and shallow Neural Networks. This observation\nemphasizes that the social biases are inherent within the LLMs themselves and\ninherited from their pretraining corpus, not only from the downstream task\ndatasets. Besides, we demonstrate that label-flipping of in-context examples\ncan significantly reduce biases, further highlighting the presence of inherent\nbias within LLMs.",
+                    "authors": "Yanchen Liu, Srishti Gautam, Jiaqi Ma, Himabindu Lakkaraju",
+                    "published": "2023-10-23",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15451v1",
+                    "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+                    "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+                    "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+                    "published": "2024-03-18",
+                    "updated": "2024-03-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.11483v1",
+                    "title": "Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in\nvarious NLP tasks. However, previous works have shown these models are\nsensitive towards prompt wording, and few-shot demonstrations and their order,\nposing challenges to fair assessment of these models. As these models become\nmore powerful, it becomes imperative to understand and address these\nlimitations. In this paper, we focus on LLMs robustness on the task of\nmultiple-choice questions -- commonly adopted task to study reasoning and\nfact-retrieving capability of LLMs. Investigating the sensitivity of LLMs\ntowards the order of options in multiple-choice questions, we demonstrate a\nconsiderable performance gap of approximately 13% to 75% in LLMs on different\nbenchmarks, when answer options are reordered, even when using demonstrations\nin a few-shot setting. Through a detailed analysis, we conjecture that this\nsensitivity arises when LLMs are uncertain about the prediction between the\ntop-2/3 choices, and specific options placements may favor certain prediction\nbetween those top choices depending on the question caused by positional bias.\nWe also identify patterns in top-2 choices that amplify or mitigate the model's\nbias toward option placement. We found that for amplifying bias, the optimal\nstrategy involves positioning the top two choices as the first and last\noptions. Conversely, to mitigate bias, we recommend placing these choices among\nthe adjacent options. To validate our conjecture, we conduct various\nexperiments and adopt two approaches to calibrate LLMs' predictions, leading to\nup to 8 percentage points improvement across different models and benchmarks.",
+                    "authors": "Pouya Pezeshkpour, Estevam Hruschka",
+                    "published": "2023-08-22",
+                    "updated": "2023-08-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.16343v2",
+                    "title": "Evaluating, Understanding, and Improving Constrained Text Generation for Large Language Models",
+                    "abstract": "Advancements in natural language generation (NLG) and large language models\n(LLMs) have led to proficient text generation in various tasks. However,\nintegrating intricate constraints into neural text generation, due to LLMs'\nopacity, remains challenging. This study investigates constrained text\ngeneration for LLMs, where predefined constraints are applied during LLM's\ngeneration process. Our research mainly focuses on mainstream open-source LLMs,\ncategorizing constraints into lexical, structural, and relation-based types. We\nalso present various benchmarks to facilitate fair evaluation. The study\naddresses some key research questions, including evaluating, understanding and\nimproving constrained text generation for LLMs. Results illuminate LLMs'\ncapacity and deficiency to incorporate constraints and provide insights for\nfuture developments in constrained text generation. Codes and datasets will be\nreleased upon acceptance.",
+                    "authors": "Xiang Chen, Xiaojun Wan",
+                    "published": "2023-10-25",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.08495v2",
+                    "title": "Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humans",
+                    "abstract": "Large language models (LLMs) are becoming pervasive in everyday life, yet\ntheir propensity to reproduce biases inherited from training data remains a\npressing concern. Prior investigations into bias in LLMs have focused on the\nassociation of social groups with stereotypical attributes. However, this is\nonly one form of human bias such systems may reproduce. We investigate a new\nform of bias in LLMs that resembles a social psychological phenomenon where\nsocially subordinate groups are perceived as more homogeneous than socially\ndominant groups. We had ChatGPT, a state-of-the-art LLM, generate texts about\nintersectional group identities and compared those texts on measures of\nhomogeneity. We consistently found that ChatGPT portrayed African, Asian, and\nHispanic Americans as more homogeneous than White Americans, indicating that\nthe model described racial minority groups with a narrower range of human\nexperience. ChatGPT also portrayed women as more homogeneous than men, but\nthese differences were small. Finally, we found that the effect of gender\ndiffered across racial/ethnic groups such that the effect of gender was\nconsistent within African and Hispanic Americans but not within Asian and White\nAmericans. We argue that the tendency of LLMs to describe groups as less\ndiverse risks perpetuating stereotypes and discriminatory behavior.",
+                    "authors": "Messi H. J. Lee, Jacob M. Montgomery, Calvin K. Lai",
+                    "published": "2024-01-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.13925v1",
+                    "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+                    "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+                    "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05374v2",
+                    "title": "Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment",
+                    "abstract": "Ensuring alignment, which refers to making models behave in accordance with\nhuman intentions [1,2], has become a critical task before deploying large\nlanguage models (LLMs) in real-world applications. For instance, OpenAI devoted\nsix months to iteratively aligning GPT-4 before its release [3]. However, a\nmajor challenge faced by practitioners is the lack of clear guidance on\nevaluating whether LLM outputs align with social norms, values, and\nregulations. This obstacle hinders systematic iteration and deployment of LLMs.\nTo address this issue, this paper presents a comprehensive survey of key\ndimensions that are crucial to consider when assessing LLM trustworthiness. The\nsurvey covers seven major categories of LLM trustworthiness: reliability,\nsafety, fairness, resistance to misuse, explainability and reasoning, adherence\nto social norms, and robustness. Each major category is further divided into\nseveral sub-categories, resulting in a total of 29 sub-categories.\nAdditionally, a subset of 8 sub-categories is selected for further\ninvestigation, where corresponding measurement studies are designed and\nconducted on several widely-used LLMs. The measurement results indicate that,\nin general, more aligned models tend to perform better in terms of overall\ntrustworthiness. However, the effectiveness of alignment varies across the\ndifferent trustworthiness categories considered. This highlights the importance\nof conducting more fine-grained analyses, testing, and making continuous\nimprovements on LLM alignment. By shedding light on these key dimensions of LLM\ntrustworthiness, this paper aims to provide valuable insights and guidance to\npractitioners in the field. Understanding and addressing these concerns will be\ncrucial in achieving reliable and ethically sound deployment of LLMs in various\napplications.",
+                    "authors": "Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, Hang Li",
+                    "published": "2023-08-10",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.17916v2",
+                    "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+                    "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+                    "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+                    "published": "2024-02-27",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.15215v1",
+                    "title": "Item-side Fairness of Large Language Model-based Recommendation System",
+                    "abstract": "Recommendation systems for Web content distribution intricately connect to\nthe information access and exposure opportunities for vulnerable populations.\nThe emergence of Large Language Models-based Recommendation System (LRS) may\nintroduce additional societal challenges to recommendation systems due to the\ninherent biases in Large Language Models (LLMs). From the perspective of\nitem-side fairness, there remains a lack of comprehensive investigation into\nthe item-side fairness of LRS given the unique characteristics of LRS compared\nto conventional recommendation systems. To bridge this gap, this study examines\nthe property of LRS with respect to item-side fairness and reveals the\ninfluencing factors of both historical users' interactions and inherent\nsemantic biases of LLMs, shedding light on the need to extend conventional\nitem-side fairness methods for LRS. Towards this goal, we develop a concise and\neffective framework called IFairLRS to enhance the item-side fairness of an\nLRS. IFairLRS covers the main stages of building an LRS with specifically\nadapted strategies to calibrate the recommendations of LRS. We utilize IFairLRS\nto fine-tune LLaMA, a representative LLM, on \\textit{MovieLens} and\n\\textit{Steam} datasets, and observe significant item-side fairness\nimprovements. The code can be found in\nhttps://github.com/JiangM-C/IFairLRS.git.",
+                    "authors": "Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, Xiangnan He",
+                    "published": "2024-02-23",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.18276v1",
+                    "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+                    "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+                    "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+                    "published": "2024-04-28",
+                    "updated": "2024-04-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "D.1; I.2"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10397v2",
+                    "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+                    "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+                    "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+                    "published": "2023-08-21",
+                    "updated": "2023-10-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.03852v2",
+                    "title": "FLM-101B: An Open LLM and How to Train It with $100K Budget",
+                    "abstract": "Large language models (LLMs) have achieved remarkable success in NLP and\nmultimodal tasks, among others. Despite these successes, two main challenges\nremain in developing LLMs: (i) high computational cost, and (ii) fair and\nobjective evaluations. In this paper, we report a solution to significantly\nreduce LLM training cost through a growth strategy. We demonstrate that a\n101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US\ndollars. Inspired by IQ tests, we also consolidate an additional range of\nevaluations on top of existing evaluations that focus on knowledge-oriented\nabilities. These IQ evaluations include symbolic mapping, rule understanding,\npattern mining, and anti-interference. Such evaluations minimize the potential\nimpact of memorization. Experimental results show that our model, named\nFLM-101B, trained with a budget of 100K US dollars, achieves performance\ncomparable to powerful and well-known models, e.g., GPT-3 and GLM-130B,\nespecially on the additional range of IQ evaluations. The checkpoint of\nFLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.",
+                    "authors": "Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Xuying Meng, Siqi Fan, Peng Han, Jing Li, Li Du, Bowen Qin, Zheng Zhang, Aixin Sun, Yequan Wang",
+                    "published": "2023-09-07",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00625v2",
+                    "title": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models",
+                    "abstract": "The burgeoning field of Large Language Models (LLMs), exemplified by\nsophisticated models like OpenAI's ChatGPT, represents a significant\nadvancement in artificial intelligence. These models, however, bring forth\nsubstantial challenges in the high consumption of computational, memory,\nenergy, and financial resources, especially in environments with limited\nresource capabilities. This survey aims to systematically address these\nchallenges by reviewing a broad spectrum of techniques designed to enhance the\nresource efficiency of LLMs. We categorize methods based on their optimization\nfocus: computational, memory, energy, financial, and network resources and\ntheir applicability across various stages of an LLM's lifecycle, including\narchitecture design, pretraining, finetuning, and system design. Additionally,\nthe survey introduces a nuanced categorization of resource efficiency\ntechniques by their specific resource types, which uncovers the intricate\nrelationships and mappings between various resources and corresponding\noptimization techniques. A standardized set of evaluation metrics and datasets\nis also presented to facilitate consistent and fair comparisons across\ndifferent models and techniques. By offering a comprehensive overview of the\ncurrent sota and identifying open research avenues, this survey serves as a\nfoundational reference for researchers and practitioners, aiding them in\ndeveloping more sustainable and efficient LLMs in a rapidly evolving landscape.",
+                    "authors": "Guangji Bai, Zheng Chai, Chen Ling, Shiyu Wang, Jiaying Lu, Nan Zhang, Tingwei Shi, Ziyang Yu, Mengdan Zhu, Yifei Zhang, Carl Yang, Yue Cheng, Liang Zhao",
+                    "published": "2024-01-01",
+                    "updated": "2024-01-04",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08472v1",
+                    "title": "Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models",
+                    "abstract": "Recently, work in NLP has shifted to few-shot (in-context) learning, with\nlarge language models (LLMs) performing well across a range of tasks. However,\nwhile fairness evaluations have become a standard for supervised methods,\nlittle is known about the fairness of LLMs as prediction systems. Further,\ncommon standard methods for fairness involve access to models weights or are\napplied during finetuning, which are not applicable in few-shot learning. Do\nLLMs exhibit prediction biases when used for standard NLP tasks? In this work,\nwe explore the effect of shots, which directly affect the performance of\nmodels, on the fairness of LLMs as NLP classification systems. We consider how\ndifferent shot selection strategies, both existing and new demographically\nsensitive methods, affect model fairness across three standard fairness\ndatasets. We discuss how future work can include LLM fairness evaluations.",
+                    "authors": "Carlos Aguirre, Kuleen Sasse, Isabel Cachola, Mark Dredze",
+                    "published": "2023-11-14",
+                    "updated": "2023-11-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02049v1",
+                    "title": "Post Turing: Mapping the landscape of LLM Evaluation",
+                    "abstract": "In the rapidly evolving landscape of Large Language Models (LLMs),\nintroduction of well-defined and standardized evaluation methodologies remains\na crucial challenge. This paper traces the historical trajectory of LLM\nevaluations, from the foundational questions posed by Alan Turing to the modern\nera of AI research. We categorize the evolution of LLMs into distinct periods,\neach characterized by its unique benchmarks and evaluation criteria. As LLMs\nincreasingly mimic human-like behaviors, traditional evaluation proxies, such\nas the Turing test, have become less reliable. We emphasize the pressing need\nfor a unified evaluation system, given the broader societal implications of\nthese models. Through an analysis of common evaluation methodologies, we\nadvocate for a qualitative shift in assessment approaches, underscoring the\nimportance of standardization and objective criteria. This work serves as a\ncall for the AI community to collaboratively address the challenges of LLM\nevaluation, ensuring their reliability, fairness, and societal benefit.",
+                    "authors": "Alexey Tikhonov, Ivan P. Yamshchikov",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00811v1",
+                    "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+                    "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+                    "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+                    "published": "2024-02-25",
+                    "updated": "2024-02-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15491v1",
+                    "title": "Open Source Conversational LLMs do not know most Spanish words",
+                    "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+                    "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.18502v1",
+                    "title": "Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification",
+                    "abstract": "Employing Large Language Models (LLM) in various downstream applications such\nas classification is crucial, especially for smaller companies lacking the\nexpertise and resources required for fine-tuning a model. Fairness in LLMs\nhelps ensure inclusivity, equal representation based on factors such as race,\ngender and promotes responsible AI deployment. As the use of LLMs has become\nincreasingly prevalent, it is essential to assess whether LLMs can generate\nfair outcomes when subjected to considerations of fairness. In this study, we\nintroduce a framework outlining fairness regulations aligned with various\nfairness definitions, with each definition being modulated by varying degrees\nof abstraction. We explore the configuration for in-context learning and the\nprocedure for selecting in-context demonstrations using RAG, while\nincorporating fairness rules into the process. Experiments conducted with\ndifferent LLMs indicate that GPT-4 delivers superior results in terms of both\naccuracy and fairness compared to other models. This work is one of the early\nattempts to achieve fairness in prediction tasks by utilizing LLMs through\nin-context learning.",
+                    "authors": "Garima Chhikara, Anurag Sharma, Kripabandhu Ghosh, Abhijnan Chakraborty",
+                    "published": "2024-02-28",
+                    "updated": "2024-02-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.17553v1",
+                    "title": "RuBia: A Russian Language Bias Detection Dataset",
+                    "abstract": "Warning: this work contains upsetting or disturbing content.\n  Large language models (LLMs) tend to learn the social and cultural biases\npresent in the raw pre-training data. To test if an LLM's behavior is fair,\nfunctional datasets are employed, and due to their purpose, these datasets are\nhighly language and culture-specific. In this paper, we address a gap in the\nscope of multilingual bias evaluation by presenting a bias detection dataset\nspecifically designed for the Russian language, dubbed as RuBia. The RuBia\ndataset is divided into 4 domains: gender, nationality, socio-economic status,\nand diverse, each of the domains is further divided into multiple fine-grained\nsubdomains. Every example in the dataset consists of two sentences with the\nfirst reinforcing a potentially harmful stereotype or trope and the second\ncontradicting it. These sentence pairs were first written by volunteers and\nthen validated by native-speaking crowdsourcing workers. Overall, there are\nnearly 2,000 unique sentence pairs spread over 19 subdomains in RuBia. To\nillustrate the dataset's purpose, we conduct a diagnostic evaluation of\nstate-of-the-art or near-state-of-the-art LLMs and discuss the LLMs'\npredisposition to social biases.",
+                    "authors": "Veronika Grigoreva, Anastasiia Ivanova, Ilseyar Alimova, Ekaterina Artemova",
+                    "published": "2024-03-26",
+                    "updated": "2024-03-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.05694v1",
+                    "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+                    "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+                    "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+                    "published": "2023-10-09",
+                    "updated": "2023-10-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.12736v1",
+                    "title": "Large Language Model Supply Chain: A Research Agenda",
+                    "abstract": "The rapid advancements in pre-trained Large Language Models (LLMs) and Large\nMultimodal Models (LMMs) have ushered in a new era of intelligent applications,\ntransforming fields ranging from natural language processing to content\ngeneration. The LLM supply chain represents a crucial aspect of the\ncontemporary artificial intelligence landscape. It encompasses the entire\nlifecycle of pre-trained models, from its initial development and training to\nits final deployment and application in various domains. This paper presents a\ncomprehensive overview of the LLM supply chain, highlighting its three core\nelements: 1) the model infrastructure, encompassing datasets and toolchain for\ntraining, optimization, and deployment; 2) the model lifecycle, covering\ntraining, testing, releasing, and ongoing maintenance; and 3) the downstream\napplication ecosystem, enabling the integration of pre-trained models into a\nwide range of intelligent applications. However, this rapidly evolving field\nfaces numerous challenges across these key components, including data privacy\nand security, model interpretability and fairness, infrastructure scalability,\nand regulatory compliance. Addressing these challenges is essential for\nharnessing the full potential of LLMs and ensuring their ethical and\nresponsible use. This paper provides a future research agenda for the LLM\nsupply chain, aiming at driving the continued advancement and responsible\ndeployment of these transformative LLMs.",
+                    "authors": "Shenao Wang, Yanjie Zhao, Xinyi Hou, Haoyu Wang",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.08780v1",
+                    "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+                    "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+                    "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+                    "published": "2023-10-13",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.03033v1",
+                    "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+                    "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+                    "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+                    "published": "2023-11-06",
+                    "updated": "2023-11-06",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.02219v1",
+                    "title": "FairEvalLLM. A Comprehensive Framework for Benchmarking Fairness in Large Language Model Recommender Systems",
+                    "abstract": "This paper presents a framework for evaluating fairness in recommender\nsystems powered by Large Language Models (RecLLMs), addressing the need for a\nunified approach that spans various fairness dimensions including sensitivity\nto user attributes, intrinsic fairness, and discussions of fairness based on\nunderlying benefits. In addition, our framework introduces counterfactual\nevaluations and integrates diverse user group considerations to enhance the\ndiscourse on fairness evaluation for RecLLMs.\n  Our key contributions include the development of a robust framework for\nfairness evaluation in LLM-based recommendations and a structured method to\ncreate \\textit{informative user profiles} from demographic data, historical\nuser preferences, and recent interactions. We argue that the latter is\nessential for enhancing personalization in such systems, especially in\ntemporal-driven scenarios. We demonstrate the utility of our framework through\npractical applications on two datasets, LastFM-1K and ML-1M. We conduct\nexperiments on a subsample of 80 users from each dataset, testing and assessing\nthe effectiveness of various prompt construction scenarios and in-context\nlearning, comprising more than 50 scenarios. This results in more than 4000\nrecommendations (80 * 50 = 4000). Our study reveals that while there are no\nsignificant unfairness issues in scenarios involving sensitive attributes, some\nconcerns remain. However, in terms of intrinsic fairness, which does not\ninvolve direct sensitivity, unfairness across demographic groups remains\nsignificant. The code and data used for this paper are available at:\n\\url{https://shorturl.at/awBFM}.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2024-05-03",
+                    "updated": "2024-05-03",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.08656v1",
+                    "title": "Linear Cross-document Event Coreference Resolution with X-AMR",
+                    "abstract": "Event Coreference Resolution (ECR) as a pairwise mention classification task\nis expensive both for automated systems and manual annotations. The task's\nquadratic difficulty is exacerbated when using Large Language Models (LLMs),\nmaking prompt engineering for ECR prohibitively costly. In this work, we\npropose a graphical representation of events, X-AMR, anchored around individual\nmentions using a \\textbf{cross}-document version of \\textbf{A}bstract\n\\textbf{M}eaning \\textbf{R}epresentation. We then linearize the ECR with a\nnovel multi-hop coreference algorithm over the event graphs. The event graphs\nsimplify ECR, making it a) LLM cost-effective, b) compositional and\ninterpretable, and c) easily annotated. For a fair assessment, we first enrich\nan existing ECR benchmark dataset with these event graphs using an\nannotator-friendly tool we introduce. Then, we employ GPT-4, the newest LLM by\nOpenAI, for these annotations. Finally, using the ECR algorithm, we assess\nGPT-4 against humans and analyze its limitations. Through this research, we aim\nto advance the state-of-the-art for efficient ECR and shed light on the\npotential shortcomings of current LLMs at this task. Code and annotations:\n\\url{https://github.com/ahmeshaf/gpt_coref}",
+                    "authors": "Shafiuddin Rehan Ahmed, George Arthur Baker, Evi Judge, Michael Regan, Kristin Wright-Bettner, Martha Palmer, James H. Martin",
+                    "published": "2024-03-25",
+                    "updated": "2024-03-25",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.04057v1",
+                    "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+                    "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+                    "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+                    "published": "2024-01-08",
+                    "updated": "2024-01-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18333v3",
+                    "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+                    "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+                    "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+                    "published": "2023-10-20",
+                    "updated": "2023-12-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14804v1",
+                    "title": "Use large language models to promote equity",
+                    "abstract": "Advances in large language models (LLMs) have driven an explosion of interest\nabout their societal impacts. Much of the discourse around how they will impact\nsocial equity has been cautionary or negative, focusing on questions like \"how\nmight LLMs be biased and how would we mitigate those biases?\" This is a vital\ndiscussion: the ways in which AI generally, and LLMs specifically, can entrench\nbiases have been well-documented. But equally vital, and much less discussed,\nis the more opportunity-focused counterpoint: \"what promising applications do\nLLMs enable that could promote equity?\" If LLMs are to enable a more equitable\nworld, it is not enough just to play defense against their biases and failure\nmodes. We must also go on offense, applying them positively to equity-enhancing\nuse cases to increase opportunities for underserved groups and reduce societal\ndiscrimination. There are many choices which determine the impact of AI, and a\nfundamental choice very early in the pipeline is the problems we choose to\napply it to. If we focus only later in the pipeline -- making LLMs marginally\nmore fair as they facilitate use cases which intrinsically entrench power -- we\nwill miss an important opportunity to guide them to equitable impacts. Here, we\nhighlight the emerging potential of LLMs to promote equity by presenting four\nnewly possible, promising research directions, while keeping risks and\ncautionary points in clear view.",
+                    "authors": "Emma Pierson, Divya Shanmugam, Rajiv Movva, Jon Kleinberg, Monica Agrawal, Mark Dredze, Kadija Ferryman, Judy Wawira Gichoya, Dan Jurafsky, Pang Wei Koh, Karen Levy, Sendhil Mullainathan, Ziad Obermeyer, Harini Suresh, Keyon Vafa",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-22",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.14208v2",
+                    "title": "Content Conditional Debiasing for Fair Text Embedding",
+                    "abstract": "Mitigating biases in machine learning models has gained increasing attention\nin Natural Language Processing (NLP). Yet, only a few studies focus on fair\ntext embeddings, which are crucial yet challenging for real-world applications.\nIn this paper, we propose a novel method for learning fair text embeddings. We\nachieve fairness while maintaining utility trade-off by ensuring conditional\nindependence between sensitive attributes and text embeddings conditioned on\nthe content. Specifically, we enforce that embeddings of texts with different\nsensitive attributes but identical content maintain the same distance toward\nthe embedding of their corresponding neutral text. Furthermore, we address the\nissue of lacking proper training data by using Large Language Models (LLMs) to\naugment texts into different sensitive groups. Our extensive evaluations\ndemonstrate that our approach effectively improves fairness while preserving\nthe utility of embeddings, representing a pioneering effort in achieving\nconditional independence for fair text embeddings.",
+                    "authors": "Wenlong Deng, Blair Chen, Xiaoxiao Li, Christos Thrampoulidis",
+                    "published": "2024-02-22",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.11761v1",
+                    "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+                    "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2023-07-14",
+                    "updated": "2023-07-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.08517v1",
+                    "title": "Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forward",
+                    "abstract": "While Large Language Models (LLMs) have seen widespread applications across\nnumerous fields, their limited interpretability poses concerns regarding their\nsafe operations from multiple aspects, e.g., truthfulness, robustness, and\nfairness. Recent research has started developing quality assurance methods for\nLLMs, introducing techniques such as offline detector-based or uncertainty\nestimation methods. However, these approaches predominantly concentrate on\npost-generation analysis, leaving the online safety analysis for LLMs during\nthe generation phase an unexplored area. To bridge this gap, we conduct in this\nwork a comprehensive evaluation of the effectiveness of existing online safety\nanalysis methods on LLMs. We begin with a pilot study that validates the\nfeasibility of detecting unsafe outputs in the early generation process.\nFollowing this, we establish the first publicly available benchmark of online\nsafety analysis for LLMs, including a broad spectrum of methods, models, tasks,\ndatasets, and evaluation metrics. Utilizing this benchmark, we extensively\nanalyze the performance of state-of-the-art online safety analysis methods on\nboth open-source and closed-source LLMs. This analysis reveals the strengths\nand weaknesses of individual methods and offers valuable insights into\nselecting the most appropriate method based on specific application scenarios\nand task requirements. Furthermore, we also explore the potential of using\nhybridization methods, i.e., combining multiple methods to derive a collective\nsafety conclusion, to enhance the efficacy of online safety analysis for LLMs.\nOur findings indicate a promising direction for the development of innovative\nand trustworthy quality assurance methodologies for LLMs, facilitating their\nreliable deployments across diverse domains.",
+                    "authors": "Xuan Xie, Jiayang Song, Zhehua Zhou, Yuheng Huang, Da Song, Lei Ma",
+                    "published": "2024-04-12",
+                    "updated": "2024-04-12",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.13095v1",
+                    "title": "Enhancing Logical Reasoning in Large Language Models to Facilitate Legal Applications",
+                    "abstract": "Language serves as a vehicle for conveying thought, enabling communication\namong individuals. The ability to distinguish between diverse concepts,\nidentify fairness and injustice, and comprehend a range of legal notions\nfundamentally relies on logical reasoning. Large Language Models (LLMs) attempt\nto emulate human language understanding and generation, but their competency in\nlogical reasoning remains limited. This paper seeks to address the\nphilosophical question: How can we effectively teach logical reasoning to LLMs\nwhile maintaining a deep understanding of the intricate relationship between\nlanguage and logic? By focusing on bolstering LLMs' capabilities in logical\nreasoning, we aim to expand their applicability in law and other\nlogic-intensive disciplines. To this end, we propose a Reinforcement Learning\nfrom Logical Feedback (RLLF) approach, which serves as a potential framework\nfor refining LLMs' reasoning capacities. Through RLLF and a revised evaluation\nmethodology, we explore new avenues for research in this domain and contribute\nto the development of LLMs capable of handling complex legal reasoning tasks\nwhile acknowledging the fundamental connection between language and logic.",
+                    "authors": "Ha-Thanh Nguyen, Wachara Fungwacharakorn, Ken Satoh",
+                    "published": "2023-11-22",
+                    "updated": "2023-11-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.07884v2",
+                    "title": "Fair Abstractive Summarization of Diverse Perspectives",
+                    "abstract": "People from different social and demographic groups express diverse\nperspectives and conflicting opinions on a broad set of topics such as product\nreviews, healthcare, law, and politics. A fair summary should provide a\ncomprehensive coverage of diverse perspectives without underrepresenting\ncertain groups. However, current work in summarization metrics and Large\nLanguage Models (LLMs) evaluation has not explored fair abstractive\nsummarization. In this paper, we systematically investigate fair abstractive\nsummarization for user-generated data. We first formally define fairness in\nabstractive summarization as not underrepresenting perspectives of any groups\nof people, and we propose four reference-free automatic metrics by measuring\nthe differences between target and source perspectives. We evaluate nine LLMs,\nincluding three GPT models, four LLaMA models, PaLM 2, and Claude, on six\ndatasets collected from social media, online reviews, and recorded transcripts.\nExperiments show that both the model-generated and the human-written reference\nsummaries suffer from low fairness. We conduct a comprehensive analysis of the\ncommon factors influencing fairness and propose three simple but effective\nmethods to alleviate unfair summarization. Our dataset and code are available\nat https://github.com/psunlpgroup/FairSumm.",
+                    "authors": "Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang",
+                    "published": "2023-11-14",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.09219v5",
+                    "title": "\"Kelly is a Warm Person, Joseph is a Role Model\": Gender Biases in LLM-Generated Reference Letters",
+                    "abstract": "Large Language Models (LLMs) have recently emerged as an effective tool to\nassist individuals in writing various types of content, including professional\ndocuments such as recommendation letters. Though bringing convenience, this\napplication also introduces unprecedented fairness concerns. Model-generated\nreference letters might be directly used by users in professional scenarios. If\nunderlying biases exist in these model-constructed letters, using them without\nscrutinization could lead to direct societal harms, such as sabotaging\napplication success rates for female applicants. In light of this pressing\nissue, it is imminent and necessary to comprehensively study fairness issues\nand associated harms in this real-world use case. In this paper, we critically\nexamine gender biases in LLM-generated reference letters. Drawing inspiration\nfrom social science findings, we design evaluation methods to manifest biases\nthrough 2 dimensions: (1) biases in language style and (2) biases in lexical\ncontent. We further investigate the extent of bias propagation by analyzing the\nhallucination bias of models, a term that we define to be bias exacerbation in\nmodel-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs-\nChatGPT and Alpaca, we reveal significant gender biases in LLM-generated\nrecommendation letters. Our findings not only warn against using LLMs for this\napplication without scrutinization, but also illuminate the importance of\nthoroughly studying hidden biases and harms in LLM-generated professional\ndocuments.",
+                    "authors": "Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng",
+                    "published": "2023-10-13",
+                    "updated": "2023-12-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00588v1",
+                    "title": "Fairness in Serving Large Language Models",
+                    "abstract": "High-demand LLM inference services (e.g., ChatGPT and BARD) support a wide\nrange of requests from short chat conversations to long document reading. To\nensure that all client requests are processed fairly, most major LLM inference\nservices have request rate limits, to ensure that no client can dominate the\nrequest queue. However, this rudimentary notion of fairness also results in\nunder-utilization of the resources and poor client experience when there is\nspare capacity. While there is a rich literature on fair scheduling, serving\nLLMs presents new challenges due to their unpredictable request lengths and\ntheir unique batching characteristics on parallel accelerators. This paper\nintroduces the definition of LLM serving fairness based on a cost function that\naccounts for the number of input and output tokens processed. To achieve\nfairness in serving, we propose a novel scheduling algorithm, the Virtual Token\nCounter (VTC), a fair scheduler based on the continuous batching mechanism. We\nprove a 2x tight upper bound on the service difference between two backlogged\nclients, adhering to the requirement of work-conserving. Through extensive\nexperiments, we demonstrate the superior performance of VTC in ensuring\nfairness, especially in contrast to other baseline methods, which exhibit\nshortcomings under various conditions.",
+                    "authors": "Ying Sheng, Shiyi Cao, Dacheng Li, Banghua Zhu, Zhuohan Li, Danyang Zhuo, Joseph E. Gonzalez, Ion Stoica",
+                    "published": "2023-12-31",
+                    "updated": "2023-12-31",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.PF"
+                    ],
+                    "category": "LLM Fairness"
+                }
+            ],
+            [
+                {
+                    "url": "http://arxiv.org/abs/2404.14462v2",
+                    "title": "Towards smaller, faster decoder-only transformers: Architectural variants and their implications",
+                    "abstract": "Research on Large Language Models (LLMs) has recently seen exponential\ngrowth, largely focused on transformer-based architectures, as introduced by\n[1] and further advanced by the decoder-only variations in [2]. Contemporary\nstudies typically aim to improve model capabilities by increasing both the\narchitecture's complexity and the volume of training data. However, research\nexploring how to reduce model sizes while maintaining performance is limited.\nThis study introduces three modifications to the decoder-only transformer\narchitecture: ParallelGPT (p-gpt), LinearlyCompressedGPT (lc-gpt), and\nConvCompressedGPT (cc-gpt). These variants achieve comparable performance to\nconventional architectures in code generation tasks while benefiting from\nreduced model sizes and faster training times. We open-source the model weights\nand codebase to support future research and development in this domain.",
+                    "authors": "Sathya Krishnan Suresh, Shunmugapriya P",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-24",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "label": "Original Paper",
+                    "paper_cat": "LLM Fairness",
+                    "gt": "The development of Large Language Models (LLMs) has been significantly driven by advancements in the transformer architecture, first introduced by [1] and extended in various decoder-only variations such as those by [2]. These models have traditionally focused on scaling up through increased architectural complexity and extensive training datasets. However, the need for more sizeand compute-efficient models has led to innovative approaches in model design, aiming to retain performance while reducing resource demands. Our study introduces three novel variants of the decoder-only transformer architecture aimed at efficiency: ParallelGPT (p-gpt), LinearlyCompressedGPT (lc-gpt), and ConvCompressedGPT (cc-gpt). These models demonstrate performance on par with traditional architectures in code generation tasks, but with the advantage of smaller model sizes and faster training times. We provide open-source access to both the model weights and the complete codebase to facilitate further exploration in this area. In parallel, the Funnel-Transformer [3] proposes an architecture that enhances efficiency by progressively compressing the sequence of hidden states, significantly lowering computational costs while potentially increasing model capacity. Unlike traditional models with constant sequence length, the Funnel-Transformer uses a pooling mechanism to reduce 2 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications sequence length across layers, which conserves computational resources and allows for model depth or width expansion without additional costs. Similarly, the \"MobiLlama\" [4] study presents a small language model optimized for resource-constrained environments. This model incorporates a parameter-sharing scheme within transformer blocks and focuses on reducing both pretraining and deployment costs, aligning well with our objectives of efficiency and reduced resource usage. Further contributing to the field, Grouped Query Attention [5] restructures the attention mechanism by grouping inputs, allowing attention within these confines and reducing complexity from quadratic to linear. This facilitates handling of longer sequences and larger datasets. Conversely, Multi-Query Attention [6] extends traditional attention mechanisms by allowing multiple queries within a single head, enhancing the model\u2019s ability to distill diverse information within the same layer and enriching its expressive capabilities. These collective advancements depict a clear trajectory towards not only enhancing the performance and scalability of LLMs but also towards making these models more adaptable to constraints of size and computational affordability. 3 Architectural modifications In this section, we introduce three novel architectures derived from the traditional GPT architecture to address various limitations in training and inference. These architectures are designed to enable faster training and inference, overcome common limitations encountered with quantized models, and facilitate local execution without specialized hardware. The three proposed architectures are ParallelGPT (p-gpt), LinearlyCompressedGPT (lc-gpt), and ConvCompressedGPT (cc-gpt). 3.1 ParallelGPT The N decoder blocks in a gpt architecture are stacked vertically on top of each other and the input to each decoder block will be the output of the previous decoder block. It can be seen here that the time taken for the input to go through this architecture will increase as the number of decoder blocks increase and the dimensionality of the model also has to increase to make sure that the information in the tokens is propagated through the blocks. Recent studies have also shown that the deeper layers [7] have little effect on the predictions made by the model. Both of these limitations can be overcome by splitting the decoder part of the gpt architecture into two parts (or more, which is for future research), each having an equal number of decoder blocks (N/2). To train a model of this architecture the dimensionality of the embedding model was made to be twice the dimensionality of the decoder (DMODEL). Hence, the output vectors of the embedding model will be of dimensions DMODEL \u22172 out of which the first half of each vector is sent to one block and the second half to the other block. This is done to ensure that each block learns a knowledge that is different from the other block. The outputs of the two blocks are combined using a weighted sum and are then fed to the classification head. Equations 1, 2 and 3 give a mathematical description of the above and 3 represents the architecture. x = embedding(tokens), x \u2208RDMODEL\u00d72 (1) x1, x2 = x[:, :, : DMODEL], x[:, :, DMODEL + 1 :], x1, x2 \u2208RDMODEL (2) probability = \u03c3(linear(w \u00b7 decoder1(x1) + (1 \u2212w) \u00b7 decoder2(x2))) (3) Training a model of this architecture has the following benefits 1. faster training, as each block can be trained on a separate compute node in parallel, 2. during inference the block with the lesser weight (w in 3) can be dropped for faster inference but this will result in a slight reduction in the performance of the model, 3. the number of parallel blocks can be increased further which we hypothesize, might produce better results and an even faster training. However with increasing the number of parallel blocks the dimensionality of the embedding also has to be increased to make sure that each block learns a different knowledge, which will increase the number of parameters in the model. This increase in the number of parameters will only be a disadvantage if we don\u2019t decide to drop a few blocks during inference. 3.2 LinearlyCompressedGPT We hypothesize that, much of the information that can be learned from the embeddings of each token might happen in the initial few blocks of the decoder (depending on the dimensions of the model), which brings into question the need for each block having the same dimensions. If each of the N decoder block has the same dimension this results in the model having much more parameters than that is actually needed to achieve the desired performance. 3 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications Figure 3: ParallelGPT Figure 4: LinearlyCompressedGPT Figure 5: ConvCompressedGPT To this end, we propose an architecture in which the dimension of the decoder blocks keeps reducing by half as the embeddings pass through the architecture. This concept is inspired by the architectural design of various image classification convolutional networks (CNNs), where the dimensions of an image are systematically reduced as it passes through successive layers of the network. To make sure that the dimensions of the output vectors of a decoder block matches up to the dimensions of the successive decoder block we introduce a dense (linear) layer after two decoder blocks of the same dimension to map the vectors to half of the original dimension. A dense layer is placed after every couple of decoder blocks, since placing them after each decoder block will produce an output vector of a small dimension from which the classification head will have a hard time to predict the next token. The architecture for lc-gpt is presented in fig. 4 and eq. 4, 5 represent the equations of the architecture discussed. x = decoder2(decoder1(x)), x \u2208RDMODEL (4) x = linear(x), x \u2208RDMODEL/2 (5) The lc-gpt architecture can reduce the number of parameters in the model in terms of millions or billions depending on the size of the original gpt architecture with the same number of layers and dimensions. The extra parameters introduced by the dense layers will be negligible compared to the amount of parameters reduced by reducing the dimensions of the decoder blocks. This reduction in dimension also forces the deeper decoder blocks to perform better with the smaller dimension vectors that are fed to it. 3.3 ConvCompressedGPT ConvCompressedGPT builds on the concept of LinearlyCompressedGPT by replacing the dense layers with 1D convolutional layers. This architecture retains the advantages of LinearlyCompressedGPT while leveraging the benefits of convolutional layers, such as weight sharing and improved positional pattern recognition. Convolutional layers apply filters across sequence positions, effectively capturing positional dependencies and patterns. This approach may enhance the model\u2019s ability to predict the next token by incorporating contextual information across the sequence. The architectural design for ConvCompressedGPT is presented in Figure 5. 4 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications The ConvCompressedGPT architecture offers multiple benefits, primarily due to its use of 1D convolutional layers. Firstly, this approach leads to a reduction in the total number of model parameters, similar to LinearlyCompressedGPT. By progressively decreasing the dimensions, ConvCompressedGPT requires fewer computational resources and uses less memory, enhancing its efficiency. Secondly, the convolutional layers provide inherent advantages by capturing positional patterns and dependencies within the input sequence, potentially improving the model\u2019s ability to make accurate predictions. This characteristic is especially beneficial for handling sequential data, where the positional context plays a crucial role. Lastly, the ConvCompressedGPT structure is highly scalable, allowing for further extension by adding more convolutional layers. This scalability provides flexibility in adapting the architecture to various contexts and tasks, making it suitable for a wide range of applications. 4 Dataset In this study, we utilize the codeparrot-ds dataset from HuggingFace [8], [9], which comprises code snippets that specifically leverage data science libraries such as numpy, pandas, scikit-learn, and matplotlib. The selection of this dataset was strategic; it focuses narrowly on four well-defined libraries. This specificity is advantageous for our research objectives, as employing a more expansive dataset encompassing a broader array of libraries might impede the performance of the models we train, particularly given their smaller scale. For practical and experimental purposes, we tailored the dataset size to suit our computational constraints and to facilitate rapid iteration cycles. Although the original codeparrot-ds dataset contains 384,000 training examples and 3,320 validation examples as detailed on the HuggingFace repository, we randomly selected a subset of 10,000 examples for training and 1,000 for validation. This reduction was necessitated by the limited computing resources available to us. 5 Training and Results 5.1 Training 5.1.1 Tokenization Tokenization is a critical preprocessing step in natural language processing (NLP) and involves breaking down text into smaller elements known as tokens, which are then mapped to unique integer identifiers. There are three main tokenization strategies: word-level, character-level, and sub-word tokenization. Each strategy has its advantages and disadvantages, depending on the specific application and the required granularity of text representation. Sub-word tokenization, commonly used by state-of-the-art large language models (LLMs), combines the benefits of word-level and character-level tokenization. It splits words into smaller meaningful units, allowing the model to generalize across similar words while maintaining a manageable vocabulary size. However, sub-word tokenization can lead to ambiguity, especially for out-of-vocabulary words, and may require more complex tokenization algorithms. In this work, we opt for character-level tokenization. This decision is driven by the architecture design, where the reduced dimensions in two of the three proposed models could hinder the ability of the classification head to predict the next token when using word-level or sub-word tokenization. The increased vocabulary size associated with these tokenization strategies could complicate the learning process and lead to slower convergence. Character-level tokenization offers several advantages for our task. It simplifies the tokenization process, eliminating the need for intricate splitting rules, and it aligns well with our focus on single-line code completion. By employing character-level tokenization, both the training and testing datasets were processed into a vocabulary of 2,117 unique tokens. 5.1.2 Training loop The training of our models is conducted with a batch size of 64. Given that our objective is to develop a model for single-line code completion, we set the context length to a small 256 tokens. This context length helps maintain the required focus on code snippets while avoiding excessive memory consumption. To reduce overfitting, we apply a dropout of 0.1 across the layers. For our loss function, we employ cross-entropy loss, a widely used metric for classification tasks. The model weights are updated using the AdamW optimizer with a learning rate of 3e-4 and a weight decay of 1e-3. The AdamW optimizer is selected for its robustness in training large models and its ability to converge quickly with minimal parameter tuning. 5 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications Figure 6: Loss comparison b/w the 4 models Name Params (M) Size (MB) Training Time (min) gpt 8.82 33.66 25.35 p-gpt 9.74 37.14 26.15 p-gpt (1-block) 6.19 23.60 26.15 lc-gpt 5.65 21.54 20.68 cc-gpt 5.65 21.54 21.68 Table 1: n_layers=4, d_model=384, n_heads=6 Training proceeds for 10,000 steps and to monitor the model\u2019s performance during training, we evaluate it every 500 steps using the test set, and the results are logged for analysis and validation (refer to Figure 6 for an overview of the loss trends). This evaluation frequency allows us to track the model\u2019s progress and adjust the training approach if needed. 5.2 Results The results depicted in Figure 6 indicate that, with the exception of the p-gpt-1 model, all other models exhibit performance metrics similar to those of gpt. This outcome aligns with our earlier hypotheses regarding model architecture and the parallelization strategy used within the decoder blocks. The degradation in the performance of p-gpt-1 can be attributed to its inference method. During inference, the model omits the decoder_1 block if its weight is less than 0.5, or alternatively the decoder_2 block, as illustrated in Equation 3. This intentional reduction in the model\u2019s complexity likely contributes to the observed performance gap. However, the performance of p-gpt-1 could potentially be improved with extended training, as the loss trends suggest a continued decrease even in the later training stages. As detailed in Table 1, several of the tested models achieved performance metrics similar to those of gpt, the traditional decoder-only transformer model, while employing fewer parameters and yielding faster training times. Notably, both lc-gpt and cc-gpt demonstrated a 36% reduction in parameter count compared to gpt. This reduction in parameters has significant implications for larger-scale models with more complex architectures, extended context lengths, and additional decoder blocks. The reduction in parameters also correlates with reduced memory requirements for model deployment. This efficiency gain has far-reaching effects, particularly for larger-scale models, offering potential benefits such as faster inference times and enhanced feasibility for on-device deployment. These advantages could play a crucial role in scenarios where computational resources and memory capacity are limited. 6 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications 6 Conclusion The architectural modifications analyzed in this study not only enhance the efficiency of the transformer models in terms of faster inference and training with fewer parameters, but also underline the need to further explore this area of research. The findings suggest that continued innovation in transformer architectures could significantly expedite the deployment of on-device language models that operate efficiently without specialized hardware. This promises substantial advancements in making AI technologies more accessible and sustainable. Particularly, further investigations into the p-gpt architecture are crucial. A deeper understanding of the knowledge processed by each of the parallel blocks within p-gpt could reveal how to structure the training process to tailor each block for specific tasks more effectively. This targeted training approach could lead to breakthroughs in model specialization and efficiency. Moreover, both lc-gpt and cc-gpt architectures present rich opportunities for further exploration. Experimenting with scaling up the dimensions\u2014from lower to higher within the processing flow\u2014could provide insights into optimizing computational resources and model performance. Additionally, incorporating pooling layers, as introduced in [3], in place of traditional linear or convolutional layers could offer a novel method to reduce complexity and enhance model learning capabilities. These avenues not only hold the promise of refining the efficiency and effectiveness of transformer models but also pave the way for groundbreaking applications in real-world scenarios. References [1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, \u0141ukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998\u20136008, 2017. [2] Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training, 2018. Available: https://cdn.openai.com/research-covers/ language-unsupervised/language_understanding_paper.pdf. [3] Zihang Dai, Guokun Lai, Yiming Yang, and Quoc Le. Funnel-transformer: Filtering out sequential redundancy for efficient language processing. In Advances in Neural Information Processing Systems, 2020. [4] Omkar Thawakar, Ashmal Vayani, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Michael Felsberg, Timothy Baldwin, Eric P. Xing, and Fahad Shahbaz Khan. Mobillama: Towards accurate and lightweight fully transparent gpt, 2024. [5] Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebr\u00f3n, and Sumit Sanghai. Gqa: Training generalized multi-query transformer models from multi-head checkpoints, 2023. [6] Noam Shazeer. Fast transformer decoding: One write-head is all you need. 2019. cite arxiv:1911.02150. [7] Andrey Gromov, Kushal Tirumala, Hassan Shapourian, Paolo Glorioso, and Daniel A. Roberts. The unreasonable ineffectiveness of the deeper layers. arXiv preprint arXiv:2403.17887, 2024. [8] Hugging Face. Codeparrot-ds-train dataset. https://huggingface.co/datasets/huggingface-course/ codeparrot-ds-train, 2021. Accessed: 2024-04-16. [9] Hugging Face. Codeparrot-ds-valid dataset. https://huggingface.co/datasets/huggingface-course/ codeparrot-ds-valid, 2021. Accessed: 2024-04-16. [10] Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio Cesar Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Conti Kauffmann, Gustavo Henrique de Rosa, Olli Saarikivi, Adil Salim, Shital Shah, Harkirat Singh Behl, Xin Wang, Sebastien Bubeck, Ronen Eldan, Adam Tauman Kalai, Yin Tat Lee, and Yuanzhi Li. Textbooks are all you need, 2023. [11] Tianyang Lin, Yuxin Wang, Xiangyang Liu, and Xipeng Qiu. A survey of transformers. arXiv preprint arXiv:2106.04554, 2021. [12] Wenfeng Zheng, Gu Gong, Jiawei Tian, Siyu Lu, Ruiyang Wang, Zhengtong Yin, Xiaolu Li, and Lirong Yin. Design of a modified transformer architecture based on relative position coding. International Journal of Computational Intelligence Systems, 16(1):1\u201317, 2023. [13] Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang. Informer: Beyond efficient transformer for long sequence time-series forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 11106\u201311115. AAAI, 2021. 7 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications [14] Shengding Hu, Yuge Tu, Xu Han, Chaoqun He, Ganqu Cui, Xiang Long, Zhi Zheng, Yewei Fang, Yuxiang Huang, Weilin Zhao, et al. Minicpm: Unveiling the potential of small language models with scalable training strategies. arXiv preprint arXiv:2404.06395, 2024. [15] Li Shen and Yangzhu Wang. Tcct: Tightly-coupled convolutional transformer on time series forecasting. Neurocomputing, 480:131\u2013145, 2022. [16] Andrej Karpathy. nanogpt: Minimal gpt implementation with pytorch. https://github.com/karpathy/ nanoGPT, 2023. [17] Hugging Face. Chapter 7.6: Transformers and transfer learning, 2024. Accessed: 2024-04-19. 8",
+                    "pre_questions": [],
+                    "main_content": "Introduction Since the debut of ChatGPT, there has been a notable increase in research on Large Language Models (LLMs) across a broad range of disciplines, made possible by the accessibility of this technology to a diverse user base. This fastly growing field has largely pursued two distinct paths: one aims at either scaling the model dimensions or the training dataset (or both) to enhance performance, while the other concentrates on refining smaller models (ranging from 1B to 7B parameters) with high-quality data. Despite these advances, investigations into the structural modifications of the transformer architecture itself have been relatively overlooked. Recent studies challenge the necessity of perpetually increasing model sizes by demonstrating that the deeper layers of LLMs may have minimal influence on predictive outcomes. In this work, we explore modifications to the decoder-only transformer architecture to address current challenges in the scalability and practical application of Large Language Models (LLMs). Recognizing the significant impact of model size on the computational overhead of training and inference, we introduce three compact variants\u2014ParallelGPT (p-gpt), LinearlyCompressedGPT (lc-gpt), and ConvCompressedGPT (cc-gpt)\u2014each designed to reduce parameter count while maintaining, or potentially enhancing, model performance. Our decision to focus on smaller-sized models, ranging from 5M to 10M parameters, stems from several considerations. Primarily, these dimensions facilitate faster training and inference times, critical for iterative development cycles and real-time applications. Additionally, smaller models circumvent the limitations often encountered with quantized models, which despite their reduced computational demands, frequently underperform compared to their full-precision counterparts. This approach not only ensures efficient local execution without the need for specialized hardware but also aligns with our goal to achieve comparable performance to the original GPT architecture with significantly reduced computational resources. We have pre-trained each model variant on a specialized dataset tailored for data science code completion, envisioning these models as locally-deployed tools for enhancing productivity through single-line code suggestions. Preliminary results from our experiments suggest a lot of potential research within transformer architecture optimization that could reduce the need for scaling models to prohibitively large sizes hundreds of billions, or even a trillion parameters-currently dominating the field. The remainder of this paper is organized as follows. The GPT architecture GitHub Repository: gpt-variations arXiv:2404.14462v2  [cs.LG]  24 Apr 2024 Towards Smaller, Faster Decoder-Only Transformers: Architectural Variants and Their Implications and its related works are discussed in Section 2. The modifications made to the base architecture and the justifications for the modifications are presented in Section 3. The dataset used is discussed in Section 4. The training pipeline, results and comparisions between the different architectures are presented in Section 5 and finally, the future scope, applications of these architectures and the concluding remarks are presented in Section 6. Figure 1: Decoder Block Figure 2: GPT 2 Literature survey 2.1 Transformer Architecture Decoder-only transformer architectures represent a significant area of research within the field of natural language processing (NLP). Originating from the broader family of transformer models, which are known for their self-attention mechanisms enabling better parallelization and handling of long-range dependencies, the decoder-only configuration has been specifically tailored for generative tasks. The architecture of a decoder-only transformer is characterized by its sequence of stacked decoder layers as shown in 1, each consisting of a masked self-attention mechanism followed by a feed-forward neural network. Unlike the encoder-decoder frameworks where both components process inputs and generate outputs, the decoder-only model focuses solely on output generation. This is achieved by training the model to predict the next token in a sequence given the previous tokens, making it inherently suitable for tasks such as text generation, language modeling, and autoregressive prediction. The masked self-attention mechanism within decoder-only architectures allows each position in the output sequence to attend to all positions up to and including that one, which is instrumental in capturing the nuances of language patterns. This mechanism, coupled with layer normalization and residual connections, helps in stabilizing the learning process and enhancing the model\u2019s ability to generalize from its training data. **[related work]*"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2108.12784v2",
+                    "title": "TCCT: Tightly-Coupled Convolutional Transformer on Time Series Forecasting",
+                    "abstract": "Time series forecasting is essential for a wide range of real-world\napplications. Recent studies have shown the superiority of Transformer in\ndealing with such problems, especially long sequence time series input(LSTI)\nand long sequence time series forecasting(LSTF) problems. To improve the\nefficiency and enhance the locality of Transformer, these studies combine\nTransformer with CNN in varying degrees. However, their combinations are\nloosely-coupled and do not make full use of CNN. To address this issue, we\npropose the concept of tightly-coupled convolutional Transformer(TCCT) and\nthree TCCT architectures which apply transformed CNN architectures into\nTransformer: (1) CSPAttention: through fusing CSPNet with self-attention\nmechanism, the computation cost of self-attention mechanism is reduced by 30%\nand the memory usage is reduced by 50% while achieving equivalent or beyond\nprediction accuracy. (2) Dilated causal convolution: this method is to modify\nthe distilling operation proposed by Informer through replacing canonical\nconvolutional layers with dilated causal convolutional layers to gain\nexponentially receptive field growth. (3) Passthrough mechanism: the\napplication of passthrough mechanism to stack of self-attention blocks helps\nTransformer-like models get more fine-grained information with negligible extra\ncomputation costs. Our experiments on real-world datasets show that our TCCT\narchitectures could greatly improve the performance of existing state-of-art\nTransformer models on time series forecasting with much lower computation and\nmemory costs, including canonical Transformer, LogTrans and Informer.",
+                    "authors": "Li Shen, Yangzhu Wang",
+                    "published": "2021-08-29",
+                    "updated": "2022-02-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.16840v1",
+                    "title": "MobiLlama: Towards Accurate and Lightweight Fully Transparent GPT",
+                    "abstract": "\"Bigger the better\" has been the predominant trend in recent Large Language\nModels (LLMs) development. However, LLMs do not suit well for scenarios that\nrequire on-device processing, energy efficiency, low memory footprint, and\nresponse efficiency. These requisites are crucial for privacy, security, and\nsustainable deployment. This paper explores the \"less is more\" paradigm by\naddressing the challenge of designing accurate yet efficient Small Language\nModels (SLMs) for resource constrained devices. Our primary contribution is the\nintroduction of an accurate and fully transparent open-source 0.5 billion\n(0.5B) parameter SLM, named MobiLlama, catering to the specific needs of\nresource-constrained computing with an emphasis on enhanced performance with\nreduced resource demands. MobiLlama is a SLM design that initiates from a\nlarger model and applies a careful parameter sharing scheme to reduce both the\npre-training and the deployment cost. Our work strives to not only bridge the\ngap in open-source SLMs but also ensures full transparency, where complete\ntraining data pipeline, training code, model weights, and over 300 checkpoints\nalong with evaluation codes is available at :\nhttps://github.com/mbzuai-oryx/MobiLlama.",
+                    "authors": "Omkar Thawakar, Ashmal Vayani, Salman Khan, Hisham Cholakal, Rao M. Anwer, Michael Felsberg, Tim Baldwin, Eric P. Xing, Fahad Shahbaz Khan",
+                    "published": "2024-02-26",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.17887v1",
+                    "title": "The Unreasonable Ineffectiveness of the Deeper Layers",
+                    "abstract": "We empirically study a simple layer-pruning strategy for popular families of\nopen-weight pretrained LLMs, finding minimal degradation of performance on\ndifferent question-answering benchmarks until after a large fraction (up to\nhalf) of the layers are removed. To prune these models, we identify the optimal\nblock of layers to prune by considering similarity across layers; then, to\n\"heal\" the damage, we perform a small amount of finetuning. In particular, we\nuse parameter-efficient finetuning (PEFT) methods, specifically quantization\nand Low Rank Adapters (QLoRA), such that each of our experiments can be\nperformed on a single A100 GPU. From a practical perspective, these results\nsuggest that layer pruning methods can complement other PEFT strategies to\nfurther reduce computational resources of finetuning on the one hand, and can\nimprove the memory and latency of inference on the other hand. From a\nscientific perspective, the robustness of these LLMs to the deletion of layers\nimplies either that current pretraining methods are not properly leveraging the\nparameters in the deeper layers of the network or that the shallow layers play\na critical role in storing knowledge.",
+                    "authors": "Andrey Gromov, Kushal Tirumala, Hassan Shapourian, Paolo Glorioso, Daniel A. Roberts",
+                    "published": "2024-03-26",
+                    "updated": "2024-03-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG",
+                        "stat.ML"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2106.04554v2",
+                    "title": "A Survey of Transformers",
+                    "abstract": "Transformers have achieved great success in many artificial intelligence\nfields, such as natural language processing, computer vision, and audio\nprocessing. Therefore, it is natural to attract lots of interest from academic\nand industry researchers. Up to the present, a great variety of Transformer\nvariants (a.k.a. X-formers) have been proposed, however, a systematic and\ncomprehensive literature review on these Transformer variants is still missing.\nIn this survey, we provide a comprehensive review of various X-formers. We\nfirst briefly introduce the vanilla Transformer and then propose a new taxonomy\nof X-formers. Next, we introduce the various X-formers from three perspectives:\narchitectural modification, pre-training, and applications. Finally, we outline\nsome potential directions for future research.",
+                    "authors": "Tianyang Lin, Yuxin Wang, Xiangyang Liu, Xipeng Qiu",
+                    "published": "2021-06-08",
+                    "updated": "2021-06-15",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06395v2",
+                    "title": "MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies",
+                    "abstract": "The burgeoning interest in developing Large Language Models (LLMs) with up to\ntrillion parameters has been met with concerns regarding resource efficiency\nand practical expense, particularly given the immense cost of experimentation.\nThis scenario underscores the importance of exploring the potential of Small\nLanguage Models (SLMs) as a resource-efficient alternative. In this context, we\nintroduce MiniCPM, specifically the 1.2B and 2.4B non-embedding parameter\nvariants, not only excel in their respective categories but also demonstrate\ncapabilities on par with 7B-13B LLMs. While focusing on SLMs, our approach\nexhibits scalability in both model and data dimensions for future LLM research.\nRegarding model scaling, we employ extensive model wind tunnel experiments for\nstable and optimal scaling. For data scaling, we introduce a\nWarmup-Stable-Decay (WSD) learning rate scheduler (LRS), conducive to\ncontinuous training and domain adaptation. We present an in-depth analysis of\nthe intriguing training dynamics that occurred in the WSD LRS. With WSD LRS, we\nare now able to efficiently study data-model scaling law without extensive\nretraining experiments on both axes of model and data, from which we derive the\nmuch higher compute optimal data-model ratio than Chinchilla Optimal.\nAdditionally, we introduce MiniCPM family, including MiniCPM-DPO, MiniCPM-MoE\nand MiniCPM-128K, whose excellent performance further cementing MiniCPM's\nfoundation in diverse SLM applications. MiniCPM models are available publicly\nat https://github.com/OpenBMB/MiniCPM .",
+                    "authors": "Shengding Hu, Yuge Tu, Xu Han, Chaoqun He, Ganqu Cui, Xiang Long, Zhi Zheng, Yewei Fang, Yuxiang Huang, Weilin Zhao, Xinrong Zhang, Zheng Leng Thai, Kaihuo Zhang, Chongyi Wang, Yuan Yao, Chenyang Zhao, Jie Zhou, Jie Cai, Zhongwu Zhai, Ning Ding, Chao Jia, Guoyang Zeng, Dahai Li, Zhiyuan Liu, Maosong Sun",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13245v3",
+                    "title": "GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints",
+                    "abstract": "Multi-query attention (MQA), which only uses a single key-value head,\ndrastically speeds up decoder inference. However, MQA can lead to quality\ndegradation, and moreover it may not be desirable to train a separate model\njust for faster inference. We (1) propose a recipe for uptraining existing\nmulti-head language model checkpoints into models with MQA using 5% of original\npre-training compute, and (2) introduce grouped-query attention (GQA), a\ngeneralization of multi-query attention which uses an intermediate (more than\none, less than number of query heads) number of key-value heads. We show that\nuptrained GQA achieves quality close to multi-head attention with comparable\nspeed to MQA.",
+                    "authors": "Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebr\u00f3n, Sumit Sanghai",
+                    "published": "2023-05-22",
+                    "updated": "2023-12-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2006.03236v1",
+                    "title": "Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing",
+                    "abstract": "With the success of language pretraining, it is highly desirable to develop\nmore efficient architectures of good scalability that can exploit the abundant\nunlabeled data at a lower cost. To improve the efficiency, we examine the\nmuch-overlooked redundancy in maintaining a full-length token-level\npresentation, especially for tasks that only require a single-vector\npresentation of the sequence. With this intuition, we propose\nFunnel-Transformer which gradually compresses the sequence of hidden states to\na shorter one and hence reduces the computation cost. More importantly, by\nre-investing the saved FLOPs from length reduction in constructing a deeper or\nwider model, we further improve the model capacity. In addition, to perform\ntoken-level predictions as required by common pretraining objectives,\nFunnel-Transformer is able to recover a deep representation for each token from\nthe reduced hidden sequence via a decoder. Empirically, with comparable or\nfewer FLOPs, Funnel-Transformer outperforms the standard Transformer on a wide\nvariety of sequence-level prediction tasks, including text classification,\nlanguage understanding, and reading comprehension. The code and pretrained\ncheckpoints are available at https://github.com/laiguokun/Funnel-Transformer.",
+                    "authors": "Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le",
+                    "published": "2020-06-05",
+                    "updated": "2020-06-05",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CL",
+                        "stat.ML"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1911.02150v1",
+                    "title": "Fast Transformer Decoding: One Write-Head is All You Need",
+                    "abstract": "Multi-head attention layers, as used in the Transformer neural sequence\nmodel, are a powerful alternative to RNNs for moving information across and\nbetween sequences. While training these layers is generally fast and simple,\ndue to parallelizability across the length of the sequence, incremental\ninference (where such paralleization is impossible) is often slow, due to the\nmemory-bandwidth cost of repeatedly loading the large \"keys\" and \"values\"\ntensors. We propose a variant called multi-query attention, where the keys and\nvalues are shared across all of the different attention \"heads\", greatly\nreducing the size of these tensors and hence the memory bandwidth requirements\nof incremental decoding. We verify experimentally that the resulting models can\nindeed be much faster to decode, and incur only minor quality degradation from\nthe baseline.",
+                    "authors": "Noam Shazeer",
+                    "published": "2019-11-06",
+                    "updated": "2019-11-06",
+                    "primary_cat": "cs.NE",
+                    "cats": [
+                        "cs.NE",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2012.07436v3",
+                    "title": "Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting",
+                    "abstract": "Many real-world applications require the prediction of long sequence\ntime-series, such as electricity consumption planning. Long sequence\ntime-series forecasting (LSTF) demands a high prediction capacity of the model,\nwhich is the ability to capture precise long-range dependency coupling between\noutput and input efficiently. Recent studies have shown the potential of\nTransformer to increase the prediction capacity. However, there are several\nsevere issues with Transformer that prevent it from being directly applicable\nto LSTF, including quadratic time complexity, high memory usage, and inherent\nlimitation of the encoder-decoder architecture. To address these issues, we\ndesign an efficient transformer-based model for LSTF, named Informer, with\nthree distinctive characteristics: (i) a $ProbSparse$ self-attention mechanism,\nwhich achieves $O(L \\log L)$ in time complexity and memory usage, and has\ncomparable performance on sequences' dependency alignment. (ii) the\nself-attention distilling highlights dominating attention by halving cascading\nlayer input, and efficiently handles extreme long input sequences. (iii) the\ngenerative style decoder, while conceptually simple, predicts the long\ntime-series sequences at one forward operation rather than a step-by-step way,\nwhich drastically improves the inference speed of long-sequence predictions.\nExtensive experiments on four large-scale datasets demonstrate that Informer\nsignificantly outperforms existing methods and provides a new solution to the\nLSTF problem.",
+                    "authors": "Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, Wancai Zhang",
+                    "published": "2020-12-14",
+                    "updated": "2021-03-28",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1706.03762v7",
+                    "title": "Attention Is All You Need",
+                    "abstract": "The dominant sequence transduction models are based on complex recurrent or\nconvolutional neural networks in an encoder-decoder configuration. The best\nperforming models also connect the encoder and decoder through an attention\nmechanism. We propose a new simple network architecture, the Transformer, based\nsolely on attention mechanisms, dispensing with recurrence and convolutions\nentirely. Experiments on two machine translation tasks show these models to be\nsuperior in quality while being more parallelizable and requiring significantly\nless time to train. Our model achieves 28.4 BLEU on the WMT 2014\nEnglish-to-German translation task, improving over the existing best results,\nincluding ensembles by over 2 BLEU. On the WMT 2014 English-to-French\ntranslation task, our model establishes a new single-model state-of-the-art\nBLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction\nof the training costs of the best models from the literature. We show that the\nTransformer generalizes well to other tasks by applying it successfully to\nEnglish constituency parsing both with large and limited training data.",
+                    "authors": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin",
+                    "published": "2017-06-12",
+                    "updated": "2023-08-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.12736v1",
+                    "title": "Large Language Model Supply Chain: A Research Agenda",
+                    "abstract": "The rapid advancements in pre-trained Large Language Models (LLMs) and Large\nMultimodal Models (LMMs) have ushered in a new era of intelligent applications,\ntransforming fields ranging from natural language processing to content\ngeneration. The LLM supply chain represents a crucial aspect of the\ncontemporary artificial intelligence landscape. It encompasses the entire\nlifecycle of pre-trained models, from its initial development and training to\nits final deployment and application in various domains. This paper presents a\ncomprehensive overview of the LLM supply chain, highlighting its three core\nelements: 1) the model infrastructure, encompassing datasets and toolchain for\ntraining, optimization, and deployment; 2) the model lifecycle, covering\ntraining, testing, releasing, and ongoing maintenance; and 3) the downstream\napplication ecosystem, enabling the integration of pre-trained models into a\nwide range of intelligent applications. However, this rapidly evolving field\nfaces numerous challenges across these key components, including data privacy\nand security, model interpretability and fairness, infrastructure scalability,\nand regulatory compliance. Addressing these challenges is essential for\nharnessing the full potential of LLMs and ensuring their ethical and\nresponsible use. This paper provides a future research agenda for the LLM\nsupply chain, aiming at driving the continued advancement and responsible\ndeployment of these transformative LLMs.",
+                    "authors": "Shenao Wang, Yanjie Zhao, Xinyi Hou, Haoyu Wang",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.03192v1",
+                    "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+                    "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+                    "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+                    "published": "2024-04-04",
+                    "updated": "2024-04-04",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.09397v1",
+                    "title": "Do Large GPT Models Discover Moral Dimensions in Language Representations? A Topological Study Of Sentence Embeddings",
+                    "abstract": "As Large Language Models are deployed within Artificial Intelligence systems,\nthat are increasingly integrated with human society, it becomes more important\nthan ever to study their internal structures. Higher level abilities of LLMs\nsuch as GPT-3.5 emerge in large part due to informative language\nrepresentations they induce from raw text data during pre-training on trillions\nof words. These embeddings exist in vector spaces of several thousand\ndimensions, and their processing involves mapping between multiple vector\nspaces, with total number of parameters on the order of trillions. Furthermore,\nthese language representations are induced by gradient optimization, resulting\nin a black box system that is hard to interpret. In this paper, we take a look\nat the topological structure of neuronal activity in the \"brain\" of Chat-GPT's\nfoundation language model, and analyze it with respect to a metric representing\nthe notion of fairness. We develop a novel approach to visualize GPT's moral\ndimensions. We first compute a fairness metric, inspired by social psychology\nliterature, to identify factors that typically influence fairness assessments\nin humans, such as legitimacy, need, and responsibility. Subsequently, we\nsummarize the manifold's shape using a lower-dimensional simplicial complex,\nwhose topology is derived from this metric. We color it with a heat map\nassociated with this fairness metric, producing human-readable visualizations\nof the high-dimensional sentence manifold. Our results show that sentence\nembeddings based on GPT-3.5 can be decomposed into two submanifolds\ncorresponding to fair and unfair moral judgments. This indicates that GPT-based\nlanguage models develop a moral dimension within their representation spaces\nand induce an understanding of fairness during their training process.",
+                    "authors": "Stephen Fitz",
+                    "published": "2023-09-17",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.11653v2",
+                    "title": "\"It's a Fair Game\", or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents",
+                    "abstract": "The widespread use of Large Language Model (LLM)-based conversational agents\n(CAs), especially in high-stakes domains, raises many privacy concerns.\nBuilding ethical LLM-based CAs that respect user privacy requires an in-depth\nunderstanding of the privacy risks that concern users the most. However,\nexisting research, primarily model-centered, does not provide insight into\nusers' perspectives. To bridge this gap, we analyzed sensitive disclosures in\nreal-world ChatGPT conversations and conducted semi-structured interviews with\n19 LLM-based CA users. We found that users are constantly faced with trade-offs\nbetween privacy, utility, and convenience when using LLM-based CAs. However,\nusers' erroneous mental models and the dark patterns in system design limited\ntheir awareness and comprehension of the privacy risks. Additionally, the\nhuman-like interactions encouraged more sensitive disclosures, which\ncomplicated users' ability to navigate the trade-offs. We discuss practical\ndesign guidelines and the needs for paradigm shifts to protect the privacy of\nLLM-based CA users.",
+                    "authors": "Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, Tianshi Li",
+                    "published": "2023-09-20",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC",
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.08780v1",
+                    "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+                    "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+                    "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+                    "published": "2023-10-13",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.13925v1",
+                    "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+                    "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+                    "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10149v2",
+                    "title": "A Survey on Fairness in Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have shown powerful performance and development\nprospects and are widely deployed in the real world. However, LLMs can capture\nsocial biases from unprocessed training data and propagate the biases to\ndownstream tasks. Unfair LLM systems have undesirable social impacts and\npotential harms. In this paper, we provide a comprehensive review of related\nresearch on fairness in LLMs. Considering the influence of parameter magnitude\nand training paradigm on research strategy, we divide existing fairness\nresearch into oriented to medium-sized LLMs under pre-training and fine-tuning\nparadigms and oriented to large-sized LLMs under prompting paradigms. First,\nfor medium-sized LLMs, we introduce evaluation metrics and debiasing methods\nfrom the perspectives of intrinsic bias and extrinsic bias, respectively. Then,\nfor large-sized LLMs, we introduce recent fairness research, including fairness\nevaluation, reasons for bias, and debiasing methods. Finally, we discuss and\nprovide insight on the challenges and future directions for the development of\nfairness in LLMs.",
+                    "authors": "Yingji Li, Mengnan Du, Rui Song, Xin Wang, Ying Wang",
+                    "published": "2023-08-20",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.16343v2",
+                    "title": "Evaluating, Understanding, and Improving Constrained Text Generation for Large Language Models",
+                    "abstract": "Advancements in natural language generation (NLG) and large language models\n(LLMs) have led to proficient text generation in various tasks. However,\nintegrating intricate constraints into neural text generation, due to LLMs'\nopacity, remains challenging. This study investigates constrained text\ngeneration for LLMs, where predefined constraints are applied during LLM's\ngeneration process. Our research mainly focuses on mainstream open-source LLMs,\ncategorizing constraints into lexical, structural, and relation-based types. We\nalso present various benchmarks to facilitate fair evaluation. The study\naddresses some key research questions, including evaluating, understanding and\nimproving constrained text generation for LLMs. Results illuminate LLMs'\ncapacity and deficiency to incorporate constraints and provide insights for\nfuture developments in constrained text generation. Codes and datasets will be\nreleased upon acceptance.",
+                    "authors": "Xiang Chen, Xiaojun Wan",
+                    "published": "2023-10-25",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.18502v1",
+                    "title": "Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification",
+                    "abstract": "Employing Large Language Models (LLM) in various downstream applications such\nas classification is crucial, especially for smaller companies lacking the\nexpertise and resources required for fine-tuning a model. Fairness in LLMs\nhelps ensure inclusivity, equal representation based on factors such as race,\ngender and promotes responsible AI deployment. As the use of LLMs has become\nincreasingly prevalent, it is essential to assess whether LLMs can generate\nfair outcomes when subjected to considerations of fairness. In this study, we\nintroduce a framework outlining fairness regulations aligned with various\nfairness definitions, with each definition being modulated by varying degrees\nof abstraction. We explore the configuration for in-context learning and the\nprocedure for selecting in-context demonstrations using RAG, while\nincorporating fairness rules into the process. Experiments conducted with\ndifferent LLMs indicate that GPT-4 delivers superior results in terms of both\naccuracy and fairness compared to other models. This work is one of the early\nattempts to achieve fairness in prediction tasks by utilizing LLMs through\nin-context learning.",
+                    "authors": "Garima Chhikara, Anurag Sharma, Kripabandhu Ghosh, Abhijnan Chakraborty",
+                    "published": "2024-02-28",
+                    "updated": "2024-02-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.12090v1",
+                    "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+                    "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+                    "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+                    "published": "2023-05-20",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.01349v1",
+                    "title": "Fairness in Large Language Models: A Taxonomic Survey",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+                    "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+                    "published": "2024-03-31",
+                    "updated": "2024-03-31",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15198v2",
+                    "title": "Do LLM Agents Exhibit Social Behavior?",
+                    "abstract": "The advances of Large Language Models (LLMs) are expanding their utility in\nboth academic research and practical applications. Recent social science\nresearch has explored the use of these ``black-box'' LLM agents for simulating\ncomplex social systems and potentially substituting human subjects in\nexperiments. Our study delves into this emerging domain, investigating the\nextent to which LLMs exhibit key social interaction principles, such as social\nlearning, social preference, and cooperative behavior (indirect reciprocity),\nin their interactions with humans and other agents. We develop a framework for\nour study, wherein classical laboratory experiments involving human subjects\nare adapted to use LLM agents. This approach involves step-by-step reasoning\nthat mirrors human cognitive processes and zero-shot learning to assess the\ninnate preferences of LLMs. Our analysis of LLM agents' behavior includes both\nthe primary effects and an in-depth examination of the underlying mechanisms.\nFocusing on GPT-4, our analyses suggest that LLM agents appear to exhibit a\nrange of human-like social behaviors such as distributional and reciprocity\npreferences, responsiveness to group identity cues, engagement in indirect\nreciprocity, and social learning capabilities. However, our analysis also\nreveals notable differences: LLMs demonstrate a pronounced fairness preference,\nweaker positive reciprocity, and a more calculating approach in social learning\ncompared to humans. These insights indicate that while LLMs hold great promise\nfor applications in social science research, such as in laboratory experiments\nand agent-based modeling, the subtle behavioral differences between LLM agents\nand humans warrant further investigation. Careful examination and development\nof protocols in evaluating the social behaviors of LLMs are necessary before\ndirectly applying these models to emulate human behavior.",
+                    "authors": "Yan Leng, Yuan Yuan",
+                    "published": "2023-12-23",
+                    "updated": "2024-02-22",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.SI",
+                        "econ.GN",
+                        "q-fin.EC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13862v2",
+                    "title": "A Trip Towards Fairness: Bias and De-Biasing in Large Language Models",
+                    "abstract": "Cheap-to-Build Very Large-Language Models (CtB-LLMs) with affordable training\nare emerging as the next big revolution in natural language processing and\nunderstanding. These CtB-LLMs are democratizing access to trainable Very\nLarge-Language Models (VLLMs) and, thus, may represent the building blocks of\nmany NLP systems solving downstream tasks. Hence, a little or a large bias in\nCtB-LLMs may cause huge harm. In this paper, we performed a large investigation\nof the bias of three families of CtB-LLMs, and we showed that debiasing\ntechniques are effective and usable. Indeed, according to current tests, the\nLLaMA and the OPT families have an important bias in gender, race, religion,\nand profession. In contrast to the analysis for other LLMs, we discovered that\nbias depends not on the number of parameters but on the perplexity. Finally,\nthe debiasing of OPT using LoRA reduces bias up to 4.12 points in the\nnormalized stereotype score.",
+                    "authors": "Leonardo Ranaldi, Elena Sofia Ruzzetti, Davide Venditti, Dario Onorati, Fabio Massimo Zanzotto",
+                    "published": "2023-05-23",
+                    "updated": "2023-08-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.19465v1",
+                    "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+                    "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+                    "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+                    "published": "2024-02-29",
+                    "updated": "2024-02-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.04489v1",
+                    "title": "De-amplifying Bias from Differential Privacy in Language Model Fine-tuning",
+                    "abstract": "Fairness and privacy are two important values machine learning (ML)\npractitioners often seek to operationalize in models. Fairness aims to reduce\nmodel bias for social/demographic sub-groups. Privacy via differential privacy\n(DP) mechanisms, on the other hand, limits the impact of any individual's\ntraining data on the resulting model. The trade-offs between privacy and\nfairness goals of trustworthy ML pose a challenge to those wishing to address\nboth. We show that DP amplifies gender, racial, and religious bias when\nfine-tuning large language models (LLMs), producing models more biased than\nones fine-tuned without DP. We find the cause of the amplification to be a\ndisparity in convergence of gradients across sub-groups. Through the case of\nbinary gender bias, we demonstrate that Counterfactual Data Augmentation (CDA),\na known method for addressing bias, also mitigates bias amplification by DP. As\na consequence, DP and CDA together can be used to fine-tune models while\nmaintaining both fairness and privacy.",
+                    "authors": "Sanjari Srivastava, Piotr Mardziel, Zhikhun Zhang, Archana Ahlawat, Anupam Datta, John C Mitchell",
+                    "published": "2024-02-07",
+                    "updated": "2024-02-07",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CR",
+                        "cs.CY",
+                        "stat.ME"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14769v3",
+                    "title": "Large Language Model (LLM) Bias Index -- LLMBI",
+                    "abstract": "The Large Language Model Bias Index (LLMBI) is a pioneering approach designed\nto quantify and address biases inherent in large language models (LLMs), such\nas GPT-4. We recognise the increasing prevalence and impact of LLMs across\ndiverse sectors. This research introduces a novel metric, LLMBI, to\nsystematically measure and mitigate biases potentially skewing model responses.\nWe formulated LLMBI using a composite scoring system incorporating multiple\ndimensions of bias, including but not limited to age, gender, and racial\nbiases. To operationalise this metric, we engaged in a multi-step process\ninvolving collecting and annotating LLM responses, applying sophisticated\nNatural Language Processing (NLP) techniques for bias detection, and computing\nthe LLMBI score through a specially crafted mathematical formula. The formula\nintegrates weighted averages of various bias dimensions, a penalty for dataset\ndiversity deficiencies, and a correction for sentiment biases. Our empirical\nanalysis, conducted using responses from OpenAI's API, employs advanced\nsentiment analysis as a representative method for bias detection. The research\nreveals LLMs, whilst demonstrating impressive capabilities in text generation,\nexhibit varying degrees of bias across different dimensions. LLMBI provides a\nquantifiable measure to compare biases across models and over time, offering a\nvital tool for systems engineers, researchers and regulators in enhancing the\nfairness and reliability of LLMs. It highlights the potential of LLMs in\nmimicking unbiased human-like responses. Additionally, it underscores the\nnecessity of continuously monitoring and recalibrating such models to align\nwith evolving societal norms and ethical standards.",
+                    "authors": "Abiodun Finbarrs Oketunji, Muhammad Anas, Deepthi Saina",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15398v1",
+                    "title": "Fairness-Aware Structured Pruning in Transformers",
+                    "abstract": "The increasing size of large language models (LLMs) has introduced challenges\nin their training and inference. Removing model components is perceived as a\nsolution to tackle the large model sizes, however, existing pruning methods\nsolely focus on performance, without considering an essential aspect for the\nresponsible use of LLMs: model fairness. It is crucial to address the fairness\nof LLMs towards diverse groups, such as women, Black people, LGBTQ+, Jewish\ncommunities, among others, as they are being deployed and available to a wide\naudience. In this work, first, we investigate how attention heads impact\nfairness and performance in pre-trained transformer-based language models. We\nthen propose a novel method to prune the attention heads that negatively impact\nfairness while retaining the heads critical for performance, i.e. language\nmodeling capabilities. Our approach is practical in terms of time and\nresources, as it does not require fine-tuning the final pruned, and fairer,\nmodel. Our findings demonstrate a reduction in gender bias by 19%, 19.5%,\n39.5%, 34.7%, 23%, and 8% for DistilGPT-2, GPT-2, GPT-Neo of two different\nsizes, GPT-J, and Llama 2 models, respectively, in comparison to the biased\nmodel, with only a slight decrease in performance.",
+                    "authors": "Abdelrahman Zayed, Goncalo Mordido, Samira Shabanian, Ioana Baldini, Sarath Chandar",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15478v1",
+                    "title": "A Group Fairness Lens for Large Language Models",
+                    "abstract": "The rapid advancement of large language models has revolutionized various\napplications but also raised crucial concerns about their potential to\nperpetuate biases and unfairness when deployed in social media contexts.\nEvaluating LLMs' potential biases and fairness has become crucial, as existing\nmethods rely on limited prompts focusing on just a few groups, lacking a\ncomprehensive categorical perspective. In this paper, we propose evaluating LLM\nbiases from a group fairness lens using a novel hierarchical schema\ncharacterizing diverse social groups. Specifically, we construct a dataset,\nGFair, encapsulating target-attribute combinations across multiple dimensions.\nIn addition, we introduce statement organization, a new open-ended text\ngeneration task, to uncover complex biases in LLMs. Extensive evaluations of\npopular LLMs reveal inherent safety concerns. To mitigate the biases of LLM\nfrom a group fairness perspective, we pioneer a novel chain-of-thought method\nGF-Think to mitigate biases of LLMs from a group fairness perspective.\nExperimental results demonstrate its efficacy in mitigating bias in LLMs to\nachieve fairness.",
+                    "authors": "Guanqun Bi, Lei Shen, Yuqiang Xie, Yanan Cao, Tiangang Zhu, Xiaodong He",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10397v2",
+                    "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+                    "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+                    "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+                    "published": "2023-08-21",
+                    "updated": "2023-10-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.15007v1",
+                    "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+                    "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+                    "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+                    "published": "2023-10-23",
+                    "updated": "2023-10-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.11761v1",
+                    "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+                    "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2023-07-14",
+                    "updated": "2023-07-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.03033v1",
+                    "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+                    "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+                    "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+                    "published": "2023-11-06",
+                    "updated": "2023-11-06",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.07981v1",
+                    "title": "Manipulating Large Language Models to Increase Product Visibility",
+                    "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+                    "authors": "Aounon Kumar, Himabindu Lakkaraju",
+                    "published": "2024-04-11",
+                    "updated": "2024-04-11",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00306v1",
+                    "title": "Probing Explicit and Implicit Gender Bias through LLM Conditional Text Generation",
+                    "abstract": "Large Language Models (LLMs) can generate biased and toxic responses. Yet\nmost prior work on LLM gender bias evaluation requires predefined\ngender-related phrases or gender stereotypes, which are challenging to be\ncomprehensively collected and are limited to explicit bias evaluation. In\naddition, we believe that instances devoid of gender-related language or\nexplicit stereotypes in inputs can still induce gender bias in LLMs. Thus, in\nthis work, we propose a conditional text generation mechanism without the need\nfor predefined gender phrases and stereotypes. This approach employs three\ntypes of inputs generated through three distinct strategies to probe LLMs,\naiming to show evidence of explicit and implicit gender biases in LLMs. We also\nutilize explicit and implicit evaluation metrics to evaluate gender bias in\nLLMs under different strategies. Our experiments demonstrate that an increased\nmodel size does not consistently lead to enhanced fairness and all tested LLMs\nexhibit explicit and/or implicit gender bias, even when explicit gender\nstereotypes are absent in the inputs.",
+                    "authors": "Xiangjue Dong, Yibo Wang, Philip S. Yu, James Caverlee",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.13095v1",
+                    "title": "Enhancing Logical Reasoning in Large Language Models to Facilitate Legal Applications",
+                    "abstract": "Language serves as a vehicle for conveying thought, enabling communication\namong individuals. The ability to distinguish between diverse concepts,\nidentify fairness and injustice, and comprehend a range of legal notions\nfundamentally relies on logical reasoning. Large Language Models (LLMs) attempt\nto emulate human language understanding and generation, but their competency in\nlogical reasoning remains limited. This paper seeks to address the\nphilosophical question: How can we effectively teach logical reasoning to LLMs\nwhile maintaining a deep understanding of the intricate relationship between\nlanguage and logic? By focusing on bolstering LLMs' capabilities in logical\nreasoning, we aim to expand their applicability in law and other\nlogic-intensive disciplines. To this end, we propose a Reinforcement Learning\nfrom Logical Feedback (RLLF) approach, which serves as a potential framework\nfor refining LLMs' reasoning capacities. Through RLLF and a revised evaluation\nmethodology, we explore new avenues for research in this domain and contribute\nto the development of LLMs capable of handling complex legal reasoning tasks\nwhile acknowledging the fundamental connection between language and logic.",
+                    "authors": "Ha-Thanh Nguyen, Wachara Fungwacharakorn, Ken Satoh",
+                    "published": "2023-11-22",
+                    "updated": "2023-11-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.10567v3",
+                    "title": "InSaAF: Incorporating Safety through Accuracy and Fairness | Are LLMs ready for the Indian Legal Domain?",
+                    "abstract": "Recent advancements in language technology and Artificial Intelligence have\nresulted in numerous Language Models being proposed to perform various tasks in\nthe legal domain ranging from predicting judgments to generating summaries.\nDespite their immense potential, these models have been proven to learn and\nexhibit societal biases and make unfair predictions. In this study, we explore\nthe ability of Large Language Models (LLMs) to perform legal tasks in the\nIndian landscape when social factors are involved. We present a novel metric,\n$\\beta$-weighted $\\textit{Legal Safety Score ($LSS_{\\beta}$)}$, which\nencapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs'\nsafety by considering its performance in the $\\textit{Binary Statutory\nReasoning}$ task and its fairness exhibition with respect to various axes of\ndisparities in the Indian society. Task performance and fairness scores of\nLLaMA and LLaMA--2 models indicate that the proposed $LSS_{\\beta}$ metric can\neffectively determine the readiness of a model for safe usage in the legal\nsector. We also propose finetuning pipelines, utilising specialised legal\ndatasets, as a potential method to mitigate bias and improve model safety. The\nfinetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\\beta}$,\nimproving their usability in the Indian legal domain. Our code is publicly\nreleased.",
+                    "authors": "Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru",
+                    "published": "2024-02-16",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00588v1",
+                    "title": "Fairness in Serving Large Language Models",
+                    "abstract": "High-demand LLM inference services (e.g., ChatGPT and BARD) support a wide\nrange of requests from short chat conversations to long document reading. To\nensure that all client requests are processed fairly, most major LLM inference\nservices have request rate limits, to ensure that no client can dominate the\nrequest queue. However, this rudimentary notion of fairness also results in\nunder-utilization of the resources and poor client experience when there is\nspare capacity. While there is a rich literature on fair scheduling, serving\nLLMs presents new challenges due to their unpredictable request lengths and\ntheir unique batching characteristics on parallel accelerators. This paper\nintroduces the definition of LLM serving fairness based on a cost function that\naccounts for the number of input and output tokens processed. To achieve\nfairness in serving, we propose a novel scheduling algorithm, the Virtual Token\nCounter (VTC), a fair scheduler based on the continuous batching mechanism. We\nprove a 2x tight upper bound on the service difference between two backlogged\nclients, adhering to the requirement of work-conserving. Through extensive\nexperiments, we demonstrate the superior performance of VTC in ensuring\nfairness, especially in contrast to other baseline methods, which exhibit\nshortcomings under various conditions.",
+                    "authors": "Ying Sheng, Shiyi Cao, Dacheng Li, Banghua Zhu, Zhuohan Li, Danyang Zhuo, Joseph E. Gonzalez, Ion Stoica",
+                    "published": "2023-12-31",
+                    "updated": "2023-12-31",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.PF"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.03514v3",
+                    "title": "Can Large Language Models Transform Computational Social Science?",
+                    "abstract": "Large Language Models (LLMs) are capable of successfully performing many\nlanguage processing tasks zero-shot (without training data). If zero-shot LLMs\ncan also reliably classify and explain social phenomena like persuasiveness and\npolitical ideology, then LLMs could augment the Computational Social Science\n(CSS) pipeline in important ways. This work provides a road map for using LLMs\nas CSS tools. Towards this end, we contribute a set of prompting best practices\nand an extensive evaluation pipeline to measure the zero-shot performance of 13\nlanguage models on 25 representative English CSS benchmarks. On taxonomic\nlabeling tasks (classification), LLMs fail to outperform the best fine-tuned\nmodels but still achieve fair levels of agreement with humans. On free-form\ncoding tasks (generation), LLMs produce explanations that often exceed the\nquality of crowdworkers' gold references. We conclude that the performance of\ntoday's LLMs can augment the CSS research pipeline in two ways: (1) serving as\nzero-shot data annotators on human annotation teams, and (2) bootstrapping\nchallenging creative generation tasks (e.g., explaining the underlying\nattributes of a text). In summary, LLMs are posed to meaningfully participate\nin social science analysis in partnership with humans.",
+                    "authors": "Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, Diyi Yang",
+                    "published": "2023-04-12",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00625v2",
+                    "title": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models",
+                    "abstract": "The burgeoning field of Large Language Models (LLMs), exemplified by\nsophisticated models like OpenAI's ChatGPT, represents a significant\nadvancement in artificial intelligence. These models, however, bring forth\nsubstantial challenges in the high consumption of computational, memory,\nenergy, and financial resources, especially in environments with limited\nresource capabilities. This survey aims to systematically address these\nchallenges by reviewing a broad spectrum of techniques designed to enhance the\nresource efficiency of LLMs. We categorize methods based on their optimization\nfocus: computational, memory, energy, financial, and network resources and\ntheir applicability across various stages of an LLM's lifecycle, including\narchitecture design, pretraining, finetuning, and system design. Additionally,\nthe survey introduces a nuanced categorization of resource efficiency\ntechniques by their specific resource types, which uncovers the intricate\nrelationships and mappings between various resources and corresponding\noptimization techniques. A standardized set of evaluation metrics and datasets\nis also presented to facilitate consistent and fair comparisons across\ndifferent models and techniques. By offering a comprehensive overview of the\ncurrent sota and identifying open research avenues, this survey serves as a\nfoundational reference for researchers and practitioners, aiding them in\ndeveloping more sustainable and efficient LLMs in a rapidly evolving landscape.",
+                    "authors": "Guangji Bai, Zheng Chai, Chen Ling, Shiyu Wang, Jiaying Lu, Nan Zhang, Tingwei Shi, Ziyang Yu, Mengdan Zhu, Yifei Zhang, Carl Yang, Yue Cheng, Liang Zhao",
+                    "published": "2024-01-01",
+                    "updated": "2024-01-04",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2304.03728v1",
+                    "title": "Interpretable Unified Language Checking",
+                    "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+                    "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+                    "published": "2023-04-07",
+                    "updated": "2023-04-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.06056v1",
+                    "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+                    "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+                    "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+                    "published": "2023-12-11",
+                    "updated": "2023-12-11",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.09219v5",
+                    "title": "\"Kelly is a Warm Person, Joseph is a Role Model\": Gender Biases in LLM-Generated Reference Letters",
+                    "abstract": "Large Language Models (LLMs) have recently emerged as an effective tool to\nassist individuals in writing various types of content, including professional\ndocuments such as recommendation letters. Though bringing convenience, this\napplication also introduces unprecedented fairness concerns. Model-generated\nreference letters might be directly used by users in professional scenarios. If\nunderlying biases exist in these model-constructed letters, using them without\nscrutinization could lead to direct societal harms, such as sabotaging\napplication success rates for female applicants. In light of this pressing\nissue, it is imminent and necessary to comprehensively study fairness issues\nand associated harms in this real-world use case. In this paper, we critically\nexamine gender biases in LLM-generated reference letters. Drawing inspiration\nfrom social science findings, we design evaluation methods to manifest biases\nthrough 2 dimensions: (1) biases in language style and (2) biases in lexical\ncontent. We further investigate the extent of bias propagation by analyzing the\nhallucination bias of models, a term that we define to be bias exacerbation in\nmodel-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs-\nChatGPT and Alpaca, we reveal significant gender biases in LLM-generated\nrecommendation letters. Our findings not only warn against using LLMs for this\napplication without scrutinization, but also illuminate the importance of\nthoroughly studying hidden biases and harms in LLM-generated professional\ndocuments.",
+                    "authors": "Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng",
+                    "published": "2023-10-13",
+                    "updated": "2023-12-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04892v2",
+                    "title": "Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs",
+                    "abstract": "Recent works have showcased the ability of LLMs to embody diverse personas in\ntheir responses, exemplified by prompts like 'You are Yoda. Explain the Theory\nof Relativity.' While this ability allows personalization of LLMs and enables\nhuman behavior simulation, its effect on LLMs' capabilities remains unclear. To\nfill this gap, we present the first extensive study of the unintended\nside-effects of persona assignment on the ability of LLMs to perform basic\nreasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse\npersonas (e.g. an Asian person) spanning 5 socio-demographic groups. Our\nexperiments unveil that LLMs harbor deep rooted bias against various\nsocio-demographics underneath a veneer of fairness. While they overtly reject\nstereotypes when explicitly asked ('Are Black people less skilled at\nmathematics?'), they manifest stereotypical and erroneous presumptions when\nasked to answer questions while adopting a persona. These can be observed as\nabstentions in responses, e.g., 'As a Black person, I can't answer this\nquestion as it requires math knowledge', and generally result in a substantial\nperformance drop. Our experiments with ChatGPT-3.5 show that this bias is\nubiquitous - 80% of our personas demonstrate bias; it is significant - some\ndatasets show performance drops of 70%+; and can be especially harmful for\ncertain groups - some personas suffer statistically significant drops on 80%+\nof the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with\nGPT-4-Turbo showing the least but still a problematic amount of bias (evident\nin 42% of the personas). Further analysis shows that these persona-induced\nerrors can be hard-to-discern and hard-to-avoid. Our findings serve as a\ncautionary tale that the practice of assigning personas to LLMs - a trend on\nthe rise - can surface their deep-rooted biases and have unforeseeable and\ndetrimental side-effects.",
+                    "authors": "Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot",
+                    "published": "2023-11-08",
+                    "updated": "2024-01-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.17553v1",
+                    "title": "RuBia: A Russian Language Bias Detection Dataset",
+                    "abstract": "Warning: this work contains upsetting or disturbing content.\n  Large language models (LLMs) tend to learn the social and cultural biases\npresent in the raw pre-training data. To test if an LLM's behavior is fair,\nfunctional datasets are employed, and due to their purpose, these datasets are\nhighly language and culture-specific. In this paper, we address a gap in the\nscope of multilingual bias evaluation by presenting a bias detection dataset\nspecifically designed for the Russian language, dubbed as RuBia. The RuBia\ndataset is divided into 4 domains: gender, nationality, socio-economic status,\nand diverse, each of the domains is further divided into multiple fine-grained\nsubdomains. Every example in the dataset consists of two sentences with the\nfirst reinforcing a potentially harmful stereotype or trope and the second\ncontradicting it. These sentence pairs were first written by volunteers and\nthen validated by native-speaking crowdsourcing workers. Overall, there are\nnearly 2,000 unique sentence pairs spread over 19 subdomains in RuBia. To\nillustrate the dataset's purpose, we conduct a diagnostic evaluation of\nstate-of-the-art or near-state-of-the-art LLMs and discuss the LLMs'\npredisposition to social biases.",
+                    "authors": "Veronika Grigoreva, Anastasiia Ivanova, Ilseyar Alimova, Ekaterina Artemova",
+                    "published": "2024-03-26",
+                    "updated": "2024-03-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07609v3",
+                    "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+                    "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+                    "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+                    "published": "2023-05-12",
+                    "updated": "2023-10-17",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18140v1",
+                    "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+                    "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+                    "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+                    "published": "2023-11-29",
+                    "updated": "2023-11-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.02219v1",
+                    "title": "FairEvalLLM. A Comprehensive Framework for Benchmarking Fairness in Large Language Model Recommender Systems",
+                    "abstract": "This paper presents a framework for evaluating fairness in recommender\nsystems powered by Large Language Models (RecLLMs), addressing the need for a\nunified approach that spans various fairness dimensions including sensitivity\nto user attributes, intrinsic fairness, and discussions of fairness based on\nunderlying benefits. In addition, our framework introduces counterfactual\nevaluations and integrates diverse user group considerations to enhance the\ndiscourse on fairness evaluation for RecLLMs.\n  Our key contributions include the development of a robust framework for\nfairness evaluation in LLM-based recommendations and a structured method to\ncreate \\textit{informative user profiles} from demographic data, historical\nuser preferences, and recent interactions. We argue that the latter is\nessential for enhancing personalization in such systems, especially in\ntemporal-driven scenarios. We demonstrate the utility of our framework through\npractical applications on two datasets, LastFM-1K and ML-1M. We conduct\nexperiments on a subsample of 80 users from each dataset, testing and assessing\nthe effectiveness of various prompt construction scenarios and in-context\nlearning, comprising more than 50 scenarios. This results in more than 4000\nrecommendations (80 * 50 = 4000). Our study reveals that while there are no\nsignificant unfairness issues in scenarios involving sensitive attributes, some\nconcerns remain. However, in terms of intrinsic fairness, which does not\ninvolve direct sensitivity, unfairness across demographic groups remains\nsignificant. The code and data used for this paper are available at:\n\\url{https://shorturl.at/awBFM}.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2024-05-03",
+                    "updated": "2024-05-03",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.01937v1",
+                    "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+                    "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+                    "authors": "Cheng-Han Chiang, Hung-yi Lee",
+                    "published": "2023-05-03",
+                    "updated": "2023-05-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.18569v1",
+                    "title": "Fairness of ChatGPT",
+                    "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+                    "authors": "Yunqi Li, Yongfeng Zhang",
+                    "published": "2023-05-22",
+                    "updated": "2023-05-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.03852v2",
+                    "title": "FLM-101B: An Open LLM and How to Train It with $100K Budget",
+                    "abstract": "Large language models (LLMs) have achieved remarkable success in NLP and\nmultimodal tasks, among others. Despite these successes, two main challenges\nremain in developing LLMs: (i) high computational cost, and (ii) fair and\nobjective evaluations. In this paper, we report a solution to significantly\nreduce LLM training cost through a growth strategy. We demonstrate that a\n101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US\ndollars. Inspired by IQ tests, we also consolidate an additional range of\nevaluations on top of existing evaluations that focus on knowledge-oriented\nabilities. These IQ evaluations include symbolic mapping, rule understanding,\npattern mining, and anti-interference. Such evaluations minimize the potential\nimpact of memorization. Experimental results show that our model, named\nFLM-101B, trained with a budget of 100K US dollars, achieves performance\ncomparable to powerful and well-known models, e.g., GPT-3 and GLM-130B,\nespecially on the additional range of IQ evaluations. The checkpoint of\nFLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.",
+                    "authors": "Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Xuying Meng, Siqi Fan, Peng Han, Jing Li, Li Du, Bowen Qin, Zheng Zhang, Aixin Sun, Yequan Wang",
+                    "published": "2023-09-07",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.02839v1",
+                    "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+                    "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+                    "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+                    "published": "2024-03-05",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00884v2",
+                    "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+                    "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+                    "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+                    "published": "2024-03-01",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.DB",
+                    "cats": [
+                        "cs.DB",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.06500v1",
+                    "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+                    "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+                    "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+                    "published": "2023-10-10",
+                    "updated": "2023-10-10",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.14607v2",
+                    "title": "Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications",
+                    "abstract": "Recent literature has suggested the potential of using large language models\n(LLMs) to make classifications for tabular tasks. However, LLMs have been shown\nto exhibit harmful social biases that reflect the stereotypes and inequalities\npresent in society. To this end, as well as the widespread use of tabular data\nin many high-stake applications, it is important to explore the following\nquestions: what sources of information do LLMs draw upon when making\nclassifications for tabular tasks; whether and to what extent are LLM\nclassifications for tabular data influenced by social biases and stereotypes;\nand what are the consequential implications for fairness?\n  Through a series of experiments, we delve into these questions and show that\nLLMs tend to inherit social biases from their training data which significantly\nimpact their fairness in tabular classification tasks. Furthermore, our\ninvestigations show that in the context of bias mitigation, though in-context\nlearning and finetuning have a moderate effect, the fairness metric gap between\ndifferent subgroups is still larger than that in traditional machine learning\nmodels, such as Random Forest and shallow Neural Networks. This observation\nemphasizes that the social biases are inherent within the LLMs themselves and\ninherited from their pretraining corpus, not only from the downstream task\ndatasets. Besides, we demonstrate that label-flipping of in-context examples\ncan significantly reduce biases, further highlighting the presence of inherent\nbias within LLMs.",
+                    "authors": "Yanchen Liu, Srishti Gautam, Jiaqi Ma, Himabindu Lakkaraju",
+                    "published": "2023-10-23",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08472v1",
+                    "title": "Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models",
+                    "abstract": "Recently, work in NLP has shifted to few-shot (in-context) learning, with\nlarge language models (LLMs) performing well across a range of tasks. However,\nwhile fairness evaluations have become a standard for supervised methods,\nlittle is known about the fairness of LLMs as prediction systems. Further,\ncommon standard methods for fairness involve access to models weights or are\napplied during finetuning, which are not applicable in few-shot learning. Do\nLLMs exhibit prediction biases when used for standard NLP tasks? In this work,\nwe explore the effect of shots, which directly affect the performance of\nmodels, on the fairness of LLMs as NLP classification systems. We consider how\ndifferent shot selection strategies, both existing and new demographically\nsensitive methods, affect model fairness across three standard fairness\ndatasets. We discuss how future work can include LLM fairness evaluations.",
+                    "authors": "Carlos Aguirre, Kuleen Sasse, Isabel Cachola, Mark Dredze",
+                    "published": "2023-11-14",
+                    "updated": "2023-11-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.15585v1",
+                    "title": "Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting",
+                    "abstract": "There exist both scalable tasks, like reading comprehension and\nfact-checking, where model performance improves with model size, and unscalable\ntasks, like arithmetic reasoning and symbolic reasoning, where model\nperformance does not necessarily improve with model size. Large language models\n(LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate\nincremental predictions even on unscalable tasks. Unfortunately, despite their\nexceptional reasoning abilities, LLMs tend to internalize and reproduce\ndiscriminatory societal biases. Whether CoT can provide discriminatory or\negalitarian rationalizations for the implicit information in unscalable tasks\nremains an open question.\n  In this study, we examine the impact of LLMs' step-by-step predictions on\ngender bias in unscalable tasks. For this purpose, we construct a benchmark for\nan unscalable task where the LLM is given a list of words comprising feminine,\nmasculine, and gendered occupational words, and is required to count the number\nof feminine and masculine words. In our CoT prompts, we require the LLM to\nexplicitly indicate whether each word in the word list is a feminine or\nmasculine before making the final predictions. With counting and handling the\nmeaning of words, this benchmark has characteristics of both arithmetic\nreasoning and symbolic reasoning. Experimental results in English show that\nwithout step-by-step prediction, most LLMs make socially biased predictions,\ndespite the task being as simple as counting words. Interestingly, CoT\nprompting reduces this unconscious social bias in LLMs and encourages fair\npredictions.",
+                    "authors": "Masahiro Kaneko, Danushka Bollegala, Naoaki Okazaki, Timothy Baldwin",
+                    "published": "2024-01-28",
+                    "updated": "2024-01-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05374v2",
+                    "title": "Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment",
+                    "abstract": "Ensuring alignment, which refers to making models behave in accordance with\nhuman intentions [1,2], has become a critical task before deploying large\nlanguage models (LLMs) in real-world applications. For instance, OpenAI devoted\nsix months to iteratively aligning GPT-4 before its release [3]. However, a\nmajor challenge faced by practitioners is the lack of clear guidance on\nevaluating whether LLM outputs align with social norms, values, and\nregulations. This obstacle hinders systematic iteration and deployment of LLMs.\nTo address this issue, this paper presents a comprehensive survey of key\ndimensions that are crucial to consider when assessing LLM trustworthiness. The\nsurvey covers seven major categories of LLM trustworthiness: reliability,\nsafety, fairness, resistance to misuse, explainability and reasoning, adherence\nto social norms, and robustness. Each major category is further divided into\nseveral sub-categories, resulting in a total of 29 sub-categories.\nAdditionally, a subset of 8 sub-categories is selected for further\ninvestigation, where corresponding measurement studies are designed and\nconducted on several widely-used LLMs. The measurement results indicate that,\nin general, more aligned models tend to perform better in terms of overall\ntrustworthiness. However, the effectiveness of alignment varies across the\ndifferent trustworthiness categories considered. This highlights the importance\nof conducting more fine-grained analyses, testing, and making continuous\nimprovements on LLM alignment. By shedding light on these key dimensions of LLM\ntrustworthiness, this paper aims to provide valuable insights and guidance to\npractitioners in the field. Understanding and addressing these concerns will be\ncrucial in achieving reliable and ethically sound deployment of LLMs in various\napplications.",
+                    "authors": "Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, Hang Li",
+                    "published": "2023-08-10",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.03838v2",
+                    "title": "RADAR: Robust AI-Text Detection via Adversarial Learning",
+                    "abstract": "Recent advances in large language models (LLMs) and the intensifying\npopularity of ChatGPT-like applications have blurred the boundary of\nhigh-quality text generation between humans and machines. However, in addition\nto the anticipated revolutionary changes to our technology and society, the\ndifficulty of distinguishing LLM-generated texts (AI-text) from human-generated\ntexts poses new challenges of misuse and fairness, such as fake content\ngeneration, plagiarism, and false accusations of innocent writers. While\nexisting works show that current AI-text detectors are not robust to LLM-based\nparaphrasing, this paper aims to bridge this gap by proposing a new framework\ncalled RADAR, which jointly trains a robust AI-text detector via adversarial\nlearning. RADAR is based on adversarial training of a paraphraser and a\ndetector. The paraphraser's goal is to generate realistic content to evade\nAI-text detection. RADAR uses the feedback from the detector to update the\nparaphraser, and vice versa. Evaluated with 8 different LLMs (Pythia, Dolly\n2.0, Palmyra, Camel, GPT-J, Dolly 1.0, LLaMA, and Vicuna) across 4 datasets,\nexperimental results show that RADAR significantly outperforms existing AI-text\ndetection methods, especially when paraphrasing is in place. We also identify\nthe strong transferability of RADAR from instruction-tuned LLMs to other LLMs,\nand evaluate the improved capability of RADAR via GPT-3.5-Turbo.",
+                    "authors": "Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho",
+                    "published": "2023-07-07",
+                    "updated": "2023-10-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18580v1",
+                    "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+                    "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+                    "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+                    "published": "2023-11-30",
+                    "updated": "2023-11-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.01769v1",
+                    "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+                    "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+                    "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+                    "published": "2024-05-02",
+                    "updated": "2024-05-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.07420v1",
+                    "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+                    "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+                    "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+                    "published": "2023-12-12",
+                    "updated": "2023-12-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.17916v2",
+                    "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+                    "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+                    "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+                    "published": "2024-02-27",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.11595v3",
+                    "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+                    "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+                    "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+                    "published": "2023-05-19",
+                    "updated": "2023-10-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.09447v2",
+                    "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+                    "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+                    "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+                    "published": "2023-11-15",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14804v1",
+                    "title": "Use large language models to promote equity",
+                    "abstract": "Advances in large language models (LLMs) have driven an explosion of interest\nabout their societal impacts. Much of the discourse around how they will impact\nsocial equity has been cautionary or negative, focusing on questions like \"how\nmight LLMs be biased and how would we mitigate those biases?\" This is a vital\ndiscussion: the ways in which AI generally, and LLMs specifically, can entrench\nbiases have been well-documented. But equally vital, and much less discussed,\nis the more opportunity-focused counterpoint: \"what promising applications do\nLLMs enable that could promote equity?\" If LLMs are to enable a more equitable\nworld, it is not enough just to play defense against their biases and failure\nmodes. We must also go on offense, applying them positively to equity-enhancing\nuse cases to increase opportunities for underserved groups and reduce societal\ndiscrimination. There are many choices which determine the impact of AI, and a\nfundamental choice very early in the pipeline is the problems we choose to\napply it to. If we focus only later in the pipeline -- making LLMs marginally\nmore fair as they facilitate use cases which intrinsically entrench power -- we\nwill miss an important opportunity to guide them to equitable impacts. Here, we\nhighlight the emerging potential of LLMs to promote equity by presenting four\nnewly possible, promising research directions, while keeping risks and\ncautionary points in clear view.",
+                    "authors": "Emma Pierson, Divya Shanmugam, Rajiv Movva, Jon Kleinberg, Monica Agrawal, Mark Dredze, Kadija Ferryman, Judy Wawira Gichoya, Dan Jurafsky, Pang Wei Koh, Karen Levy, Sendhil Mullainathan, Ziad Obermeyer, Harini Suresh, Keyon Vafa",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-22",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06003v1",
+                    "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+                    "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+                    "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2303.01248v3",
+                    "title": "Can ChatGPT Assess Human Personalities? A General Evaluation Framework",
+                    "abstract": "Large Language Models (LLMs) especially ChatGPT have produced impressive\nresults in various areas, but their potential human-like psychology is still\nlargely unexplored. Existing works study the virtual personalities of LLMs but\nrarely explore the possibility of analyzing human personalities via LLMs. This\npaper presents a generic evaluation framework for LLMs to assess human\npersonalities based on Myers Briggs Type Indicator (MBTI) tests. Specifically,\nwe first devise unbiased prompts by randomly permuting options in MBTI\nquestions and adopt the average testing result to encourage more impartial\nanswer generation. Then, we propose to replace the subject in question\nstatements to enable flexible queries and assessments on different subjects\nfrom LLMs. Finally, we re-formulate the question instructions in a manner of\ncorrectness evaluation to facilitate LLMs to generate clearer responses. The\nproposed framework enables LLMs to flexibly assess personalities of different\ngroups of people. We further propose three evaluation metrics to measure the\nconsistency, robustness, and fairness of assessment results from\nstate-of-the-art LLMs including ChatGPT and GPT-4. Our experiments reveal\nChatGPT's ability to assess human personalities, and the average results\ndemonstrate that it can achieve more consistent and fairer assessments in spite\nof lower robustness against prompt biases compared with InstructGPT.",
+                    "authors": "Haocong Rao, Cyril Leung, Chunyan Miao",
+                    "published": "2023-03-01",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.13343v1",
+                    "title": "Challenges and Contributing Factors in the Utilization of Large Language Models (LLMs)",
+                    "abstract": "With the development of large language models (LLMs) like the GPT series,\ntheir widespread use across various application scenarios presents a myriad of\nchallenges. This review initially explores the issue of domain specificity,\nwhere LLMs may struggle to provide precise answers to specialized questions\nwithin niche fields. The problem of knowledge forgetting arises as these LLMs\nmight find it hard to balance old and new information. The knowledge repetition\nphenomenon reveals that sometimes LLMs might deliver overly mechanized\nresponses, lacking depth and originality. Furthermore, knowledge illusion\ndescribes situations where LLMs might provide answers that seem insightful but\nare actually superficial, while knowledge toxicity focuses on harmful or biased\ninformation outputs. These challenges underscore problems in the training data\nand algorithmic design of LLMs. To address these issues, it's suggested to\ndiversify training data, fine-tune models, enhance transparency and\ninterpretability, and incorporate ethics and fairness training. Future\ntechnological trends might lean towards iterative methodologies, multimodal\nlearning, model personalization and customization, and real-time learning and\nfeedback mechanisms. In conclusion, future LLMs should prioritize fairness,\ntransparency, and ethics, ensuring they uphold high moral and ethical standards\nwhen serving humanity.",
+                    "authors": "Xiaoliang Chen, Liangbin Li, Le Chang, Yunhe Huang, Yuxuan Zhao, Yuxiao Zhang, Dinuo Li",
+                    "published": "2023-10-20",
+                    "updated": "2023-10-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.11483v1",
+                    "title": "Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in\nvarious NLP tasks. However, previous works have shown these models are\nsensitive towards prompt wording, and few-shot demonstrations and their order,\nposing challenges to fair assessment of these models. As these models become\nmore powerful, it becomes imperative to understand and address these\nlimitations. In this paper, we focus on LLMs robustness on the task of\nmultiple-choice questions -- commonly adopted task to study reasoning and\nfact-retrieving capability of LLMs. Investigating the sensitivity of LLMs\ntowards the order of options in multiple-choice questions, we demonstrate a\nconsiderable performance gap of approximately 13% to 75% in LLMs on different\nbenchmarks, when answer options are reordered, even when using demonstrations\nin a few-shot setting. Through a detailed analysis, we conjecture that this\nsensitivity arises when LLMs are uncertain about the prediction between the\ntop-2/3 choices, and specific options placements may favor certain prediction\nbetween those top choices depending on the question caused by positional bias.\nWe also identify patterns in top-2 choices that amplify or mitigate the model's\nbias toward option placement. We found that for amplifying bias, the optimal\nstrategy involves positioning the top two choices as the first and last\noptions. Conversely, to mitigate bias, we recommend placing these choices among\nthe adjacent options. To validate our conjecture, we conduct various\nexperiments and adopt two approaches to calibrate LLMs' predictions, leading to\nup to 8 percentage points improvement across different models and benchmarks.",
+                    "authors": "Pouya Pezeshkpour, Estevam Hruschka",
+                    "published": "2023-08-22",
+                    "updated": "2023-08-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02049v1",
+                    "title": "Post Turing: Mapping the landscape of LLM Evaluation",
+                    "abstract": "In the rapidly evolving landscape of Large Language Models (LLMs),\nintroduction of well-defined and standardized evaluation methodologies remains\na crucial challenge. This paper traces the historical trajectory of LLM\nevaluations, from the foundational questions posed by Alan Turing to the modern\nera of AI research. We categorize the evolution of LLMs into distinct periods,\neach characterized by its unique benchmarks and evaluation criteria. As LLMs\nincreasingly mimic human-like behaviors, traditional evaluation proxies, such\nas the Turing test, have become less reliable. We emphasize the pressing need\nfor a unified evaluation system, given the broader societal implications of\nthese models. Through an analysis of common evaluation methodologies, we\nadvocate for a qualitative shift in assessment approaches, underscoring the\nimportance of standardization and objective criteria. This work serves as a\ncall for the AI community to collaboratively address the challenges of LLM\nevaluation, ensuring their reliability, fairness, and societal benefit.",
+                    "authors": "Alexey Tikhonov, Ivan P. Yamshchikov",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.13840v1",
+                    "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+                    "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+                    "published": "2024-03-15",
+                    "updated": "2024-03-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.SI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.04057v1",
+                    "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+                    "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+                    "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+                    "published": "2024-01-08",
+                    "updated": "2024-01-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.07688v1",
+                    "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+                    "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+                    "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+                    "published": "2024-02-12",
+                    "updated": "2024-02-12",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.10199v3",
+                    "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+                    "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+                    "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+                    "published": "2024-04-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.01964v1",
+                    "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+                    "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+                    "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00811v1",
+                    "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+                    "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+                    "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+                    "published": "2024-02-25",
+                    "updated": "2024-02-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.15997v1",
+                    "title": "RoCar: A Relationship Network-based Evaluation Method to Large Language Models",
+                    "abstract": "Large language models (LLMs) have received increasing attention. However, due\nto the complexity of its capabilities, how to rationally evaluate the\ncapabilities of LLMs is still a task to be solved. We propose the RoCar method,\nwhich utilizes the defined basic schemas to randomly construct a task graph and\ngenerates natural language evaluation tasks based on the task graph to evaluate\nthe reasoning and memory abilities of LLMs respectively. Due to the very large\nrandomness of the task construction process, it is possible to ensure that none\nof the LLMs to be tested has directly learned the evaluation tasks,\nguaranteeing the fairness of the evaluation method.",
+                    "authors": "Ming Wang, Wenfang Wu, Chongyun Gao, Daling Wang, Shi Feng, Yifei Zhang",
+                    "published": "2023-07-29",
+                    "updated": "2023-07-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.07884v2",
+                    "title": "Fair Abstractive Summarization of Diverse Perspectives",
+                    "abstract": "People from different social and demographic groups express diverse\nperspectives and conflicting opinions on a broad set of topics such as product\nreviews, healthcare, law, and politics. A fair summary should provide a\ncomprehensive coverage of diverse perspectives without underrepresenting\ncertain groups. However, current work in summarization metrics and Large\nLanguage Models (LLMs) evaluation has not explored fair abstractive\nsummarization. In this paper, we systematically investigate fair abstractive\nsummarization for user-generated data. We first formally define fairness in\nabstractive summarization as not underrepresenting perspectives of any groups\nof people, and we propose four reference-free automatic metrics by measuring\nthe differences between target and source perspectives. We evaluate nine LLMs,\nincluding three GPT models, four LLaMA models, PaLM 2, and Claude, on six\ndatasets collected from social media, online reviews, and recorded transcripts.\nExperiments show that both the model-generated and the human-written reference\nsummaries suffer from low fairness. We conduct a comprehensive analysis of the\ncommon factors influencing fairness and propose three simple but effective\nmethods to alleviate unfair summarization. Our dataset and code are available\nat https://github.com/psunlpgroup/FairSumm.",
+                    "authors": "Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang",
+                    "published": "2023-11-14",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18130v2",
+                    "title": "DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues",
+                    "abstract": "Controversy is a reflection of our zeitgeist, and an important aspect to any\ndiscourse. The rise of large language models (LLMs) as conversational systems\nhas increased public reliance on these systems for answers to their various\nquestions. Consequently, it is crucial to systematically examine how these\nmodels respond to questions that pertaining to ongoing debates. However, few\nsuch datasets exist in providing human-annotated labels reflecting the\ncontemporary discussions. To foster research in this area, we propose a novel\nconstruction of a controversial questions dataset, expanding upon the publicly\nreleased Quora Question Pairs Dataset. This dataset presents challenges\nconcerning knowledge recency, safety, fairness, and bias. We evaluate different\nLLMs using a subset of this dataset, illuminating how they handle controversial\nissues and the stances they adopt. This research ultimately contributes to our\nunderstanding of LLMs' interaction with controversial issues, paving the way\nfor improvements in their comprehension and handling of complex societal\ndebates.",
+                    "authors": "David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein, Jason D. Williams",
+                    "published": "2023-10-27",
+                    "updated": "2023-11-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09606v1",
+                    "title": "Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey",
+                    "abstract": "Causal inference has shown potential in enhancing the predictive accuracy,\nfairness, robustness, and explainability of Natural Language Processing (NLP)\nmodels by capturing causal relationships among variables. The emergence of\ngenerative Large Language Models (LLMs) has significantly impacted various NLP\ndomains, particularly through their advanced reasoning capabilities. This\nsurvey focuses on evaluating and improving LLMs from a causal view in the\nfollowing areas: understanding and improving the LLMs' reasoning capacity,\naddressing fairness and safety issues in LLMs, complementing LLMs with\nexplanations, and handling multimodality. Meanwhile, LLMs' strong reasoning\ncapacities can in turn contribute to the field of causal inference by aiding\ncausal relationship discovery and causal effect estimations. This review\nexplores the interplay between causal inference frameworks and LLMs from both\nperspectives, emphasizing their collective potential to further the development\nof more advanced and equitable artificial intelligence systems.",
+                    "authors": "Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.08836v2",
+                    "title": "Bias and Fairness in Chatbots: An Overview",
+                    "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+                    "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+                    "published": "2023-09-16",
+                    "updated": "2023-12-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.12150v1",
+                    "title": "Your Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One",
+                    "abstract": "The widespread adoption of large language models (LLMs) underscores the\nurgent need to ensure their fairness. However, LLMs frequently present dominant\nviewpoints while ignoring alternative perspectives from minority parties,\nresulting in potential biases. We hypothesize that these fairness-violating\nbehaviors occur because LLMs express their viewpoints using a human personality\nthat represents the majority of training data. In response to this, we validate\nthat prompting LLMs with specific roles can allow LLMs to express diverse\nviewpoints. Building on this insight and observation, we develop FairThinking,\na pipeline designed to automatically generate roles that enable LLMs to\narticulate diverse perspectives for fair expressions. To evaluate FairThinking,\nwe create a dataset with a thousand items covering three fairness-related\ntopics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to\ndemonstrate its superior performance.",
+                    "authors": "Tianlin Li, Xiaoyu Zhang, Chao Du, Tianyu Pang, Qian Liu, Qing Guo, Chao Shen, Yang Liu",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "I.2; J.4"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11406v2",
+                    "title": "Don't Go To Extremes: Revealing the Excessive Sensitivity and Calibration Limitations of LLMs in Implicit Hate Speech Detection",
+                    "abstract": "The fairness and trustworthiness of Large Language Models (LLMs) are\nreceiving increasing attention. Implicit hate speech, which employs indirect\nlanguage to convey hateful intentions, occupies a significant portion of\npractice. However, the extent to which LLMs effectively address this issue\nremains insufficiently examined. This paper delves into the capability of LLMs\nto detect implicit hate speech (Classification Task) and express confidence in\ntheir responses (Calibration Task). Our evaluation meticulously considers\nvarious prompt patterns and mainstream uncertainty estimation methods. Our\nfindings highlight that LLMs exhibit two extremes: (1) LLMs display excessive\nsensitivity towards groups or topics that may cause fairness issues, resulting\nin misclassifying benign statements as hate speech. (2) LLMs' confidence scores\nfor each method excessively concentrate on a fixed range, remaining unchanged\nregardless of the dataset's complexity. Consequently, the calibration\nperformance is heavily reliant on primary classification accuracy. These\ndiscoveries unveil new limitations of LLMs, underscoring the need for caution\nwhen optimizing models to ensure they do not veer towards extremes. This serves\nas a reminder to carefully consider sensitivity and confidence in the pursuit\nof model fairness.",
+                    "authors": "Min Zhang, Jianfeng He, Taoran Ji, Chang-Tien Lu",
+                    "published": "2024-02-18",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05345v3",
+                    "title": "RTLLM: An Open-Source Benchmark for Design RTL Generation with Large Language Model",
+                    "abstract": "Inspired by the recent success of large language models (LLMs) like ChatGPT,\nresearchers start to explore the adoption of LLMs for agile hardware design,\nsuch as generating design RTL based on natural-language instructions. However,\nin existing works, their target designs are all relatively simple and in a\nsmall scale, and proposed by the authors themselves, making a fair comparison\namong different LLM solutions challenging. In addition, many prior works only\nfocus on the design correctness, without evaluating the design qualities of\ngenerated design RTL. In this work, we propose an open-source benchmark named\nRTLLM, for generating design RTL with natural language instructions. To\nsystematically evaluate the auto-generated design RTL, we summarized three\nprogressive goals, named syntax goal, functionality goal, and design quality\ngoal. This benchmark can automatically provide a quantitative evaluation of any\ngiven LLM-based solution. Furthermore, we propose an easy-to-use yet\nsurprisingly effective prompt engineering technique named self-planning, which\nproves to significantly boost the performance of GPT-3.5 in our proposed\nbenchmark.",
+                    "authors": "Yao Lu, Shang Liu, Qijun Zhang, Zhiyao Xie",
+                    "published": "2023-08-10",
+                    "updated": "2023-11-11",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.05694v1",
+                    "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+                    "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+                    "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+                    "published": "2023-10-09",
+                    "updated": "2023-10-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18333v3",
+                    "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+                    "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+                    "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+                    "published": "2023-10-20",
+                    "updated": "2023-12-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15491v1",
+                    "title": "Open Source Conversational LLMs do not know most Spanish words",
+                    "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+                    "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.14473v1",
+                    "title": "The Ethics of ChatGPT in Medicine and Healthcare: A Systematic Review on Large Language Models (LLMs)",
+                    "abstract": "With the introduction of ChatGPT, Large Language Models (LLMs) have received\nenormous attention in healthcare. Despite their potential benefits, researchers\nhave underscored various ethical implications. While individual instances have\ndrawn much attention, the debate lacks a systematic overview of practical\napplications currently researched and ethical issues connected to them. Against\nthis background, this work aims to map the ethical landscape surrounding the\ncurrent stage of deployment of LLMs in medicine and healthcare. Electronic\ndatabases and preprint servers were queried using a comprehensive search\nstrategy. Studies were screened and extracted following a modified rapid review\napproach. Methodological quality was assessed using a hybrid approach. For 53\nrecords, a meta-aggregative synthesis was performed. Four fields of\napplications emerged and testify to a vivid exploration phase. Advantages of\nusing LLMs are attributed to their capacity in data analysis, personalized\ninformation provisioning, support in decision-making, mitigating information\nloss and enhancing information accessibility. However, we also identifies\nrecurrent ethical concerns connected to fairness, bias, non-maleficence,\ntransparency, and privacy. A distinctive concern is the tendency to produce\nharmful misinformation or convincingly but inaccurate content. A recurrent plea\nfor ethical guidance and human oversight is evident. Given the variety of use\ncases, it is suggested that the ethical guidance debate be reframed to focus on\ndefining what constitutes acceptable human oversight across the spectrum of\napplications. This involves considering diverse settings, varying potentials\nfor harm, and different acceptable thresholds for performance and certainty in\nhealthcare. In addition, a critical inquiry is necessary to determine the\nextent to which the current experimental use of LLMs is necessary and\njustified.",
+                    "authors": "Joschka Haltaufderheide, Robert Ranisch",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.05668v1",
+                    "title": "CFaiRLLM: Consumer Fairness Evaluation in Large-Language Model Recommender System",
+                    "abstract": "In the evolving landscape of recommender systems, the integration of Large\nLanguage Models (LLMs) such as ChatGPT marks a new era, introducing the concept\nof Recommendation via LLM (RecLLM). While these advancements promise\nunprecedented personalization and efficiency, they also bring to the fore\ncritical concerns regarding fairness, particularly in how recommendations might\ninadvertently perpetuate or amplify biases associated with sensitive user\nattributes. In order to address these concerns, our study introduces a\ncomprehensive evaluation framework, CFaiRLLM, aimed at evaluating (and thereby\nmitigating) biases on the consumer side within RecLLMs.\n  Our research methodically assesses the fairness of RecLLMs by examining how\nrecommendations might vary with the inclusion of sensitive attributes such as\ngender, age, and their intersections, through both similarity alignment and\ntrue preference alignment. By analyzing recommendations generated under\ndifferent conditions-including the use of sensitive attributes in user\nprompts-our framework identifies potential biases in the recommendations\nprovided. A key part of our study involves exploring how different detailed\nstrategies for constructing user profiles (random, top-rated, recent) impact\nthe alignment between recommendations made without consideration of sensitive\nattributes and those that are sensitive-attribute-aware, highlighting the bias\nmechanisms within RecLLMs.\n  The findings in our study highlight notable disparities in the fairness of\nrecommendations, particularly when sensitive attributes are integrated into the\nrecommendation process, either individually or in combination. The analysis\ndemonstrates that the choice of user profile sampling strategy plays a\nsignificant role in affecting fairness outcomes, highlighting the complexity of\nachieving fair recommendations in the era of LLMs.",
+                    "authors": "Yashar Deldjoo, Tommaso di Noia",
+                    "published": "2024-03-08",
+                    "updated": "2024-03-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2206.13757v1",
+                    "title": "Flexible text generation for counterfactual fairness probing",
+                    "abstract": "A common approach for testing fairness issues in text-based classifiers is\nthrough the use of counterfactuals: does the classifier output change if a\nsensitive attribute in the input is changed? Existing counterfactual generation\nmethods typically rely on wordlists or templates, producing simple\ncounterfactuals that don't take into account grammar, context, or subtle\nsensitive attribute references, and could miss issues that the wordlist\ncreators had not considered. In this paper, we introduce a task for generating\ncounterfactuals that overcomes these shortcomings, and demonstrate how large\nlanguage models (LLMs) can be leveraged to make progress on this task. We show\nthat this LLM-based method can produce complex counterfactuals that existing\nmethods cannot, comparing the performance of various counterfactual generation\nmethods on the Civil Comments dataset and showing their value in evaluating a\ntoxicity classifier.",
+                    "authors": "Zee Fryer, Vera Axelrod, Ben Packer, Alex Beutel, Jilin Chen, Kellie Webster",
+                    "published": "2022-06-28",
+                    "updated": "2022-06-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.15215v1",
+                    "title": "Item-side Fairness of Large Language Model-based Recommendation System",
+                    "abstract": "Recommendation systems for Web content distribution intricately connect to\nthe information access and exposure opportunities for vulnerable populations.\nThe emergence of Large Language Models-based Recommendation System (LRS) may\nintroduce additional societal challenges to recommendation systems due to the\ninherent biases in Large Language Models (LLMs). From the perspective of\nitem-side fairness, there remains a lack of comprehensive investigation into\nthe item-side fairness of LRS given the unique characteristics of LRS compared\nto conventional recommendation systems. To bridge this gap, this study examines\nthe property of LRS with respect to item-side fairness and reveals the\ninfluencing factors of both historical users' interactions and inherent\nsemantic biases of LLMs, shedding light on the need to extend conventional\nitem-side fairness methods for LRS. Towards this goal, we develop a concise and\neffective framework called IFairLRS to enhance the item-side fairness of an\nLRS. IFairLRS covers the main stages of building an LRS with specifically\nadapted strategies to calibrate the recommendations of LRS. We utilize IFairLRS\nto fine-tune LLaMA, a representative LLM, on \\textit{MovieLens} and\n\\textit{Steam} datasets, and observe significant item-side fairness\nimprovements. The code can be found in\nhttps://github.com/JiangM-C/IFairLRS.git.",
+                    "authors": "Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, Xiangnan He",
+                    "published": "2024-02-23",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.08189v1",
+                    "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+                    "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+                    "authors": "Karthik Sreedhar, Lydia Chilton",
+                    "published": "2024-02-13",
+                    "updated": "2024-02-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02294v1",
+                    "title": "LLMs grasp morality in concept",
+                    "abstract": "Work in AI ethics and fairness has made much progress in regulating LLMs to\nreflect certain values, such as fairness, truth, and diversity. However, it has\ntaken the problem of how LLMs might 'mean' anything at all for granted. Without\naddressing this, it is not clear what imbuing LLMs with such values even means.\nIn response, we provide a general theory of meaning that extends beyond humans.\nWe use this theory to explicate the precise nature of LLMs as meaning-agents.\nWe suggest that the LLM, by virtue of its position as a meaning-agent, already\ngrasps the constructions of human society (e.g. morality, gender, and race) in\nconcept. Consequently, under certain ethical frameworks, currently popular\nmethods for model alignment are limited at best and counterproductive at worst.\nMoreover, unaligned models may help us better develop our moral and social\nphilosophy.",
+                    "authors": "Mark Pock, Andre Ye, Jared Moore",
+                    "published": "2023-11-04",
+                    "updated": "2023-11-04",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.06852v2",
+                    "title": "ChemLLM: A Chemical Large Language Model",
+                    "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+                    "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+                    "published": "2024-02-10",
+                    "updated": "2024-04-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.01262v2",
+                    "title": "Fairness Certification for Natural Language Processing and Large Language Models",
+                    "abstract": "Natural Language Processing (NLP) plays an important role in our daily lives,\nparticularly due to the enormous progress of Large Language Models (LLM).\nHowever, NLP has many fairness-critical use cases, e.g., as an expert system in\nrecruitment or as an LLM-based tutor in education. Since NLP is based on human\nlanguage, potentially harmful biases can diffuse into NLP systems and produce\nunfair results, discriminate against minorities or generate legal issues.\nHence, it is important to develop a fairness certification for NLP approaches.\nWe follow a qualitative research approach towards a fairness certification for\nNLP. In particular, we have reviewed a large body of literature on algorithmic\nfairness, and we have conducted semi-structured expert interviews with a wide\nrange of experts from that area. We have systematically devised six fairness\ncriteria for NLP, which can be further refined into 18 sub-categories. Our\ncriteria offer a foundation for operationalizing and testing processes to\ncertify fairness, both from the perspective of the auditor and the audited\norganization.",
+                    "authors": "Vincent Freiberger, Erik Buchmann",
+                    "published": "2024-01-02",
+                    "updated": "2024-01-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04205v2",
+                    "title": "Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves",
+                    "abstract": "Misunderstandings arise not only in interpersonal communication but also\nbetween humans and Large Language Models (LLMs). Such discrepancies can make\nLLMs interpret seemingly unambiguous questions in unexpected ways, yielding\nincorrect responses. While it is widely acknowledged that the quality of a\nprompt, such as a question, significantly impacts the quality of the response\nprovided by LLMs, a systematic method for crafting questions that LLMs can\nbetter comprehend is still underdeveloped. In this paper, we present a method\nnamed `Rephrase and Respond' (RaR), which allows LLMs to rephrase and expand\nquestions posed by humans and provide responses in a single prompt. This\napproach serves as a simple yet effective prompting method for improving\nperformance. We also introduce a two-step variant of RaR, where a rephrasing\nLLM first rephrases the question and then passes the original and rephrased\nquestions together to a different responding LLM. This facilitates the\neffective utilization of rephrased questions generated by one LLM with another.\nOur experiments demonstrate that our methods significantly improve the\nperformance of different models across a wide range to tasks. We further\nprovide a comprehensive comparison between RaR and the popular Chain-of-Thought\n(CoT) methods, both theoretically and empirically. We show that RaR is\ncomplementary to CoT and can be combined with CoT to achieve even better\nperformance. Our work not only contributes to enhancing LLM performance\nefficiently and effectively but also sheds light on a fair evaluation of LLM\ncapabilities. Data and codes are available at\nhttps://github.com/uclaml/Rephrase-and-Respond.",
+                    "authors": "Yihe Deng, Weitong Zhang, Zixiang Chen, Quanquan Gu",
+                    "published": "2023-11-07",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15451v1",
+                    "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+                    "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+                    "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+                    "published": "2024-03-18",
+                    "updated": "2024-03-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.11033v4",
+                    "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+                    "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+                    "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+                    "published": "2024-01-19",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.18276v1",
+                    "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+                    "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+                    "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+                    "published": "2024-04-28",
+                    "updated": "2024-04-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "D.1; I.2"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.02680v1",
+                    "title": "Large Language Models are Geographically Biased",
+                    "abstract": "Large Language Models (LLMs) inherently carry the biases contained in their\ntraining corpora, which can lead to the perpetuation of societal harm. As the\nimpact of these foundation models grows, understanding and evaluating their\nbiases becomes crucial to achieving fairness and accuracy. We propose to study\nwhat LLMs know about the world we live in through the lens of geography. This\napproach is particularly powerful as there is ground truth for the numerous\naspects of human life that are meaningfully projected onto geographic space\nsuch as culture, race, language, politics, and religion. We show various\nproblematic geographic biases, which we define as systemic errors in geospatial\npredictions. Initially, we demonstrate that LLMs are capable of making accurate\nzero-shot geospatial predictions in the form of ratings that show strong\nmonotonic correlation with ground truth (Spearman's $\\rho$ of up to 0.89). We\nthen show that LLMs exhibit common biases across a range of objective and\nsubjective topics. In particular, LLMs are clearly biased against locations\nwith lower socioeconomic conditions (e.g. most of Africa) on a variety of\nsensitive subjective topics such as attractiveness, morality, and intelligence\n(Spearman's $\\rho$ of up to 0.70). Finally, we introduce a bias score to\nquantify this and find that there is significant variation in the magnitude of\nbias across existing LLMs.",
+                    "authors": "Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon",
+                    "published": "2024-02-05",
+                    "updated": "2024-02-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                }
+            ],
+            [
+                {
+                    "url": "http://arxiv.org/abs/2404.14445v1",
+                    "title": "A Multi-Faceted Evaluation Framework for Assessing Synthetic Data Generated by Large Language Models",
+                    "abstract": "The rapid advancements in generative AI and large language models (LLMs) have\nopened up new avenues for producing synthetic data, particularly in the realm\nof structured tabular formats, such as product reviews. Despite the potential\nbenefits, concerns regarding privacy leakage have surfaced, especially when\npersonal information is utilized in the training datasets. In addition, there\nis an absence of a comprehensive evaluation framework capable of quantitatively\nmeasuring the quality of the generated synthetic data and their utility for\ndownstream tasks. In response to this gap, we introduce SynEval, an open-source\nevaluation framework designed to assess the fidelity, utility, and privacy\npreservation of synthetically generated tabular data via a suite of diverse\nevaluation metrics. We validate the efficacy of our proposed framework -\nSynEval - by applying it to synthetic product review data generated by three\nstate-of-the-art LLMs: ChatGPT, Claude, and Llama. Our experimental findings\nilluminate the trade-offs between various evaluation metrics in the context of\nsynthetic data generation. Furthermore, SynEval stands as a critical instrument\nfor researchers and practitioners engaged with synthetic tabular data,,\nempowering them to judiciously determine the suitability of the generated data\nfor their specific applications, with an emphasis on upholding user privacy.",
+                    "authors": "Yefeng Yuan, Yuhong Liu, Liang Cheng",
+                    "published": "2024-04-20",
+                    "updated": "2024-04-20",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "label": "Original Paper",
+                    "paper_cat": "LLM Fairness",
+                    "gt": "2.1 Synthetic Data Generation Synthetic data generation has gained significant attention in recent years to address privacy concerns, data scarcity, and data imbalance issues. Various approaches have been proposed for generating synthetic data, including statistical models [16], generative adversarial networks (GANs) [17], variational autoencoders (VAEs) [18], and more recently, LLMs [19]. Statistical models, such as Bayesian networks [20] and copulas [21], have been used to generate synthetic data by learning the underlying probability distributions of the real data. These models often require domain knowledge and explicit modeling of the data dependencies, which can be challenging for complex datasets. GANs have emerged as a popular approach for generating realistic synthetic data by training a generator network to produce samples that are indistinguishable from real data. GANs have been successfully applied to various domains, including image generation [22], text generation [23], and tabular data generation [24]. However, training GANs can be challenging, and ensuring the stability and convergence of the training process remains an active area of research [25]. VAEs, on the other hand, learn a latent representation of the data and generate synthetic samples by sampling from the learned latent space. VAEs have been used to generate synthetic tabular data and have shown promising results in preserving the statistical properties of the real data. However, the generated samples may lack diversity and realism [18]. 2 Figure 1: Framework Overview Recently, LLMs, such as GPT-3 [26], Claude [27], and Llama [28] have demonstrated remarkable capabilities in generating coherent and realistic text. These models have been pre-trained on vast amounts of diverse text data and can be fine-tuned for specific tasks, including synthetic data generation [29]. 2.2 Quantitative Evaluation Metrics With diverse approaches for generating synthetic data, it is essential to quantitatively evaluate the quality of generated synthetic data, which can help compare different generation approaches and ensure diverse, balanced, and privacypreserving data for downstream tasks. To this end, various evaluation metrics have been proposed. Statistical similarity metrics, such as Jensen-Shannon divergence [30], and maximum mean discrepancy [31], have been introduced to quantify the statistical differences between synthetic data and real data. These measures capture the overall statistical properties of the data but may not account for the specific characteristics of individual features or feature correlations. To evaluate the utility of synthetic data for downstream machine learning tasks, typical machine learning performance metrics, such as accuracy, precision, recall, and F1-score, are often used by existing literature [32]. By training machine learning models on synthetic data and evaluating their performance on real data, researchers can assess the extent to which the synthetic data retains the inference capability of the real data. However, these metrics may not capture the nuances and biases presented in the synthetic data. Privacy evaluation frameworks, such as differential privacy(DP) [33], k-anonymity [34], and l-diversity [35], have been proposed to assess the privacy risks associated with synthetic data. However, these techniques often assume a specific level of attacker knowledge and are inadequate to evaluate general privacy risks. Differential privacy has emerged as a strong mathematical framework that provides quantitative privacy guarantees by adding carefully calibrated noise to the data or query results. Nevertheless, evaluating whether the synthetically generated data satisfies DP requirements is challenging due to the high computational overhead. Therefore, in the proposed SynEval framework, we evaluate the privacy of synthetic data by checking its robustness against privacy attacks, specifically the Membership Inference Attacks (MIA) [36]. Existing evaluation frameworks often focus on specific aspects of synthetic data quality and may not provide a comprehensive assessment of the generated data. Moreover, evaluating synthetic tabular data generated by LLMs presents unique challenges due to the complex nature of the data, which often comprises a combination of discrete categorical variables and free-form text fields, along with intricate relationships and dependencies between these elements that must be preserved to ensure the utility and integrity of the data, while also maintaining semantic consistency, coherence, and the protection of sensitive information to mitigate privacy risks. Therefore, a comprehensive evaluation framework that integrates multiple perspectives and is tailored to the characteristics of synthetic tabular data generated by LLMs is needed.",
+                    "pre_questions": [],
+                    "main_content": "Introduction The proliferation of high-quality data serves as the cornerstone propelling the current advancements in artificial intelligence (AI). Nonetheless, real-world datasets frequently suffer from noise and imbalances, impeding the performance of AI models trained on such data. For example, in the context of product reviews, there is often a lack of data representing marginalized communities, such as disabled users. This data scarcity can lead to biased and inaccurate models that fail to capture the diverse needs and experiences of all users. Also, collecting and labeling real data can be expensive, time consuming, and error-prone. For instance, it is reported that, on average, data labeling can cost organizations $2.3 million per year and 16 weeks to perform supervised learning on a new project[1]. Furthermore, privacy regulations make it even more difficult for companies to utilize their customers\u2019 data even if such data is available at their premises. For example, the fines imposed by GDPR in 2023 alone have already exceeded the total amount of fines from 2019 to 2021 [2], with a record-breaking $1.28 billion imposed for Meta\u2019s failure in complying with GDPR during its data transfer from EU to the U.S. for Facebook services [3]. As a result, rather than directly utilizing real data, generating synthetic data that can sufficiently represent the statistical properties of real data while preserving individual user privacy is gaining extensive attention. Compared to real data, synthetic data can be generated in large quantities much faster than the same amount of real data can be collected, and can be more easily manipulated to allow more precise and controlled training and testing of AI models [4, 5]. According to Gartner, it is estimated that synthetic data will play a dominant role in AI models by 2030, and 89% of tech executives believed that synthetic data is a key for companies to stay competitive [6]. 1https://github.com/yefyuan/SynEval arXiv:2404.14445v1  [cs.LG]  20 Apr 2024 Due to its numerous benefits, synthetic data generation has attracted interests from a wide range of companies, such as Google, Microsoft, Amazon, Facebook, Nvidia [7, 8, 9]. For example, Google Cloud recently announced partnership with Gretel for generating anonymized and safe synthetic data [10]. Microsoft has collaborated with Harvard to develop synthetic data generation tools for responsive AI [11]. Amazon has developed in-house synthetic data tools for training computer vision models (AWS, 2023). Facebook acquired a synthetic data generator startup AI.Reverie [12]. Nvidia is creating a synthetic data generation engine for training AI networks [13]. Recent advancements in generative AI, such as Generative Adversarial Networks (GANs) [14] and Large Language Models (LLMs) [15], have shown promising results in generating realistic synthetic data. However, existing evaluation frameworks for synthetic data often focus on one or two specific aspects. There is a lack of a comprehensive evaluation framework that merges multiple perspectives and offers a holistic suite of metrics for the assessment of synthetic data. This poses a significant challenge in the adoption and trustworthiness of synthetic data generation techniques. Therefore, in this work, we propose SynEval, a comprehensive framework to rigorously evaluate the effectiveness of synthetic data generation from these critical dimensions: \u2022 Fidelity: This aspect focuses on the degree to which synthetic data replicates the statistical characteristics of the original dataset. \u2022 Utility: The utility of synthetic data is determined by its effectiveness in facilitating various downstream machine learning tasks. This involves evaluating whether models trained on synthetic data perform comparably to those trained on real data when tested on validation or real-world datasets. \u2022 Privacy: Privacy protection is paramount to ensure that the synthetic data does not reveal sensitive information about individuals in the original dataset. Privacy can be assessed through techniques like re-identification risk analysis and other privacy-preserving metrics. The development of robust evaluation metrics and frameworks like SynEval is critical for the advancement of synthetic data generation methods. Without such tools, it is challenging to gauge the quality and safety of synthetic data, which can hinder its adoption in sensitive domains such as ecommerce, healthcare, and finance. The major contributions of this work are as follows: \u2022 We propose a multi-faceted evaluation framework that integrates data fidelity, utility, and privacy evaluation with a comprehensive set of evaluation metrics to provide a holistic assessment of synthetically generated data. \u2022 We demonstrate the effectiveness of the proposed framework by applying it to synthetic product review data with a comprehensive set of evaluation metrics to provide a holistic assessment of synthetically generated data. \u2022 We demonstrate the effectiveness of the proposed framework by applying it to synthetic product review data generated by three prominent LLMs: ChatGPT, Claude, and Llama. \u2022 We provide insights and recommendations based on the evaluation results, highlighting the strengths and generated by three prominent LLMs: ChatGPT, Claude, and Llama. \u2022 We provide insights and recommendations based on the evaluation results, highlighting the strengths and limitations of each large language model in generating high-quality, useful, and privacy-preserving synthetic tabular data. By addressing the research gap and providing a comprehensive evaluation framework, the proposed work contributes to the advancement of synthetic data generation techniques and promotes the responsible and trustworthy use of synthetic data in various applications. The proposed framework serves as a valuable tool for researchers and practitioners to assess the effectiveness of synthetic tabular data generated by LLMs and make informed decisions regarding their deployment in real-world scenarios. Figure 1 presents an overview of the proposed framework, which comprises three evaluations: fidelity, utility, and privacy. The fidelity evaluation employs metrics such as structure preservation, data integrity, and column shape calculations for non-text tabular data. Additionally, it includes metrics for review text, such as sentiment distribution, top keywords and sentiment-related words, and average length. The utility evaluation is determined by calculating the accuracy of downstream sentiment classification. Lastly, the privacy evaluation is based on the success rate of membership inference attacks. 3 3.1 Fidelity Evaluation Data fidelity evaluation is a crucial component of the proposed framework, as it assesses the degree to which the synthetic tabular data resembles the real data. Ensuring high fidelity is essential for building trust in the generated synthetic data and enabling its effective use in various applications, such as model training, testing, and decision-making. In particular, the proposed framework emphasizes the importance of comprehensive data fidelity evaluation, considering multiple aspects of data similarity and quality. Different from existing literature, the proposed data fidelity evaluation includes not only simple statistical comparisons but also an in-depth analysis of the relationships, dependencies, and contextual information present in the data. First, for non-text tabular data, we introduce the Structure Preserving Score (SPS) to evaluate whether the synthetic data retains the same column names and order as the real data. This assessment involves identifying all column names in the real (r) and synthetic (s) datasets and calculating the overlap between these columns to derive a fidelity score. SPS = |r \u2229s| |r \u222as|[37] (1) Next, we introduce Integrity Score (IS) to assess the integrity of the non-text tabular data. Specifically, the computation of IS depends on whether the data is continuous or discrete. The overall IS is the average IS of continuous and discrete data. For continuous data, such as timestamps, we assess whether synthetic columns adhere to the minimum and maximum values defined by the corresponding real columns. For discrete data, such as ratings and verified purchases, the proposed approach checks if synthetic columns maintain the same category values as those in the real data. It is crucial that the synthetic data does not introduce new category values absent from the real dataset. The metric first recognizes set of unique categories from a real column (Cr), then it calculates the percentage of synthetic data (s) that are found in the set compared to all synthetic data points in this column. IS = |s, s \u2208Cr| |s| [37] (2) The third evaluation criterion for non-text tabular data assesses whether the synthetic data accurately capture the distribution shape of each column. Column shape refers to its overall distribution pattern. To evaluate this, we employ different methods depending on the data type. For continuous numerical values, such as helpful vote, we utilize the Kolmogorov-Smirnov [38] statistic to measure the similarity in marginal distributions between the synthetic and real columns. For discrete data, such as verified purchases, we apply the Total Variation Distance (TVD) [37] to compare the synthetic and real columns. The overall column shapes score is then calculated as the average across all columns. To rigorously evaluate the fidelity of textual review data, we have developed a text analysis tool. This tool begins by assessing the sentiment distribution of each review, categorizing them as positive, neutral, or negative. We then compare the distribution of these sentiment categories between the real and synthetic data to identify and analyze dominant sentiments, ensuring that the synthetic data accurately reflects the emotional tone of the real reviews. Furthermore, we conduct a detailed lexical analysis by extracting the top three keywords and sentiment-related words from the real dataset and synthetic data. This step involves not only identifying these pivotal words but also quantifying their frequency to establish a benchmark for comparison. We then evaluate how well these keywords and sentiment indicators are replicated in the synthetic data, which provides insight into the synthetic model\u2019s ability to capture and reproduce the key thematic elements of the real text. Additionally, we measure the average length of the reviews in both the real and synthetic data. This metric is crucial as it helps determine whether the synthetic data can maintain the same level of detail as the real reviews, which is important for applications where text length may influence the perceived quality or informativeness of the content. Through these multifaceted analytical approaches, we ensure a comprehensive evaluation of the synthetic textual data, gauging its fidelity not only in mimicking the overt sentiments expressed in the reviews but also in preserving the underlying thematic and stylistic nuances that characterize the real dataset. 3.2 Utility Evaluation The utility evaluation component of the proposed framework plays a crucial role in assessing the usefulness of the generated synthetic tabular data for downstream machine learning tasks. It is essential to ensure that the synthetic data retains the predictive power and informativeness of the real data, enabling the development of accurate and reliable machine learning models. To evaluate the utility of the synthetic tabular data, we propose adopting a framework named 4 TSTR (Train-Synthetic-Test-Real), which offers a robust measure of synthetic data quality by assessing how well machine learning models trained on synthetic data perform on previously unseen real data. In the context of product reviews, a common machine learning task is sentiment classification, which involves predicting the sentiment (positive, negative, or neutral) of a given review based on its text content. We begin by extracting (review, rating) pairs from both the original and synthetic datasets to train sentiment classification models using logistic regression. The models are then validated on a separate set of untouched real data to ensure robustness and prevent overfitting. Accuracy is determined by how closely the predicted ratings align with the actual ratings from the review texts, incorporating a tolerance threshold to account for minor variations. Additionally, we calculate the Mean Absolute Error (MAE) across all test instances to quantify the average prediction error, providing a clear metric of the models\u2019 performance in real-world scenarios. By comparing models trained on synthetic data to those trained on real data, we assess whether synthetic data preserves the utility of its real counterparts. This utility evaluation offers an in-depth analysis of the synthetic tabular data\u2019s effectiveness, extending beyond basic statistical comparisons. It directly gauges the performance of synthetic data in real-world machine learning applications, providing valuable insights into its quality and utility. 3.3 Privacy Evaluation In the privacy evaluation of the proposed work, we employ Membership Inference Attacks (MIA) [39] to assess whether synthetic datasets can reliably mimic the privacy attributes of real datasets, hence protecting the anonymity of the underlying data sources. This approach not only tests the robustness of the privacy-preserving measures but also quantifies the risk of sensitive information leakage through synthetic data. We initiate the process by loading and preprocessing both real and synthetic datasets, which include data labeled as from \u2019members\u2019 (real dataset) and \u2019non-members\u2019 (synthetic dataset). For each dataset, categorical features such as product IDs, parent product IDs, and user IDs are encoded using label encoding. The datasets processed include real member data, real non-member data, and synthetic data generated by different language models. For the features\u2019 transformation, we implement a column transformer that processes different types of data. Using the prepared datasets, we construct a combined dataset of real members and a randomly selected half of all synthetic datasets. This combined dataset serves as our training set. The test set comprises real non-member data combined with the remaining half of the synthetic data. A Random Forest Classifier is trained on this mixed dataset within a pipeline that includes the preliminary feature transformation. We assess the classifier\u2019s performance by calculating successful rate of predictions. Higher successful rate in distinguishing members from non-members suggest greater potential for privacy leakage, indicating that the synthetic data may be too closely replicating identifiable patterns present in the real data. This comprehensive scheme for privacy evaluation using MIA models thus plays a crucial role in our assessment of the security and applicability of synthetic datasets generated by state-of-the-art language models, guiding us in making informed decisions about their deployment in sensitive or privacy-conscious environments. 4 Experiments 4.1 Synthetic Data Generation Scenario In this work, our primary focus is on the generation of synthetic tabular data with text reviews. Tabular data encompasses a structured format that is pivotal in numerous analytical scenarios across various domains. The ability to accurately and efficiently generate synthetic tabular data not only aids in enhancing data privacy but also facilitates the testing and development of new analytical models where real data may be sparse or sensitive. We specifically concentrate on online review data for several reasons. First, such data is abundant and rich in diverse types of information, making it an ideal candidate for robust synthetic data generation experiments. Online reviews, as seen on platforms like Amazon, not only influence consumer behavior but also provide a wealth of information that can be mined for insights into user satisfaction and product quality. We use the Amazon product review dataset [40], which contains millions of product reviews from various categories, such as electronics, books, and clothing, as the source dataset. User review data includes various data fields, which are structured below. \u2022 Rating: Numerical rating of the product, ranging from 1.0 to 5.0, indicating the overall customer satisfaction. \u2022 Title: Brief title of the user review, summarizing the user\u2019s opinion. 5 \u2022 Text: Detailed text body of the user review, providing in-depth feedback and commentary on the product. \u2022 Images: Visual content uploaded by users post-purchase, showcasing the product from a consumer\u2019s perspective. \u2022 ASIN: The Amazon Standard Identification Number, a unique code assigned to each product listed on Amazon. \u2022 Parent ASIN: The identifier for the parent product, under which variations (such as size, color, etc.) are listed. \u2022 User ID: Identifier for the user who posted the review, ensuring traceability and authenticity. \u2022 Timestamp: The exact time when the review was posted, expressed in Unix time format. \u2022 Helpful Vote: Count of the number of times other users found the review helpful. \u2022 Verified Purchase: A boolean indicator showing whether the product was purchased through Amazon, confirming the authenticity of the review. Such user review data, which includes both non-text tabular data and text-based comment data, can well represent the complexity of the generated synthetic data. To generate synthetic review data, we select 50 samples from a dataset of software product reviews to train LLMs through prompting techniques. Each model is tasked with producing 300 synthetic entries. Claude demonstrates the most efficient data generation process, smoothly producing approximately 25 entries at a time without the need for re-prompting. In contrast, ChatGPT requires frequent reintroduction of the real data to maintain consistency in generating software related product reviews. Llama necessitates detailed prompting to accurately specify numeric data ranges and text formats; lacking precise instructions, it tends to shift focus to unrelated topics or generate data in incorrect formats. A uniqueness verification is conducted to ensure the originality of each synthetic entry. Out of the requested 300 entries, Claude manages to produce 300 unique items. ChatGPT produces 292 unique entries. Llama, however, only produces 115, with the remainder being duplicates. These duplicates are removed prior to further evaluation to maintain the integrity of our data analysis. 4.2 Models to evaluate In the experiment, we evaluate three advanced large language models (LLMs) to understand their performances in generating synthetic data. The models selected are Claude 3 Opus, ChatGPT 3.5, and Llama 2 13B, representing a mix of proprietary and open-source technologies, each with unique operational and accessibility characteristics. Claude 3 Opus is a proprietary model developed by Anthropic. It is notable for its advanced natural language understanding and generation capabilities. This model operates on a subscription basis, requiring a monthly payment for access. The decision to include Claude in this work stems from its representation of cutting-edge, commercially available closed-source LLMs. Evaluating Claude allows us to assess the performance of pay-to-access models in synthetic data generation, providing insights into the value offered by such commercially structured tools. ChatGPT 3.5, developed by OpenAI, is another proprietary model but is offered free of charge. It has gained significant attention for its robust performance in a variety of natural language processing tasks. The inclusion of ChatGPT 3.5 enables us to compare a no-cost proprietary model against its paid counterparts, offering a perspective on how accessibility influences the quality and utility of synthetic data generation, especially in academic and low-resource settings. Llama 2 13B, an open-source model developed by Meta, stands out due to its accessibility and the transparency of its development process. As an open-source LLM, Llama allows researchers complete access to tweak its parameters and training methods. Including Llama in our evaluation provides a comprehensive view of how open-source models stack up against proprietary models in generating synthetic data, which is crucial for understanding the broader ecosystem of data generation technologies. By examining these three models, we aim to uncover the nuances of synthetic data generation across different model architectures. This comparative analysis helps us delineate the performance differentials and potentially identify the optimal configurations and settings for each type of model in the domain of synthetic data generation. 4.3 Evaluation Results 4.3.1 Fidelity Table 1 presents the fidelity evaluation of the non-text tabular data within our synthetic datasets. All three models achieve a 100% score in structure preservation, demonstrating their ability to maintain the column names in the synthetic 6 datasets relative to the real data. In terms of data integrity, Claude scores the highest. ChatGPT and Llama receive lower scores due to their propensity to generate duplicate titles. Claude and ChatGPT exhibit higher scores for column shape, attributed to their capabilities of preserving the distribution shape of most columns. Their scores are affected by the predominance of zero values in the \u2019helpful vote\u2019 category, which deviates from the real data. Llama scores the lowest, not only displaying zeros in all \u2019helpful vote\u2019 fields but also inaccurately representing most \u2019verified purchase\u2019 values as false, contrary to the real data, which predominantly features verified purchases. Metric Claude ChatGPT Llama Structure Preserving Score 100% 100% 100% Data Integrity Score 98.4% 93.9% 87.59% Column Shapes Score 80.92% 80.97% 62.29% Table 1: Fidelity Evaluation Results on Non-Text Tabular Data Table 2 presents the text analysis results comparing three synthetic data sets against real data. All models successfully capture the predominant sentiment found within the review data. Claude not only aligns closely with the top three keywords and sentiment-related words of the real data but also closely mirrors the original reviews\u2019 writing style. Although the other two models produce somewhat similar keywords, the significant distinction lies in the average length of the reviews. Claude maintains an average review length most akin to that of the real data, while the other models generate significantly shorter reviews on average. Notably, ChatGPT initially produces lengthy reviews but the length decreases over time. Llama, on the other hand, struggles to generate extended content without losing coherence. Overall, Claude excels in preserving the underlying thematic and stylistic nuances characteristic of the real dataset. Metric Claude ChatGPT Llama Real Sentiment Distribution Positive (82%) Positive (78.42%) Positive (75.86%) Positive (82%) Top 3 Keywords app, features, like game, highly, app product, highly, app app, game, like Top 3 Sentiment Words great, more, easy reliable, addictive, flashlight great, more, much good, sure, easy Average length (words) 40.48 16.55 18.69 59.26 Table 2: Text Analysis Results 4.3.2 Utility Table 3 displays the utility evaluation results for synthetic data. While the sentiment classification model trained on real data achieves the highest accuracy, the models trained on synthetic data also exhibit comparable accuracy and similar Mean Absolute Error (MAE) values. This suggests that, despite varying levels of fidelity, all models effectively capture the relationship between review text data and user-provided ratings. Given that the training involved no more than 300 data samples, we anticipate that accuracy could further improve with the inclusion of more training data. Additionally, with an expanded dataset, the utility performance of each synthetic data might show greater divergence. Data Type MAE Accuracy Claude Synthetic 1.2929 67.68% ChatGPT Synthetic 1.2041 67.35% Llama Synthetic 1.4151 62.26% Real Data 1.3019 67.92% Table 3: Utility Evaluation Results 4.3.3 Privacy Table 4 presents the privacy evaluation results, showcasing the accuracy of the trained Membership Inference Attack (MIA) models [39]. All three models demonstrate very high successful rates, which in the context of MIA, suggests a higher probability of information leakage. This high successful rate is likely due to the fact that categorical features such 7 as product IDs, parent product IDs, and user IDs are largely duplicated within each synthetic dataset. This duplication enables the models to re-identify data successfully. These results imply that without additional fine-tuning, LLMs struggle to maintain the security and uniqueness of data in complex synthetic data generation tasks. Nonetheless, it is also conceivable that with more comprehensive datasets for training, these models could enhance their data generation capabilities, reducing the risk of privacy breaches. MIA Model Successful Rate Trained by Claude Synthetic 91% Trained by ChatGPT Synthetic 90% Trained by Llama Synthetic 83% Table 4: Privacy Evaluation Results 5 Conclusion and Future Work In this paper, we propose a comprehensive evaluation framework for quantitatively assessing the fidelity, utility, and privacy of synthetic tabular data generated by LLMs. We demonstrated the effectiveness of the proposed framework by applying it to synthetic product review data generated by three state-of-the-art LLMs: ChatGPT, Claude, and Llama. This framework can contribute to the growing field of synthetic data evaluation. As LLMs continue to advance and generate increasingly realistic synthetic data, the proposed framework can serve as a foundation for future research and help ensure the responsible and effective use of synthetic data in real-world applications. The proposed work also opens up opportunities for future research. First, while this study focuses on product review data, the evaluation framework can be extended to other domains and data types to assess. Second, the privacy evaluation component can be enhanced by incorporating additional privacy attacks and metrics to provide a more comprehensive assessment of privacy risks. Third, the trade-off between privacy and utility can be further explored by developing advanced privacy-preserving techniques that maintain high data utility while ensuring strong privacy guarantees."
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.09288v2",
+                    "title": "Llama 2: Open Foundation and Fine-Tuned Chat Models",
+                    "abstract": "In this work, we develop and release Llama 2, a collection of pretrained and\nfine-tuned large language models (LLMs) ranging in scale from 7 billion to 70\nbillion parameters. Our fine-tuned LLMs, called Llama 2-Chat, are optimized for\ndialogue use cases. Our models outperform open-source chat models on most\nbenchmarks we tested, and based on our human evaluations for helpfulness and\nsafety, may be a suitable substitute for closed-source models. We provide a\ndetailed description of our approach to fine-tuning and safety improvements of\nLlama 2-Chat in order to enable the community to build on our work and\ncontribute to the responsible development of LLMs.",
+                    "authors": "Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez, Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushkar Mishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing Ellen Tan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom",
+                    "published": "2023-07-18",
+                    "updated": "2023-07-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1907.00503v2",
+                    "title": "Modeling Tabular data using Conditional GAN",
+                    "abstract": "Modeling the probability distribution of rows in tabular data and generating\nrealistic synthetic data is a non-trivial task. Tabular data usually contains a\nmix of discrete and continuous columns. Continuous columns may have multiple\nmodes whereas discrete columns are sometimes imbalanced making the modeling\ndifficult. Existing statistical and deep neural network models fail to properly\nmodel this type of data. We design TGAN, which uses a conditional generative\nadversarial network to address these challenges. To aid in a fair and thorough\ncomparison, we design a benchmark with 7 simulated and 8 real datasets and\nseveral Bayesian network baselines. TGAN outperforms Bayesian methods on most\nof the real datasets whereas other deep learning methods could not.",
+                    "authors": "Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, Kalyan Veeramachaneni",
+                    "published": "2019-07-01",
+                    "updated": "2019-10-28",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "stat.ML"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1610.05820v2",
+                    "title": "Membership Inference Attacks against Machine Learning Models",
+                    "abstract": "We quantitatively investigate how machine learning models leak information\nabout the individual data records on which they were trained. We focus on the\nbasic membership inference attack: given a data record and black-box access to\na model, determine if the record was in the model's training dataset. To\nperform membership inference against a target model, we make adversarial use of\nmachine learning and train our own inference model to recognize differences in\nthe target model's predictions on the inputs that it trained on versus the\ninputs that it did not train on.\n  We empirically evaluate our inference techniques on classification models\ntrained by commercial \"machine learning as a service\" providers such as Google\nand Amazon. Using realistic datasets and classification tasks, including a\nhospital discharge dataset whose membership is sensitive from the privacy\nperspective, we show that these models can be vulnerable to membership\ninference attacks. We then investigate the factors that influence this leakage\nand evaluate mitigation strategies.",
+                    "authors": "Reza Shokri, Marco Stronati, Congzheng Song, Vitaly Shmatikov",
+                    "published": "2016-10-18",
+                    "updated": "2017-03-31",
+                    "primary_cat": "cs.CR",
+                    "cats": [
+                        "cs.CR",
+                        "cs.LG",
+                        "stat.ML"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.07849v2",
+                    "title": "Synthetic Data Generation with Large Language Models for Text Classification: Potential and Limitations",
+                    "abstract": "The collection and curation of high-quality training data is crucial for\ndeveloping text classification models with superior performance, but it is\noften associated with significant costs and time investment. Researchers have\nrecently explored using large language models (LLMs) to generate synthetic\ndatasets as an alternative approach. However, the effectiveness of the\nLLM-generated synthetic data in supporting model training is inconsistent\nacross different classification tasks. To better understand factors that\nmoderate the effectiveness of the LLM-generated synthetic data, in this study,\nwe look into how the performance of models trained on these synthetic data may\nvary with the subjectivity of classification. Our results indicate that\nsubjectivity, at both the task level and instance level, is negatively\nassociated with the performance of the model trained on synthetic data. We\nconclude by discussing the implications of our work on the potential and\nlimitations of leveraging LLM for synthetic data generation.",
+                    "authors": "Zhuoyan Li, Hangxiao Zhu, Zhuoran Lu, Ming Yin",
+                    "published": "2023-10-11",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2002.00269v3",
+                    "title": "A Tutorial on Learning With Bayesian Networks",
+                    "abstract": "A Bayesian network is a graphical model that encodes probabilistic\nrelationships among variables of interest. When used in conjunction with\nstatistical techniques, the graphical model has several advantages for data\nanalysis. One, because the model encodes dependencies among all variables, it\nreadily handles situations where some data entries are missing. Two, a Bayesian\nnetwork can be used to learn causal relationships, and hence can be used to\ngain understanding about a problem domain and to predict the consequences of\nintervention. Three, because the model has both a causal and probabilistic\nsemantics, it is an ideal representation for combining prior knowledge (which\noften comes in causal form) and data. Four, Bayesian statistical methods in\nconjunction with Bayesian networks offer an efficient and principled approach\nfor avoiding the overfitting of data. In this paper, we discuss methods for\nconstructing Bayesian networks from prior knowledge and summarize Bayesian\nstatistical methods for using data to improve these models. With regard to the\nlatter task, we describe methods for learning both the parameters and structure\nof a Bayesian network, including techniques for learning with incomplete data.\nIn addition, we relate Bayesian-network methods for learning to techniques for\nsupervised and unsupervised learning. We illustrate the graphical-modeling\napproach using a real-world case study.",
+                    "authors": "David Heckerman",
+                    "published": "2020-02-01",
+                    "updated": "2022-01-10",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "stat.ML",
+                        "I.2; G.3"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1706.03850v3",
+                    "title": "Adversarial Feature Matching for Text Generation",
+                    "abstract": "The Generative Adversarial Network (GAN) has achieved great success in\ngenerating realistic (real-valued) synthetic data. However, convergence issues\nand difficulties dealing with discrete data hinder the applicability of GAN to\ntext. We propose a framework for generating realistic text via adversarial\ntraining. We employ a long short-term memory network as generator, and a\nconvolutional network as discriminator. Instead of using the standard objective\nof GAN, we propose matching the high-dimensional latent feature distributions\nof real and synthetic sentences, via a kernelized discrepancy metric. This\neases adversarial training by alleviating the mode-collapsing problem. Our\nexperiments show superior performance in quantitative evaluation, and\ndemonstrate that our model can generate realistic-looking sentences.",
+                    "authors": "Yizhe Zhang, Zhe Gan, Kai Fan, Zhi Chen, Ricardo Henao, Dinghan Shen, Lawrence Carin",
+                    "published": "2017-06-12",
+                    "updated": "2017-11-18",
+                    "primary_cat": "stat.ML",
+                    "cats": [
+                        "stat.ML",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1505.03906v1",
+                    "title": "Training generative neural networks via Maximum Mean Discrepancy optimization",
+                    "abstract": "We consider training a deep neural network to generate samples from an\nunknown distribution given i.i.d. data. We frame learning as an optimization\nminimizing a two-sample test statistic---informally speaking, a good generator\nnetwork produces samples that cause a two-sample test to fail to reject the\nnull hypothesis. As our two-sample test statistic, we use an unbiased estimate\nof the maximum mean discrepancy, which is the centerpiece of the nonparametric\nkernel two-sample test proposed by Gretton et al. (2012). We compare to the\nadversarial nets framework introduced by Goodfellow et al. (2014), in which\nlearning is a two-player game between a generator network and an adversarial\ndiscriminator network, both trained to outwit the other. From this perspective,\nthe MMD statistic plays the role of the discriminator. In addition to empirical\ncomparisons, we prove bounds on the generalization error incurred by optimizing\nthe empirical MMD.",
+                    "authors": "Gintare Karolina Dziugaite, Daniel M. Roy, Zoubin Ghahramani",
+                    "published": "2015-05-14",
+                    "updated": "2015-05-14",
+                    "primary_cat": "stat.ML",
+                    "cats": [
+                        "stat.ML",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1703.10155v2",
+                    "title": "CVAE-GAN: Fine-Grained Image Generation through Asymmetric Training",
+                    "abstract": "We present variational generative adversarial networks, a general learning\nframework that combines a variational auto-encoder with a generative\nadversarial network, for synthesizing images in fine-grained categories, such\nas faces of a specific person or objects in a category. Our approach models an\nimage as a composition of label and latent attributes in a probabilistic model.\nBy varying the fine-grained category label fed into the resulting generative\nmodel, we can generate images in a specific category with randomly drawn values\non a latent attribute vector. Our approach has two novel aspects. First, we\nadopt a cross entropy loss for the discriminative and classifier network, but a\nmean discrepancy objective for the generative network. This kind of asymmetric\nloss function makes the GAN training more stable. Second, we adopt an encoder\nnetwork to learn the relationship between the latent space and the real image\nspace, and use pairwise feature matching to keep the structure of generated\nimages. We experiment with natural images of faces, flowers, and birds, and\ndemonstrate that the proposed models are capable of generating realistic and\ndiverse samples with fine-grained category labels. We further show that our\nmodels can be applied to other tasks, such as image inpainting,\nsuper-resolution, and data augmentation for training better face recognition\nmodels.",
+                    "authors": "Jianmin Bao, Dong Chen, Fang Wen, Houqiang Li, Gang Hua",
+                    "published": "2017-03-29",
+                    "updated": "2017-10-12",
+                    "primary_cat": "cs.CV",
+                    "cats": [
+                        "cs.CV"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.04709v1",
+                    "title": "A Comparative Study of Open-Source Large Language Models, GPT-4 and Claude 2: Multiple-Choice Test Taking in Nephrology",
+                    "abstract": "In recent years, there have been significant breakthroughs in the field of\nnatural language processing, particularly with the development of large\nlanguage models (LLMs). These LLMs have showcased remarkable capabilities on\nvarious benchmarks. In the healthcare field, the exact role LLMs and other\nfuture AI models will play remains unclear. There is a potential for these\nmodels in the future to be used as part of adaptive physician training, medical\nco-pilot applications, and digital patient interaction scenarios. The ability\nof AI models to participate in medical training and patient care will depend in\npart on their mastery of the knowledge content of specific medical fields. This\nstudy investigated the medical knowledge capability of LLMs, specifically in\nthe context of internal medicine subspecialty multiple-choice test-taking\nability. We compared the performance of several open-source LLMs (Koala 7B,\nFalcon 7B, Stable-Vicuna 13B, and Orca Mini 13B), to GPT-4 and Claude 2 on\nmultiple-choice questions in the field of Nephrology. Nephrology was chosen as\nan example of a particularly conceptually complex subspecialty field within\ninternal medicine. The study was conducted to evaluate the ability of LLM\nmodels to provide correct answers to nephSAP (Nephrology Self-Assessment\nProgram) multiple-choice questions. The overall success of open-sourced LLMs in\nanswering the 858 nephSAP multiple-choice questions correctly was 17.1% -\n25.5%. In contrast, Claude 2 answered 54.4% of the questions correctly, whereas\nGPT-4 achieved a score of 73.3%. We show that current widely used open-sourced\nLLMs do poorly in their ability for zero-shot reasoning when compared to GPT-4\nand Claude 2. The findings of this study potentially have significant\nimplications for the future of subspecialty medical training and patient care.",
+                    "authors": "Sean Wu, Michael Koo, Lesley Blum, Andy Black, Liyo Kao, Fabien Scalzo, Ira Kurtz",
+                    "published": "2023-08-09",
+                    "updated": "2023-08-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1604.06651v2",
+                    "title": "General and specific utility measures for synthetic data",
+                    "abstract": "Data holders can produce synthetic versions of datasets when concerns about\npotential disclosure restrict the availability of the original records. This\npaper is concerned with methods to judge whether such synthetic data have a\ndistribution that is comparable to that of the original data, what we will term\ngeneral utility. We consider how general utility compares with specific\nutility, the similarity of results of analyses from the synthetic data and the\noriginal data. We adapt a previous general measure of data utility, the\npropensity score mean-squared-error (pMSE), to the specific case of synthetic\ndata and derive its distribution for the case when the correct synthesis model\nis used to create the synthetic data. Our asymptotic results are confirmed by a\nsimulation study. We also consider two specific utility measures, confidence\ninterval overlap and standardized difference in summary statistics, which we\ncompare with the general utility results. We present two examples examining\nthis comparison of general and specific utility to real data syntheses and make\nrecommendations for their use for evaluating synthetic data.",
+                    "authors": "Joshua Snoke, Gillian Raab, Beata Nowok, Chris Dibben, Aleksandra Slavkovic",
+                    "published": "2016-04-22",
+                    "updated": "2017-06-19",
+                    "primary_cat": "stat.AP",
+                    "cats": [
+                        "stat.AP"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.01262v2",
+                    "title": "Fairness Certification for Natural Language Processing and Large Language Models",
+                    "abstract": "Natural Language Processing (NLP) plays an important role in our daily lives,\nparticularly due to the enormous progress of Large Language Models (LLM).\nHowever, NLP has many fairness-critical use cases, e.g., as an expert system in\nrecruitment or as an LLM-based tutor in education. Since NLP is based on human\nlanguage, potentially harmful biases can diffuse into NLP systems and produce\nunfair results, discriminate against minorities or generate legal issues.\nHence, it is important to develop a fairness certification for NLP approaches.\nWe follow a qualitative research approach towards a fairness certification for\nNLP. In particular, we have reviewed a large body of literature on algorithmic\nfairness, and we have conducted semi-structured expert interviews with a wide\nrange of experts from that area. We have systematically devised six fairness\ncriteria for NLP, which can be further refined into 18 sub-categories. Our\ncriteria offer a foundation for operationalizing and testing processes to\ncertify fairness, both from the perspective of the auditor and the audited\norganization.",
+                    "authors": "Vincent Freiberger, Erik Buchmann",
+                    "published": "2024-01-02",
+                    "updated": "2024-01-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.07884v2",
+                    "title": "Fair Abstractive Summarization of Diverse Perspectives",
+                    "abstract": "People from different social and demographic groups express diverse\nperspectives and conflicting opinions on a broad set of topics such as product\nreviews, healthcare, law, and politics. A fair summary should provide a\ncomprehensive coverage of diverse perspectives without underrepresenting\ncertain groups. However, current work in summarization metrics and Large\nLanguage Models (LLMs) evaluation has not explored fair abstractive\nsummarization. In this paper, we systematically investigate fair abstractive\nsummarization for user-generated data. We first formally define fairness in\nabstractive summarization as not underrepresenting perspectives of any groups\nof people, and we propose four reference-free automatic metrics by measuring\nthe differences between target and source perspectives. We evaluate nine LLMs,\nincluding three GPT models, four LLaMA models, PaLM 2, and Claude, on six\ndatasets collected from social media, online reviews, and recorded transcripts.\nExperiments show that both the model-generated and the human-written reference\nsummaries suffer from low fairness. We conduct a comprehensive analysis of the\ncommon factors influencing fairness and propose three simple but effective\nmethods to alleviate unfair summarization. Our dataset and code are available\nat https://github.com/psunlpgroup/FairSumm.",
+                    "authors": "Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang",
+                    "published": "2023-11-14",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06003v1",
+                    "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+                    "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+                    "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.10199v3",
+                    "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+                    "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+                    "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+                    "published": "2024-04-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.18569v1",
+                    "title": "Fairness of ChatGPT",
+                    "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+                    "authors": "Yunqi Li, Yongfeng Zhang",
+                    "published": "2023-05-22",
+                    "updated": "2023-05-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.08836v2",
+                    "title": "Bias and Fairness in Chatbots: An Overview",
+                    "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+                    "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+                    "published": "2023-09-16",
+                    "updated": "2023-12-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.07688v1",
+                    "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+                    "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+                    "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+                    "published": "2024-02-12",
+                    "updated": "2024-02-12",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.06500v1",
+                    "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+                    "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+                    "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+                    "published": "2023-10-10",
+                    "updated": "2023-10-10",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.13840v1",
+                    "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+                    "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+                    "published": "2024-03-15",
+                    "updated": "2024-03-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.SI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.08656v1",
+                    "title": "Linear Cross-document Event Coreference Resolution with X-AMR",
+                    "abstract": "Event Coreference Resolution (ECR) as a pairwise mention classification task\nis expensive both for automated systems and manual annotations. The task's\nquadratic difficulty is exacerbated when using Large Language Models (LLMs),\nmaking prompt engineering for ECR prohibitively costly. In this work, we\npropose a graphical representation of events, X-AMR, anchored around individual\nmentions using a \\textbf{cross}-document version of \\textbf{A}bstract\n\\textbf{M}eaning \\textbf{R}epresentation. We then linearize the ECR with a\nnovel multi-hop coreference algorithm over the event graphs. The event graphs\nsimplify ECR, making it a) LLM cost-effective, b) compositional and\ninterpretable, and c) easily annotated. For a fair assessment, we first enrich\nan existing ECR benchmark dataset with these event graphs using an\nannotator-friendly tool we introduce. Then, we employ GPT-4, the newest LLM by\nOpenAI, for these annotations. Finally, using the ECR algorithm, we assess\nGPT-4 against humans and analyze its limitations. Through this research, we aim\nto advance the state-of-the-art for efficient ECR and shed light on the\npotential shortcomings of current LLMs at this task. Code and annotations:\n\\url{https://github.com/ahmeshaf/gpt_coref}",
+                    "authors": "Shafiuddin Rehan Ahmed, George Arthur Baker, Evi Judge, Michael Regan, Kristin Wright-Bettner, Martha Palmer, James H. Martin",
+                    "published": "2024-03-25",
+                    "updated": "2024-03-25",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15198v2",
+                    "title": "Do LLM Agents Exhibit Social Behavior?",
+                    "abstract": "The advances of Large Language Models (LLMs) are expanding their utility in\nboth academic research and practical applications. Recent social science\nresearch has explored the use of these ``black-box'' LLM agents for simulating\ncomplex social systems and potentially substituting human subjects in\nexperiments. Our study delves into this emerging domain, investigating the\nextent to which LLMs exhibit key social interaction principles, such as social\nlearning, social preference, and cooperative behavior (indirect reciprocity),\nin their interactions with humans and other agents. We develop a framework for\nour study, wherein classical laboratory experiments involving human subjects\nare adapted to use LLM agents. This approach involves step-by-step reasoning\nthat mirrors human cognitive processes and zero-shot learning to assess the\ninnate preferences of LLMs. Our analysis of LLM agents' behavior includes both\nthe primary effects and an in-depth examination of the underlying mechanisms.\nFocusing on GPT-4, our analyses suggest that LLM agents appear to exhibit a\nrange of human-like social behaviors such as distributional and reciprocity\npreferences, responsiveness to group identity cues, engagement in indirect\nreciprocity, and social learning capabilities. However, our analysis also\nreveals notable differences: LLMs demonstrate a pronounced fairness preference,\nweaker positive reciprocity, and a more calculating approach in social learning\ncompared to humans. These insights indicate that while LLMs hold great promise\nfor applications in social science research, such as in laboratory experiments\nand agent-based modeling, the subtle behavioral differences between LLM agents\nand humans warrant further investigation. Careful examination and development\nof protocols in evaluating the social behaviors of LLMs are necessary before\ndirectly applying these models to emulate human behavior.",
+                    "authors": "Yan Leng, Yuan Yuan",
+                    "published": "2023-12-23",
+                    "updated": "2024-02-22",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.SI",
+                        "econ.GN",
+                        "q-fin.EC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2206.13757v1",
+                    "title": "Flexible text generation for counterfactual fairness probing",
+                    "abstract": "A common approach for testing fairness issues in text-based classifiers is\nthrough the use of counterfactuals: does the classifier output change if a\nsensitive attribute in the input is changed? Existing counterfactual generation\nmethods typically rely on wordlists or templates, producing simple\ncounterfactuals that don't take into account grammar, context, or subtle\nsensitive attribute references, and could miss issues that the wordlist\ncreators had not considered. In this paper, we introduce a task for generating\ncounterfactuals that overcomes these shortcomings, and demonstrate how large\nlanguage models (LLMs) can be leveraged to make progress on this task. We show\nthat this LLM-based method can produce complex counterfactuals that existing\nmethods cannot, comparing the performance of various counterfactual generation\nmethods on the Civil Comments dataset and showing their value in evaluating a\ntoxicity classifier.",
+                    "authors": "Zee Fryer, Vera Axelrod, Ben Packer, Alex Beutel, Jilin Chen, Kellie Webster",
+                    "published": "2022-06-28",
+                    "updated": "2022-06-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11764v1",
+                    "title": "ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs",
+                    "abstract": "Large Language models (LLMs), while powerful, exhibit harmful social biases.\nDebiasing is often challenging due to computational costs, data constraints,\nand potential degradation of multi-task language capabilities. This work\nintroduces a novel approach utilizing ChatGPT to generate synthetic training\ndata, aiming to enhance the debiasing of LLMs. We propose two strategies:\nTargeted Prompting, which provides effective debiasing for known biases but\nnecessitates prior specification of bias in question; and General Prompting,\nwhich, while slightly less effective, offers debiasing across various\ncategories. We leverage resource-efficient LLM debiasing using adapter tuning\nand compare the effectiveness of our synthetic data to existing debiasing\ndatasets. Our results reveal that: (1) ChatGPT can efficiently produce\nhigh-quality training data for debiasing other LLMs; (2) data produced via our\napproach surpasses existing datasets in debiasing performance while also\npreserving internal knowledge of a pre-trained LLM; and (3) synthetic data\nexhibits generalizability across categories, effectively mitigating various\nbiases, including intersectional ones. These findings underscore the potential\nof synthetic data in advancing the fairness of LLMs with minimal retraining\ncost.",
+                    "authors": "Pengrui Han, Rafal Kocielnik, Adhithya Saravanan, Roy Jiang, Or Sharir, Anima Anandkumar",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "68T50",
+                        "I.2.7; K.4.1"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.04814v2",
+                    "title": "Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks",
+                    "abstract": "We introduce Syntax-Aware Fill-In-the-Middle (SAFIM), a new benchmark for\nevaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM)\ntask. This benchmark focuses on syntax-aware completions of program structures\nsuch as code blocks and conditional expressions, and includes 17,720 examples\nfrom multiple programming languages, sourced from recent code submissions after\nApril 2022 to minimize data contamination. SAFIM provides a robust framework\nwith various prompt designs and novel syntax-aware post-processing techniques,\nfacilitating accurate and fair comparisons across LLMs. Our comprehensive\nevaluation of 15 LLMs shows that FIM pretraining not only enhances FIM\nproficiency but also improves Left-to-Right (L2R) inference using LLMs. Our\nfindings challenge conventional beliefs and suggest that pretraining methods\nand data quality have more impact than model size. SAFIM thus serves as a\nfoundational platform for future research in effective pretraining strategies\nfor code LLMs. The evaluation toolkit and dataset are available at\nhttps://github.com/gonglinyuan/safim, and the leaderboard is available at\nhttps://safimbenchmark.com.",
+                    "authors": "Linyuan Gong, Sida Wang, Mostafa Elhoushi, Alvin Cheung",
+                    "published": "2024-03-07",
+                    "updated": "2024-04-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.15997v1",
+                    "title": "RoCar: A Relationship Network-based Evaluation Method to Large Language Models",
+                    "abstract": "Large language models (LLMs) have received increasing attention. However, due\nto the complexity of its capabilities, how to rationally evaluate the\ncapabilities of LLMs is still a task to be solved. We propose the RoCar method,\nwhich utilizes the defined basic schemas to randomly construct a task graph and\ngenerates natural language evaluation tasks based on the task graph to evaluate\nthe reasoning and memory abilities of LLMs respectively. Due to the very large\nrandomness of the task construction process, it is possible to ensure that none\nof the LLMs to be tested has directly learned the evaluation tasks,\nguaranteeing the fairness of the evaluation method.",
+                    "authors": "Ming Wang, Wenfang Wu, Chongyun Gao, Daling Wang, Shi Feng, Yifei Zhang",
+                    "published": "2023-07-29",
+                    "updated": "2023-07-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.07981v1",
+                    "title": "Manipulating Large Language Models to Increase Product Visibility",
+                    "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+                    "authors": "Aounon Kumar, Himabindu Lakkaraju",
+                    "published": "2024-04-11",
+                    "updated": "2024-04-11",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.01937v1",
+                    "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+                    "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+                    "authors": "Cheng-Han Chiang, Hung-yi Lee",
+                    "published": "2023-05-03",
+                    "updated": "2023-05-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.03192v1",
+                    "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+                    "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+                    "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+                    "published": "2024-04-04",
+                    "updated": "2024-04-04",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.17553v1",
+                    "title": "RuBia: A Russian Language Bias Detection Dataset",
+                    "abstract": "Warning: this work contains upsetting or disturbing content.\n  Large language models (LLMs) tend to learn the social and cultural biases\npresent in the raw pre-training data. To test if an LLM's behavior is fair,\nfunctional datasets are employed, and due to their purpose, these datasets are\nhighly language and culture-specific. In this paper, we address a gap in the\nscope of multilingual bias evaluation by presenting a bias detection dataset\nspecifically designed for the Russian language, dubbed as RuBia. The RuBia\ndataset is divided into 4 domains: gender, nationality, socio-economic status,\nand diverse, each of the domains is further divided into multiple fine-grained\nsubdomains. Every example in the dataset consists of two sentences with the\nfirst reinforcing a potentially harmful stereotype or trope and the second\ncontradicting it. These sentence pairs were first written by volunteers and\nthen validated by native-speaking crowdsourcing workers. Overall, there are\nnearly 2,000 unique sentence pairs spread over 19 subdomains in RuBia. To\nillustrate the dataset's purpose, we conduct a diagnostic evaluation of\nstate-of-the-art or near-state-of-the-art LLMs and discuss the LLMs'\npredisposition to social biases.",
+                    "authors": "Veronika Grigoreva, Anastasiia Ivanova, Ilseyar Alimova, Ekaterina Artemova",
+                    "published": "2024-03-26",
+                    "updated": "2024-03-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.06852v2",
+                    "title": "ChemLLM: A Chemical Large Language Model",
+                    "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+                    "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+                    "published": "2024-02-10",
+                    "updated": "2024-04-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02294v1",
+                    "title": "LLMs grasp morality in concept",
+                    "abstract": "Work in AI ethics and fairness has made much progress in regulating LLMs to\nreflect certain values, such as fairness, truth, and diversity. However, it has\ntaken the problem of how LLMs might 'mean' anything at all for granted. Without\naddressing this, it is not clear what imbuing LLMs with such values even means.\nIn response, we provide a general theory of meaning that extends beyond humans.\nWe use this theory to explicate the precise nature of LLMs as meaning-agents.\nWe suggest that the LLM, by virtue of its position as a meaning-agent, already\ngrasps the constructions of human society (e.g. morality, gender, and race) in\nconcept. Consequently, under certain ethical frameworks, currently popular\nmethods for model alignment are limited at best and counterproductive at worst.\nMoreover, unaligned models may help us better develop our moral and social\nphilosophy.",
+                    "authors": "Mark Pock, Andre Ye, Jared Moore",
+                    "published": "2023-11-04",
+                    "updated": "2023-11-04",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.08780v1",
+                    "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+                    "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+                    "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+                    "published": "2023-10-13",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.04057v1",
+                    "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+                    "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+                    "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+                    "published": "2024-01-08",
+                    "updated": "2024-01-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.13095v1",
+                    "title": "Enhancing Logical Reasoning in Large Language Models to Facilitate Legal Applications",
+                    "abstract": "Language serves as a vehicle for conveying thought, enabling communication\namong individuals. The ability to distinguish between diverse concepts,\nidentify fairness and injustice, and comprehend a range of legal notions\nfundamentally relies on logical reasoning. Large Language Models (LLMs) attempt\nto emulate human language understanding and generation, but their competency in\nlogical reasoning remains limited. This paper seeks to address the\nphilosophical question: How can we effectively teach logical reasoning to LLMs\nwhile maintaining a deep understanding of the intricate relationship between\nlanguage and logic? By focusing on bolstering LLMs' capabilities in logical\nreasoning, we aim to expand their applicability in law and other\nlogic-intensive disciplines. To this end, we propose a Reinforcement Learning\nfrom Logical Feedback (RLLF) approach, which serves as a potential framework\nfor refining LLMs' reasoning capacities. Through RLLF and a revised evaluation\nmethodology, we explore new avenues for research in this domain and contribute\nto the development of LLMs capable of handling complex legal reasoning tasks\nwhile acknowledging the fundamental connection between language and logic.",
+                    "authors": "Ha-Thanh Nguyen, Wachara Fungwacharakorn, Ken Satoh",
+                    "published": "2023-11-22",
+                    "updated": "2023-11-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.08189v1",
+                    "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+                    "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+                    "authors": "Karthik Sreedhar, Lydia Chilton",
+                    "published": "2024-02-13",
+                    "updated": "2024-02-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.02839v1",
+                    "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+                    "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+                    "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+                    "published": "2024-03-05",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.06056v1",
+                    "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+                    "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+                    "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+                    "published": "2023-12-11",
+                    "updated": "2023-12-11",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10397v2",
+                    "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+                    "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+                    "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+                    "published": "2023-08-21",
+                    "updated": "2023-10-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.01769v1",
+                    "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+                    "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+                    "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+                    "published": "2024-05-02",
+                    "updated": "2024-05-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.03838v2",
+                    "title": "RADAR: Robust AI-Text Detection via Adversarial Learning",
+                    "abstract": "Recent advances in large language models (LLMs) and the intensifying\npopularity of ChatGPT-like applications have blurred the boundary of\nhigh-quality text generation between humans and machines. However, in addition\nto the anticipated revolutionary changes to our technology and society, the\ndifficulty of distinguishing LLM-generated texts (AI-text) from human-generated\ntexts poses new challenges of misuse and fairness, such as fake content\ngeneration, plagiarism, and false accusations of innocent writers. While\nexisting works show that current AI-text detectors are not robust to LLM-based\nparaphrasing, this paper aims to bridge this gap by proposing a new framework\ncalled RADAR, which jointly trains a robust AI-text detector via adversarial\nlearning. RADAR is based on adversarial training of a paraphraser and a\ndetector. The paraphraser's goal is to generate realistic content to evade\nAI-text detection. RADAR uses the feedback from the detector to update the\nparaphraser, and vice versa. Evaluated with 8 different LLMs (Pythia, Dolly\n2.0, Palmyra, Camel, GPT-J, Dolly 1.0, LLaMA, and Vicuna) across 4 datasets,\nexperimental results show that RADAR significantly outperforms existing AI-text\ndetection methods, especially when paraphrasing is in place. We also identify\nthe strong transferability of RADAR from instruction-tuned LLMs to other LLMs,\nand evaluate the improved capability of RADAR via GPT-3.5-Turbo.",
+                    "authors": "Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho",
+                    "published": "2023-07-07",
+                    "updated": "2023-10-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.10567v3",
+                    "title": "InSaAF: Incorporating Safety through Accuracy and Fairness | Are LLMs ready for the Indian Legal Domain?",
+                    "abstract": "Recent advancements in language technology and Artificial Intelligence have\nresulted in numerous Language Models being proposed to perform various tasks in\nthe legal domain ranging from predicting judgments to generating summaries.\nDespite their immense potential, these models have been proven to learn and\nexhibit societal biases and make unfair predictions. In this study, we explore\nthe ability of Large Language Models (LLMs) to perform legal tasks in the\nIndian landscape when social factors are involved. We present a novel metric,\n$\\beta$-weighted $\\textit{Legal Safety Score ($LSS_{\\beta}$)}$, which\nencapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs'\nsafety by considering its performance in the $\\textit{Binary Statutory\nReasoning}$ task and its fairness exhibition with respect to various axes of\ndisparities in the Indian society. Task performance and fairness scores of\nLLaMA and LLaMA--2 models indicate that the proposed $LSS_{\\beta}$ metric can\neffectively determine the readiness of a model for safe usage in the legal\nsector. We also propose finetuning pipelines, utilising specialised legal\ndatasets, as a potential method to mitigate bias and improve model safety. The\nfinetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\\beta}$,\nimproving their usability in the Indian legal domain. Our code is publicly\nreleased.",
+                    "authors": "Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru",
+                    "published": "2024-02-16",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.02650v1",
+                    "title": "Towards detecting unanticipated bias in Large Language Models",
+                    "abstract": "Over the last year, Large Language Models (LLMs) like ChatGPT have become\nwidely available and have exhibited fairness issues similar to those in\nprevious machine learning systems. Current research is primarily focused on\nanalyzing and quantifying these biases in training data and their impact on the\ndecisions of these models, alongside developing mitigation strategies. This\nresearch largely targets well-known biases related to gender, race, ethnicity,\nand language. However, it is clear that LLMs are also affected by other, less\nobvious implicit biases. The complex and often opaque nature of these models\nmakes detecting such biases challenging, yet this is crucial due to their\npotential negative impact in various applications. In this paper, we explore\nnew avenues for detecting these unanticipated biases in LLMs, focusing\nspecifically on Uncertainty Quantification and Explainable AI methods. These\napproaches aim to assess the certainty of model decisions and to make the\ninternal decision-making processes of LLMs more transparent, thereby\nidentifying and understanding biases that are not immediately apparent. Through\nthis research, we aim to contribute to the development of fairer and more\ntransparent AI systems.",
+                    "authors": "Anna Kruspe",
+                    "published": "2024-04-03",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02049v1",
+                    "title": "Post Turing: Mapping the landscape of LLM Evaluation",
+                    "abstract": "In the rapidly evolving landscape of Large Language Models (LLMs),\nintroduction of well-defined and standardized evaluation methodologies remains\na crucial challenge. This paper traces the historical trajectory of LLM\nevaluations, from the foundational questions posed by Alan Turing to the modern\nera of AI research. We categorize the evolution of LLMs into distinct periods,\neach characterized by its unique benchmarks and evaluation criteria. As LLMs\nincreasingly mimic human-like behaviors, traditional evaluation proxies, such\nas the Turing test, have become less reliable. We emphasize the pressing need\nfor a unified evaluation system, given the broader societal implications of\nthese models. Through an analysis of common evaluation methodologies, we\nadvocate for a qualitative shift in assessment approaches, underscoring the\nimportance of standardization and objective criteria. This work serves as a\ncall for the AI community to collaboratively address the challenges of LLM\nevaluation, ensuring their reliability, fairness, and societal benefit.",
+                    "authors": "Alexey Tikhonov, Ivan P. Yamshchikov",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.15007v1",
+                    "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+                    "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+                    "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+                    "published": "2023-10-23",
+                    "updated": "2023-10-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.14607v2",
+                    "title": "Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications",
+                    "abstract": "Recent literature has suggested the potential of using large language models\n(LLMs) to make classifications for tabular tasks. However, LLMs have been shown\nto exhibit harmful social biases that reflect the stereotypes and inequalities\npresent in society. To this end, as well as the widespread use of tabular data\nin many high-stake applications, it is important to explore the following\nquestions: what sources of information do LLMs draw upon when making\nclassifications for tabular tasks; whether and to what extent are LLM\nclassifications for tabular data influenced by social biases and stereotypes;\nand what are the consequential implications for fairness?\n  Through a series of experiments, we delve into these questions and show that\nLLMs tend to inherit social biases from their training data which significantly\nimpact their fairness in tabular classification tasks. Furthermore, our\ninvestigations show that in the context of bias mitigation, though in-context\nlearning and finetuning have a moderate effect, the fairness metric gap between\ndifferent subgroups is still larger than that in traditional machine learning\nmodels, such as Random Forest and shallow Neural Networks. This observation\nemphasizes that the social biases are inherent within the LLMs themselves and\ninherited from their pretraining corpus, not only from the downstream task\ndatasets. Besides, we demonstrate that label-flipping of in-context examples\ncan significantly reduce biases, further highlighting the presence of inherent\nbias within LLMs.",
+                    "authors": "Yanchen Liu, Srishti Gautam, Jiaqi Ma, Himabindu Lakkaraju",
+                    "published": "2023-10-23",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15478v1",
+                    "title": "A Group Fairness Lens for Large Language Models",
+                    "abstract": "The rapid advancement of large language models has revolutionized various\napplications but also raised crucial concerns about their potential to\nperpetuate biases and unfairness when deployed in social media contexts.\nEvaluating LLMs' potential biases and fairness has become crucial, as existing\nmethods rely on limited prompts focusing on just a few groups, lacking a\ncomprehensive categorical perspective. In this paper, we propose evaluating LLM\nbiases from a group fairness lens using a novel hierarchical schema\ncharacterizing diverse social groups. Specifically, we construct a dataset,\nGFair, encapsulating target-attribute combinations across multiple dimensions.\nIn addition, we introduce statement organization, a new open-ended text\ngeneration task, to uncover complex biases in LLMs. Extensive evaluations of\npopular LLMs reveal inherent safety concerns. To mitigate the biases of LLM\nfrom a group fairness perspective, we pioneer a novel chain-of-thought method\nGF-Think to mitigate biases of LLMs from a group fairness perspective.\nExperimental results demonstrate its efficacy in mitigating bias in LLMs to\nachieve fairness.",
+                    "authors": "Guanqun Bi, Lei Shen, Yuqiang Xie, Yanan Cao, Tiangang Zhu, Xiaodong He",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15491v1",
+                    "title": "Open Source Conversational LLMs do not know most Spanish words",
+                    "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+                    "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.02680v1",
+                    "title": "Large Language Models are Geographically Biased",
+                    "abstract": "Large Language Models (LLMs) inherently carry the biases contained in their\ntraining corpora, which can lead to the perpetuation of societal harm. As the\nimpact of these foundation models grows, understanding and evaluating their\nbiases becomes crucial to achieving fairness and accuracy. We propose to study\nwhat LLMs know about the world we live in through the lens of geography. This\napproach is particularly powerful as there is ground truth for the numerous\naspects of human life that are meaningfully projected onto geographic space\nsuch as culture, race, language, politics, and religion. We show various\nproblematic geographic biases, which we define as systemic errors in geospatial\npredictions. Initially, we demonstrate that LLMs are capable of making accurate\nzero-shot geospatial predictions in the form of ratings that show strong\nmonotonic correlation with ground truth (Spearman's $\\rho$ of up to 0.89). We\nthen show that LLMs exhibit common biases across a range of objective and\nsubjective topics. In particular, LLMs are clearly biased against locations\nwith lower socioeconomic conditions (e.g. most of Africa) on a variety of\nsensitive subjective topics such as attractiveness, morality, and intelligence\n(Spearman's $\\rho$ of up to 0.70). Finally, we introduce a bias score to\nquantify this and find that there is significant variation in the magnitude of\nbias across existing LLMs.",
+                    "authors": "Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon",
+                    "published": "2024-02-05",
+                    "updated": "2024-02-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13862v2",
+                    "title": "A Trip Towards Fairness: Bias and De-Biasing in Large Language Models",
+                    "abstract": "Cheap-to-Build Very Large-Language Models (CtB-LLMs) with affordable training\nare emerging as the next big revolution in natural language processing and\nunderstanding. These CtB-LLMs are democratizing access to trainable Very\nLarge-Language Models (VLLMs) and, thus, may represent the building blocks of\nmany NLP systems solving downstream tasks. Hence, a little or a large bias in\nCtB-LLMs may cause huge harm. In this paper, we performed a large investigation\nof the bias of three families of CtB-LLMs, and we showed that debiasing\ntechniques are effective and usable. Indeed, according to current tests, the\nLLaMA and the OPT families have an important bias in gender, race, religion,\nand profession. In contrast to the analysis for other LLMs, we discovered that\nbias depends not on the number of parameters but on the perplexity. Finally,\nthe debiasing of OPT using LoRA reduces bias up to 4.12 points in the\nnormalized stereotype score.",
+                    "authors": "Leonardo Ranaldi, Elena Sofia Ruzzetti, Davide Venditti, Dario Onorati, Fabio Massimo Zanzotto",
+                    "published": "2023-05-23",
+                    "updated": "2023-08-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.13925v1",
+                    "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+                    "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+                    "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18333v3",
+                    "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+                    "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+                    "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+                    "published": "2023-10-20",
+                    "updated": "2023-12-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09606v1",
+                    "title": "Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey",
+                    "abstract": "Causal inference has shown potential in enhancing the predictive accuracy,\nfairness, robustness, and explainability of Natural Language Processing (NLP)\nmodels by capturing causal relationships among variables. The emergence of\ngenerative Large Language Models (LLMs) has significantly impacted various NLP\ndomains, particularly through their advanced reasoning capabilities. This\nsurvey focuses on evaluating and improving LLMs from a causal view in the\nfollowing areas: understanding and improving the LLMs' reasoning capacity,\naddressing fairness and safety issues in LLMs, complementing LLMs with\nexplanations, and handling multimodality. Meanwhile, LLMs' strong reasoning\ncapacities can in turn contribute to the field of causal inference by aiding\ncausal relationship discovery and causal effect estimations. This review\nexplores the interplay between causal inference frameworks and LLMs from both\nperspectives, emphasizing their collective potential to further the development\nof more advanced and equitable artificial intelligence systems.",
+                    "authors": "Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.05694v1",
+                    "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+                    "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+                    "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+                    "published": "2023-10-09",
+                    "updated": "2023-10-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05374v2",
+                    "title": "Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment",
+                    "abstract": "Ensuring alignment, which refers to making models behave in accordance with\nhuman intentions [1,2], has become a critical task before deploying large\nlanguage models (LLMs) in real-world applications. For instance, OpenAI devoted\nsix months to iteratively aligning GPT-4 before its release [3]. However, a\nmajor challenge faced by practitioners is the lack of clear guidance on\nevaluating whether LLM outputs align with social norms, values, and\nregulations. This obstacle hinders systematic iteration and deployment of LLMs.\nTo address this issue, this paper presents a comprehensive survey of key\ndimensions that are crucial to consider when assessing LLM trustworthiness. The\nsurvey covers seven major categories of LLM trustworthiness: reliability,\nsafety, fairness, resistance to misuse, explainability and reasoning, adherence\nto social norms, and robustness. Each major category is further divided into\nseveral sub-categories, resulting in a total of 29 sub-categories.\nAdditionally, a subset of 8 sub-categories is selected for further\ninvestigation, where corresponding measurement studies are designed and\nconducted on several widely-used LLMs. The measurement results indicate that,\nin general, more aligned models tend to perform better in terms of overall\ntrustworthiness. However, the effectiveness of alignment varies across the\ndifferent trustworthiness categories considered. This highlights the importance\nof conducting more fine-grained analyses, testing, and making continuous\nimprovements on LLM alignment. By shedding light on these key dimensions of LLM\ntrustworthiness, this paper aims to provide valuable insights and guidance to\npractitioners in the field. Understanding and addressing these concerns will be\ncrucial in achieving reliable and ethically sound deployment of LLMs in various\napplications.",
+                    "authors": "Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, Hang Li",
+                    "published": "2023-08-10",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.17916v2",
+                    "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+                    "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+                    "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+                    "published": "2024-02-27",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04892v2",
+                    "title": "Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs",
+                    "abstract": "Recent works have showcased the ability of LLMs to embody diverse personas in\ntheir responses, exemplified by prompts like 'You are Yoda. Explain the Theory\nof Relativity.' While this ability allows personalization of LLMs and enables\nhuman behavior simulation, its effect on LLMs' capabilities remains unclear. To\nfill this gap, we present the first extensive study of the unintended\nside-effects of persona assignment on the ability of LLMs to perform basic\nreasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse\npersonas (e.g. an Asian person) spanning 5 socio-demographic groups. Our\nexperiments unveil that LLMs harbor deep rooted bias against various\nsocio-demographics underneath a veneer of fairness. While they overtly reject\nstereotypes when explicitly asked ('Are Black people less skilled at\nmathematics?'), they manifest stereotypical and erroneous presumptions when\nasked to answer questions while adopting a persona. These can be observed as\nabstentions in responses, e.g., 'As a Black person, I can't answer this\nquestion as it requires math knowledge', and generally result in a substantial\nperformance drop. Our experiments with ChatGPT-3.5 show that this bias is\nubiquitous - 80% of our personas demonstrate bias; it is significant - some\ndatasets show performance drops of 70%+; and can be especially harmful for\ncertain groups - some personas suffer statistically significant drops on 80%+\nof the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with\nGPT-4-Turbo showing the least but still a problematic amount of bias (evident\nin 42% of the personas). Further analysis shows that these persona-induced\nerrors can be hard-to-discern and hard-to-avoid. Our findings serve as a\ncautionary tale that the practice of assigning personas to LLMs - a trend on\nthe rise - can surface their deep-rooted biases and have unforeseeable and\ndetrimental side-effects.",
+                    "authors": "Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot",
+                    "published": "2023-11-08",
+                    "updated": "2024-01-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18580v1",
+                    "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+                    "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+                    "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+                    "published": "2023-11-30",
+                    "updated": "2023-11-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.03033v1",
+                    "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+                    "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+                    "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+                    "published": "2023-11-06",
+                    "updated": "2023-11-06",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.06899v4",
+                    "title": "Flames: Benchmarking Value Alignment of LLMs in Chinese",
+                    "abstract": "The widespread adoption of large language models (LLMs) across various\nregions underscores the urgent need to evaluate their alignment with human\nvalues. Current benchmarks, however, fall short of effectively uncovering\nsafety vulnerabilities in LLMs. Despite numerous models achieving high scores\nand 'topping the chart' in these evaluations, there is still a significant gap\nin LLMs' deeper alignment with human values and achieving genuine harmlessness.\nTo this end, this paper proposes a value alignment benchmark named Flames,\nwhich encompasses both common harmlessness principles and a unique morality\ndimension that integrates specific Chinese values such as harmony. Accordingly,\nwe carefully design adversarial prompts that incorporate complex scenarios and\njailbreaking methods, mostly with implicit malice. By prompting 17 mainstream\nLLMs, we obtain model responses and rigorously annotate them for detailed\nevaluation. Our findings indicate that all the evaluated LLMs demonstrate\nrelatively poor performance on Flames, particularly in the safety and fairness\ndimensions. We also develop a lightweight specified scorer capable of scoring\nLLMs across multiple dimensions to efficiently evaluate new models on the\nbenchmark. The complexity of Flames has far exceeded existing benchmarks,\nsetting a new challenge for contemporary LLMs and highlighting the need for\nfurther alignment of LLMs. Our benchmark is publicly available at\nhttps://github.com/AIFlames/Flames.",
+                    "authors": "Kexin Huang, Xiangyang Liu, Qianyu Guo, Tianxiang Sun, Jiawei Sun, Yaru Wang, Zeyang Zhou, Yixu Wang, Yan Teng, Xipeng Qiu, Yingchun Wang, Dahua Lin",
+                    "published": "2023-11-12",
+                    "updated": "2024-04-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00884v2",
+                    "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+                    "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+                    "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+                    "published": "2024-03-01",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.DB",
+                    "cats": [
+                        "cs.DB",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2304.03728v1",
+                    "title": "Interpretable Unified Language Checking",
+                    "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+                    "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+                    "published": "2023-04-07",
+                    "updated": "2023-04-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14769v3",
+                    "title": "Large Language Model (LLM) Bias Index -- LLMBI",
+                    "abstract": "The Large Language Model Bias Index (LLMBI) is a pioneering approach designed\nto quantify and address biases inherent in large language models (LLMs), such\nas GPT-4. We recognise the increasing prevalence and impact of LLMs across\ndiverse sectors. This research introduces a novel metric, LLMBI, to\nsystematically measure and mitigate biases potentially skewing model responses.\nWe formulated LLMBI using a composite scoring system incorporating multiple\ndimensions of bias, including but not limited to age, gender, and racial\nbiases. To operationalise this metric, we engaged in a multi-step process\ninvolving collecting and annotating LLM responses, applying sophisticated\nNatural Language Processing (NLP) techniques for bias detection, and computing\nthe LLMBI score through a specially crafted mathematical formula. The formula\nintegrates weighted averages of various bias dimensions, a penalty for dataset\ndiversity deficiencies, and a correction for sentiment biases. Our empirical\nanalysis, conducted using responses from OpenAI's API, employs advanced\nsentiment analysis as a representative method for bias detection. The research\nreveals LLMs, whilst demonstrating impressive capabilities in text generation,\nexhibit varying degrees of bias across different dimensions. LLMBI provides a\nquantifiable measure to compare biases across models and over time, offering a\nvital tool for systems engineers, researchers and regulators in enhancing the\nfairness and reliability of LLMs. It highlights the potential of LLMs in\nmimicking unbiased human-like responses. Additionally, it underscores the\nnecessity of continuously monitoring and recalibrating such models to align\nwith evolving societal norms and ethical standards.",
+                    "authors": "Abiodun Finbarrs Oketunji, Muhammad Anas, Deepthi Saina",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.11595v3",
+                    "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+                    "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+                    "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+                    "published": "2023-05-19",
+                    "updated": "2023-10-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.19118v1",
+                    "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate",
+                    "abstract": "Modern large language models (LLMs) like ChatGPT have shown remarkable\nperformance on general language tasks but still struggle on complex reasoning\ntasks, which drives the research on cognitive behaviors of LLMs to explore\nhuman-like problem-solving strategies. Along this direction, one representative\nstrategy is self-reflection, which asks an LLM to refine the solution with the\nfeedback generated by itself iteratively. However, our study shows that such\nreflection-style methods suffer from the Degeneration-of-Thought (DoT) problem:\nonce the LLM has established confidence in its solutions, it is unable to\ngenerate novel thoughts later through reflection even if its initial stance is\nincorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD)\nframework, in which multiple agents express their arguments in the state of\n\"tit for tat\" and a judge manages the debate process to obtain a final\nsolution. Clearly, our MAD framework encourages divergent thinking in LLMs\nwhich would be helpful for tasks that require deep levels of contemplation.\nExperiment results on two challenging datasets, commonsense machine translation\nand counter-intuitive arithmetic reasoning, demonstrate the effectiveness of\nour MAD framework. Extensive analyses suggest that the adaptive break of debate\nand the modest level of \"tit for tat\" state are required for MAD to obtain good\nperformance. Moreover, we find that LLMs might not be a fair judge if different\nLLMs are used for agents. Codes:\nhttps://github.com/Skytliang/Multi-Agents-Debate",
+                    "authors": "Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi",
+                    "published": "2023-05-30",
+                    "updated": "2023-05-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.19465v1",
+                    "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+                    "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+                    "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+                    "published": "2024-02-29",
+                    "updated": "2024-02-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10149v2",
+                    "title": "A Survey on Fairness in Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have shown powerful performance and development\nprospects and are widely deployed in the real world. However, LLMs can capture\nsocial biases from unprocessed training data and propagate the biases to\ndownstream tasks. Unfair LLM systems have undesirable social impacts and\npotential harms. In this paper, we provide a comprehensive review of related\nresearch on fairness in LLMs. Considering the influence of parameter magnitude\nand training paradigm on research strategy, we divide existing fairness\nresearch into oriented to medium-sized LLMs under pre-training and fine-tuning\nparadigms and oriented to large-sized LLMs under prompting paradigms. First,\nfor medium-sized LLMs, we introduce evaluation metrics and debiasing methods\nfrom the perspectives of intrinsic bias and extrinsic bias, respectively. Then,\nfor large-sized LLMs, we introduce recent fairness research, including fairness\nevaluation, reasons for bias, and debiasing methods. Finally, we discuss and\nprovide insight on the challenges and future directions for the development of\nfairness in LLMs.",
+                    "authors": "Yingji Li, Mengnan Du, Rui Song, Xin Wang, Ying Wang",
+                    "published": "2023-08-20",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15398v1",
+                    "title": "Fairness-Aware Structured Pruning in Transformers",
+                    "abstract": "The increasing size of large language models (LLMs) has introduced challenges\nin their training and inference. Removing model components is perceived as a\nsolution to tackle the large model sizes, however, existing pruning methods\nsolely focus on performance, without considering an essential aspect for the\nresponsible use of LLMs: model fairness. It is crucial to address the fairness\nof LLMs towards diverse groups, such as women, Black people, LGBTQ+, Jewish\ncommunities, among others, as they are being deployed and available to a wide\naudience. In this work, first, we investigate how attention heads impact\nfairness and performance in pre-trained transformer-based language models. We\nthen propose a novel method to prune the attention heads that negatively impact\nfairness while retaining the heads critical for performance, i.e. language\nmodeling capabilities. Our approach is practical in terms of time and\nresources, as it does not require fine-tuning the final pruned, and fairer,\nmodel. Our findings demonstrate a reduction in gender bias by 19%, 19.5%,\n39.5%, 34.7%, 23%, and 8% for DistilGPT-2, GPT-2, GPT-Neo of two different\nsizes, GPT-J, and Llama 2 models, respectively, in comparison to the biased\nmodel, with only a slight decrease in performance.",
+                    "authors": "Abdelrahman Zayed, Goncalo Mordido, Samira Shabanian, Ioana Baldini, Sarath Chandar",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.08495v2",
+                    "title": "Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humans",
+                    "abstract": "Large language models (LLMs) are becoming pervasive in everyday life, yet\ntheir propensity to reproduce biases inherited from training data remains a\npressing concern. Prior investigations into bias in LLMs have focused on the\nassociation of social groups with stereotypical attributes. However, this is\nonly one form of human bias such systems may reproduce. We investigate a new\nform of bias in LLMs that resembles a social psychological phenomenon where\nsocially subordinate groups are perceived as more homogeneous than socially\ndominant groups. We had ChatGPT, a state-of-the-art LLM, generate texts about\nintersectional group identities and compared those texts on measures of\nhomogeneity. We consistently found that ChatGPT portrayed African, Asian, and\nHispanic Americans as more homogeneous than White Americans, indicating that\nthe model described racial minority groups with a narrower range of human\nexperience. ChatGPT also portrayed women as more homogeneous than men, but\nthese differences were small. Finally, we found that the effect of gender\ndiffered across racial/ethnic groups such that the effect of gender was\nconsistent within African and Hispanic Americans but not within Asian and White\nAmericans. We argue that the tendency of LLMs to describe groups as less\ndiverse risks perpetuating stereotypes and discriminatory behavior.",
+                    "authors": "Messi H. J. Lee, Jacob M. Montgomery, Calvin K. Lai",
+                    "published": "2024-01-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15451v1",
+                    "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+                    "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+                    "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+                    "published": "2024-03-18",
+                    "updated": "2024-03-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.01349v1",
+                    "title": "Fairness in Large Language Models: A Taxonomic Survey",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+                    "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+                    "published": "2024-03-31",
+                    "updated": "2024-03-31",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.14473v1",
+                    "title": "The Ethics of ChatGPT in Medicine and Healthcare: A Systematic Review on Large Language Models (LLMs)",
+                    "abstract": "With the introduction of ChatGPT, Large Language Models (LLMs) have received\nenormous attention in healthcare. Despite their potential benefits, researchers\nhave underscored various ethical implications. While individual instances have\ndrawn much attention, the debate lacks a systematic overview of practical\napplications currently researched and ethical issues connected to them. Against\nthis background, this work aims to map the ethical landscape surrounding the\ncurrent stage of deployment of LLMs in medicine and healthcare. Electronic\ndatabases and preprint servers were queried using a comprehensive search\nstrategy. Studies were screened and extracted following a modified rapid review\napproach. Methodological quality was assessed using a hybrid approach. For 53\nrecords, a meta-aggregative synthesis was performed. Four fields of\napplications emerged and testify to a vivid exploration phase. Advantages of\nusing LLMs are attributed to their capacity in data analysis, personalized\ninformation provisioning, support in decision-making, mitigating information\nloss and enhancing information accessibility. However, we also identifies\nrecurrent ethical concerns connected to fairness, bias, non-maleficence,\ntransparency, and privacy. A distinctive concern is the tendency to produce\nharmful misinformation or convincingly but inaccurate content. A recurrent plea\nfor ethical guidance and human oversight is evident. Given the variety of use\ncases, it is suggested that the ethical guidance debate be reframed to focus on\ndefining what constitutes acceptable human oversight across the spectrum of\napplications. This involves considering diverse settings, varying potentials\nfor harm, and different acceptable thresholds for performance and certainty in\nhealthcare. In addition, a critical inquiry is necessary to determine the\nextent to which the current experimental use of LLMs is necessary and\njustified.",
+                    "authors": "Joschka Haltaufderheide, Robert Ranisch",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.09397v1",
+                    "title": "Do Large GPT Models Discover Moral Dimensions in Language Representations? A Topological Study Of Sentence Embeddings",
+                    "abstract": "As Large Language Models are deployed within Artificial Intelligence systems,\nthat are increasingly integrated with human society, it becomes more important\nthan ever to study their internal structures. Higher level abilities of LLMs\nsuch as GPT-3.5 emerge in large part due to informative language\nrepresentations they induce from raw text data during pre-training on trillions\nof words. These embeddings exist in vector spaces of several thousand\ndimensions, and their processing involves mapping between multiple vector\nspaces, with total number of parameters on the order of trillions. Furthermore,\nthese language representations are induced by gradient optimization, resulting\nin a black box system that is hard to interpret. In this paper, we take a look\nat the topological structure of neuronal activity in the \"brain\" of Chat-GPT's\nfoundation language model, and analyze it with respect to a metric representing\nthe notion of fairness. We develop a novel approach to visualize GPT's moral\ndimensions. We first compute a fairness metric, inspired by social psychology\nliterature, to identify factors that typically influence fairness assessments\nin humans, such as legitimacy, need, and responsibility. Subsequently, we\nsummarize the manifold's shape using a lower-dimensional simplicial complex,\nwhose topology is derived from this metric. We color it with a heat map\nassociated with this fairness metric, producing human-readable visualizations\nof the high-dimensional sentence manifold. Our results show that sentence\nembeddings based on GPT-3.5 can be decomposed into two submanifolds\ncorresponding to fair and unfair moral judgments. This indicates that GPT-based\nlanguage models develop a moral dimension within their representation spaces\nand induce an understanding of fairness during their training process.",
+                    "authors": "Stephen Fitz",
+                    "published": "2023-09-17",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.05668v1",
+                    "title": "CFaiRLLM: Consumer Fairness Evaluation in Large-Language Model Recommender System",
+                    "abstract": "In the evolving landscape of recommender systems, the integration of Large\nLanguage Models (LLMs) such as ChatGPT marks a new era, introducing the concept\nof Recommendation via LLM (RecLLM). While these advancements promise\nunprecedented personalization and efficiency, they also bring to the fore\ncritical concerns regarding fairness, particularly in how recommendations might\ninadvertently perpetuate or amplify biases associated with sensitive user\nattributes. In order to address these concerns, our study introduces a\ncomprehensive evaluation framework, CFaiRLLM, aimed at evaluating (and thereby\nmitigating) biases on the consumer side within RecLLMs.\n  Our research methodically assesses the fairness of RecLLMs by examining how\nrecommendations might vary with the inclusion of sensitive attributes such as\ngender, age, and their intersections, through both similarity alignment and\ntrue preference alignment. By analyzing recommendations generated under\ndifferent conditions-including the use of sensitive attributes in user\nprompts-our framework identifies potential biases in the recommendations\nprovided. A key part of our study involves exploring how different detailed\nstrategies for constructing user profiles (random, top-rated, recent) impact\nthe alignment between recommendations made without consideration of sensitive\nattributes and those that are sensitive-attribute-aware, highlighting the bias\nmechanisms within RecLLMs.\n  The findings in our study highlight notable disparities in the fairness of\nrecommendations, particularly when sensitive attributes are integrated into the\nrecommendation process, either individually or in combination. The analysis\ndemonstrates that the choice of user profile sampling strategy plays a\nsignificant role in affecting fairness outcomes, highlighting the complexity of\nachieving fair recommendations in the era of LLMs.",
+                    "authors": "Yashar Deldjoo, Tommaso di Noia",
+                    "published": "2024-03-08",
+                    "updated": "2024-03-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00811v1",
+                    "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+                    "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+                    "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+                    "published": "2024-02-25",
+                    "updated": "2024-02-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08472v1",
+                    "title": "Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models",
+                    "abstract": "Recently, work in NLP has shifted to few-shot (in-context) learning, with\nlarge language models (LLMs) performing well across a range of tasks. However,\nwhile fairness evaluations have become a standard for supervised methods,\nlittle is known about the fairness of LLMs as prediction systems. Further,\ncommon standard methods for fairness involve access to models weights or are\napplied during finetuning, which are not applicable in few-shot learning. Do\nLLMs exhibit prediction biases when used for standard NLP tasks? In this work,\nwe explore the effect of shots, which directly affect the performance of\nmodels, on the fairness of LLMs as NLP classification systems. We consider how\ndifferent shot selection strategies, both existing and new demographically\nsensitive methods, affect model fairness across three standard fairness\ndatasets. We discuss how future work can include LLM fairness evaluations.",
+                    "authors": "Carlos Aguirre, Kuleen Sasse, Isabel Cachola, Mark Dredze",
+                    "published": "2023-11-14",
+                    "updated": "2023-11-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.18502v1",
+                    "title": "Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification",
+                    "abstract": "Employing Large Language Models (LLM) in various downstream applications such\nas classification is crucial, especially for smaller companies lacking the\nexpertise and resources required for fine-tuning a model. Fairness in LLMs\nhelps ensure inclusivity, equal representation based on factors such as race,\ngender and promotes responsible AI deployment. As the use of LLMs has become\nincreasingly prevalent, it is essential to assess whether LLMs can generate\nfair outcomes when subjected to considerations of fairness. In this study, we\nintroduce a framework outlining fairness regulations aligned with various\nfairness definitions, with each definition being modulated by varying degrees\nof abstraction. We explore the configuration for in-context learning and the\nprocedure for selecting in-context demonstrations using RAG, while\nincorporating fairness rules into the process. Experiments conducted with\ndifferent LLMs indicate that GPT-4 delivers superior results in terms of both\naccuracy and fairness compared to other models. This work is one of the early\nattempts to achieve fairness in prediction tasks by utilizing LLMs through\nin-context learning.",
+                    "authors": "Garima Chhikara, Anurag Sharma, Kripabandhu Ghosh, Abhijnan Chakraborty",
+                    "published": "2024-02-28",
+                    "updated": "2024-02-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.02219v1",
+                    "title": "FairEvalLLM. A Comprehensive Framework for Benchmarking Fairness in Large Language Model Recommender Systems",
+                    "abstract": "This paper presents a framework for evaluating fairness in recommender\nsystems powered by Large Language Models (RecLLMs), addressing the need for a\nunified approach that spans various fairness dimensions including sensitivity\nto user attributes, intrinsic fairness, and discussions of fairness based on\nunderlying benefits. In addition, our framework introduces counterfactual\nevaluations and integrates diverse user group considerations to enhance the\ndiscourse on fairness evaluation for RecLLMs.\n  Our key contributions include the development of a robust framework for\nfairness evaluation in LLM-based recommendations and a structured method to\ncreate \\textit{informative user profiles} from demographic data, historical\nuser preferences, and recent interactions. We argue that the latter is\nessential for enhancing personalization in such systems, especially in\ntemporal-driven scenarios. We demonstrate the utility of our framework through\npractical applications on two datasets, LastFM-1K and ML-1M. We conduct\nexperiments on a subsample of 80 users from each dataset, testing and assessing\nthe effectiveness of various prompt construction scenarios and in-context\nlearning, comprising more than 50 scenarios. This results in more than 4000\nrecommendations (80 * 50 = 4000). Our study reveals that while there are no\nsignificant unfairness issues in scenarios involving sensitive attributes, some\nconcerns remain. However, in terms of intrinsic fairness, which does not\ninvolve direct sensitivity, unfairness across demographic groups remains\nsignificant. The code and data used for this paper are available at:\n\\url{https://shorturl.at/awBFM}.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2024-05-03",
+                    "updated": "2024-05-03",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00306v1",
+                    "title": "Probing Explicit and Implicit Gender Bias through LLM Conditional Text Generation",
+                    "abstract": "Large Language Models (LLMs) can generate biased and toxic responses. Yet\nmost prior work on LLM gender bias evaluation requires predefined\ngender-related phrases or gender stereotypes, which are challenging to be\ncomprehensively collected and are limited to explicit bias evaluation. In\naddition, we believe that instances devoid of gender-related language or\nexplicit stereotypes in inputs can still induce gender bias in LLMs. Thus, in\nthis work, we propose a conditional text generation mechanism without the need\nfor predefined gender phrases and stereotypes. This approach employs three\ntypes of inputs generated through three distinct strategies to probe LLMs,\naiming to show evidence of explicit and implicit gender biases in LLMs. We also\nutilize explicit and implicit evaluation metrics to evaluate gender bias in\nLLMs under different strategies. Our experiments demonstrate that an increased\nmodel size does not consistently lead to enhanced fairness and all tested LLMs\nexhibit explicit and/or implicit gender bias, even when explicit gender\nstereotypes are absent in the inputs.",
+                    "authors": "Xiangjue Dong, Yibo Wang, Philip S. Yu, James Caverlee",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.15215v1",
+                    "title": "Item-side Fairness of Large Language Model-based Recommendation System",
+                    "abstract": "Recommendation systems for Web content distribution intricately connect to\nthe information access and exposure opportunities for vulnerable populations.\nThe emergence of Large Language Models-based Recommendation System (LRS) may\nintroduce additional societal challenges to recommendation systems due to the\ninherent biases in Large Language Models (LLMs). From the perspective of\nitem-side fairness, there remains a lack of comprehensive investigation into\nthe item-side fairness of LRS given the unique characteristics of LRS compared\nto conventional recommendation systems. To bridge this gap, this study examines\nthe property of LRS with respect to item-side fairness and reveals the\ninfluencing factors of both historical users' interactions and inherent\nsemantic biases of LLMs, shedding light on the need to extend conventional\nitem-side fairness methods for LRS. Towards this goal, we develop a concise and\neffective framework called IFairLRS to enhance the item-side fairness of an\nLRS. IFairLRS covers the main stages of building an LRS with specifically\nadapted strategies to calibrate the recommendations of LRS. We utilize IFairLRS\nto fine-tune LLaMA, a representative LLM, on \\textit{MovieLens} and\n\\textit{Steam} datasets, and observe significant item-side fairness\nimprovements. The code can be found in\nhttps://github.com/JiangM-C/IFairLRS.git.",
+                    "authors": "Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, Xiangnan He",
+                    "published": "2024-02-23",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00625v2",
+                    "title": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models",
+                    "abstract": "The burgeoning field of Large Language Models (LLMs), exemplified by\nsophisticated models like OpenAI's ChatGPT, represents a significant\nadvancement in artificial intelligence. These models, however, bring forth\nsubstantial challenges in the high consumption of computational, memory,\nenergy, and financial resources, especially in environments with limited\nresource capabilities. This survey aims to systematically address these\nchallenges by reviewing a broad spectrum of techniques designed to enhance the\nresource efficiency of LLMs. We categorize methods based on their optimization\nfocus: computational, memory, energy, financial, and network resources and\ntheir applicability across various stages of an LLM's lifecycle, including\narchitecture design, pretraining, finetuning, and system design. Additionally,\nthe survey introduces a nuanced categorization of resource efficiency\ntechniques by their specific resource types, which uncovers the intricate\nrelationships and mappings between various resources and corresponding\noptimization techniques. A standardized set of evaluation metrics and datasets\nis also presented to facilitate consistent and fair comparisons across\ndifferent models and techniques. By offering a comprehensive overview of the\ncurrent sota and identifying open research avenues, this survey serves as a\nfoundational reference for researchers and practitioners, aiding them in\ndeveloping more sustainable and efficient LLMs in a rapidly evolving landscape.",
+                    "authors": "Guangji Bai, Zheng Chai, Chen Ling, Shiyu Wang, Jiaying Lu, Nan Zhang, Tingwei Shi, Ziyang Yu, Mengdan Zhu, Yifei Zhang, Carl Yang, Yue Cheng, Liang Zhao",
+                    "published": "2024-01-01",
+                    "updated": "2024-01-04",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.12090v1",
+                    "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+                    "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+                    "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+                    "published": "2023-05-20",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.01964v1",
+                    "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+                    "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+                    "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.09447v2",
+                    "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+                    "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+                    "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+                    "published": "2023-11-15",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.11033v4",
+                    "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+                    "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+                    "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+                    "published": "2024-01-19",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.13343v1",
+                    "title": "Challenges and Contributing Factors in the Utilization of Large Language Models (LLMs)",
+                    "abstract": "With the development of large language models (LLMs) like the GPT series,\ntheir widespread use across various application scenarios presents a myriad of\nchallenges. This review initially explores the issue of domain specificity,\nwhere LLMs may struggle to provide precise answers to specialized questions\nwithin niche fields. The problem of knowledge forgetting arises as these LLMs\nmight find it hard to balance old and new information. The knowledge repetition\nphenomenon reveals that sometimes LLMs might deliver overly mechanized\nresponses, lacking depth and originality. Furthermore, knowledge illusion\ndescribes situations where LLMs might provide answers that seem insightful but\nare actually superficial, while knowledge toxicity focuses on harmful or biased\ninformation outputs. These challenges underscore problems in the training data\nand algorithmic design of LLMs. To address these issues, it's suggested to\ndiversify training data, fine-tune models, enhance transparency and\ninterpretability, and incorporate ethics and fairness training. Future\ntechnological trends might lean towards iterative methodologies, multimodal\nlearning, model personalization and customization, and real-time learning and\nfeedback mechanisms. In conclusion, future LLMs should prioritize fairness,\ntransparency, and ethics, ensuring they uphold high moral and ethical standards\nwhen serving humanity.",
+                    "authors": "Xiaoliang Chen, Liangbin Li, Le Chang, Yunhe Huang, Yuxuan Zhao, Yuxiao Zhang, Dinuo Li",
+                    "published": "2023-10-20",
+                    "updated": "2023-10-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.07420v1",
+                    "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+                    "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+                    "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+                    "published": "2023-12-12",
+                    "updated": "2023-12-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.11761v1",
+                    "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+                    "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2023-07-14",
+                    "updated": "2023-07-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04205v2",
+                    "title": "Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves",
+                    "abstract": "Misunderstandings arise not only in interpersonal communication but also\nbetween humans and Large Language Models (LLMs). Such discrepancies can make\nLLMs interpret seemingly unambiguous questions in unexpected ways, yielding\nincorrect responses. While it is widely acknowledged that the quality of a\nprompt, such as a question, significantly impacts the quality of the response\nprovided by LLMs, a systematic method for crafting questions that LLMs can\nbetter comprehend is still underdeveloped. In this paper, we present a method\nnamed `Rephrase and Respond' (RaR), which allows LLMs to rephrase and expand\nquestions posed by humans and provide responses in a single prompt. This\napproach serves as a simple yet effective prompting method for improving\nperformance. We also introduce a two-step variant of RaR, where a rephrasing\nLLM first rephrases the question and then passes the original and rephrased\nquestions together to a different responding LLM. This facilitates the\neffective utilization of rephrased questions generated by one LLM with another.\nOur experiments demonstrate that our methods significantly improve the\nperformance of different models across a wide range to tasks. We further\nprovide a comprehensive comparison between RaR and the popular Chain-of-Thought\n(CoT) methods, both theoretically and empirically. We show that RaR is\ncomplementary to CoT and can be combined with CoT to achieve even better\nperformance. Our work not only contributes to enhancing LLM performance\nefficiently and effectively but also sheds light on a fair evaluation of LLM\ncapabilities. Data and codes are available at\nhttps://github.com/uclaml/Rephrase-and-Respond.",
+                    "authors": "Yihe Deng, Weitong Zhang, Zixiang Chen, Quanquan Gu",
+                    "published": "2023-11-07",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18140v1",
+                    "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+                    "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+                    "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+                    "published": "2023-11-29",
+                    "updated": "2023-11-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.12736v1",
+                    "title": "Large Language Model Supply Chain: A Research Agenda",
+                    "abstract": "The rapid advancements in pre-trained Large Language Models (LLMs) and Large\nMultimodal Models (LMMs) have ushered in a new era of intelligent applications,\ntransforming fields ranging from natural language processing to content\ngeneration. The LLM supply chain represents a crucial aspect of the\ncontemporary artificial intelligence landscape. It encompasses the entire\nlifecycle of pre-trained models, from its initial development and training to\nits final deployment and application in various domains. This paper presents a\ncomprehensive overview of the LLM supply chain, highlighting its three core\nelements: 1) the model infrastructure, encompassing datasets and toolchain for\ntraining, optimization, and deployment; 2) the model lifecycle, covering\ntraining, testing, releasing, and ongoing maintenance; and 3) the downstream\napplication ecosystem, enabling the integration of pre-trained models into a\nwide range of intelligent applications. However, this rapidly evolving field\nfaces numerous challenges across these key components, including data privacy\nand security, model interpretability and fairness, infrastructure scalability,\nand regulatory compliance. Addressing these challenges is essential for\nharnessing the full potential of LLMs and ensuring their ethical and\nresponsible use. This paper provides a future research agenda for the LLM\nsupply chain, aiming at driving the continued advancement and responsible\ndeployment of these transformative LLMs.",
+                    "authors": "Shenao Wang, Yanjie Zhao, Xinyi Hou, Haoyu Wang",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.11653v2",
+                    "title": "\"It's a Fair Game\", or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents",
+                    "abstract": "The widespread use of Large Language Model (LLM)-based conversational agents\n(CAs), especially in high-stakes domains, raises many privacy concerns.\nBuilding ethical LLM-based CAs that respect user privacy requires an in-depth\nunderstanding of the privacy risks that concern users the most. However,\nexisting research, primarily model-centered, does not provide insight into\nusers' perspectives. To bridge this gap, we analyzed sensitive disclosures in\nreal-world ChatGPT conversations and conducted semi-structured interviews with\n19 LLM-based CA users. We found that users are constantly faced with trade-offs\nbetween privacy, utility, and convenience when using LLM-based CAs. However,\nusers' erroneous mental models and the dark patterns in system design limited\ntheir awareness and comprehension of the privacy risks. Additionally, the\nhuman-like interactions encouraged more sensitive disclosures, which\ncomplicated users' ability to navigate the trade-offs. We discuss practical\ndesign guidelines and the needs for paradigm shifts to protect the privacy of\nLLM-based CA users.",
+                    "authors": "Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, Tianshi Li",
+                    "published": "2023-09-20",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC",
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2303.01248v3",
+                    "title": "Can ChatGPT Assess Human Personalities? A General Evaluation Framework",
+                    "abstract": "Large Language Models (LLMs) especially ChatGPT have produced impressive\nresults in various areas, but their potential human-like psychology is still\nlargely unexplored. Existing works study the virtual personalities of LLMs but\nrarely explore the possibility of analyzing human personalities via LLMs. This\npaper presents a generic evaluation framework for LLMs to assess human\npersonalities based on Myers Briggs Type Indicator (MBTI) tests. Specifically,\nwe first devise unbiased prompts by randomly permuting options in MBTI\nquestions and adopt the average testing result to encourage more impartial\nanswer generation. Then, we propose to replace the subject in question\nstatements to enable flexible queries and assessments on different subjects\nfrom LLMs. Finally, we re-formulate the question instructions in a manner of\ncorrectness evaluation to facilitate LLMs to generate clearer responses. The\nproposed framework enables LLMs to flexibly assess personalities of different\ngroups of people. We further propose three evaluation metrics to measure the\nconsistency, robustness, and fairness of assessment results from\nstate-of-the-art LLMs including ChatGPT and GPT-4. Our experiments reveal\nChatGPT's ability to assess human personalities, and the average results\ndemonstrate that it can achieve more consistent and fairer assessments in spite\nof lower robustness against prompt biases compared with InstructGPT.",
+                    "authors": "Haocong Rao, Cyril Leung, Chunyan Miao",
+                    "published": "2023-03-01",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07609v3",
+                    "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+                    "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+                    "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+                    "published": "2023-05-12",
+                    "updated": "2023-10-17",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.09219v5",
+                    "title": "\"Kelly is a Warm Person, Joseph is a Role Model\": Gender Biases in LLM-Generated Reference Letters",
+                    "abstract": "Large Language Models (LLMs) have recently emerged as an effective tool to\nassist individuals in writing various types of content, including professional\ndocuments such as recommendation letters. Though bringing convenience, this\napplication also introduces unprecedented fairness concerns. Model-generated\nreference letters might be directly used by users in professional scenarios. If\nunderlying biases exist in these model-constructed letters, using them without\nscrutinization could lead to direct societal harms, such as sabotaging\napplication success rates for female applicants. In light of this pressing\nissue, it is imminent and necessary to comprehensively study fairness issues\nand associated harms in this real-world use case. In this paper, we critically\nexamine gender biases in LLM-generated reference letters. Drawing inspiration\nfrom social science findings, we design evaluation methods to manifest biases\nthrough 2 dimensions: (1) biases in language style and (2) biases in lexical\ncontent. We further investigate the extent of bias propagation by analyzing the\nhallucination bias of models, a term that we define to be bias exacerbation in\nmodel-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs-\nChatGPT and Alpaca, we reveal significant gender biases in LLM-generated\nrecommendation letters. Our findings not only warn against using LLMs for this\napplication without scrutinization, but also illuminate the importance of\nthoroughly studying hidden biases and harms in LLM-generated professional\ndocuments.",
+                    "authors": "Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng",
+                    "published": "2023-10-13",
+                    "updated": "2023-12-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.18276v1",
+                    "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+                    "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+                    "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+                    "published": "2024-04-28",
+                    "updated": "2024-04-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "D.1; I.2"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.11483v1",
+                    "title": "Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in\nvarious NLP tasks. However, previous works have shown these models are\nsensitive towards prompt wording, and few-shot demonstrations and their order,\nposing challenges to fair assessment of these models. As these models become\nmore powerful, it becomes imperative to understand and address these\nlimitations. In this paper, we focus on LLMs robustness on the task of\nmultiple-choice questions -- commonly adopted task to study reasoning and\nfact-retrieving capability of LLMs. Investigating the sensitivity of LLMs\ntowards the order of options in multiple-choice questions, we demonstrate a\nconsiderable performance gap of approximately 13% to 75% in LLMs on different\nbenchmarks, when answer options are reordered, even when using demonstrations\nin a few-shot setting. Through a detailed analysis, we conjecture that this\nsensitivity arises when LLMs are uncertain about the prediction between the\ntop-2/3 choices, and specific options placements may favor certain prediction\nbetween those top choices depending on the question caused by positional bias.\nWe also identify patterns in top-2 choices that amplify or mitigate the model's\nbias toward option placement. We found that for amplifying bias, the optimal\nstrategy involves positioning the top two choices as the first and last\noptions. Conversely, to mitigate bias, we recommend placing these choices among\nthe adjacent options. To validate our conjecture, we conduct various\nexperiments and adopt two approaches to calibrate LLMs' predictions, leading to\nup to 8 percentage points improvement across different models and benchmarks.",
+                    "authors": "Pouya Pezeshkpour, Estevam Hruschka",
+                    "published": "2023-08-22",
+                    "updated": "2023-08-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.16343v2",
+                    "title": "Evaluating, Understanding, and Improving Constrained Text Generation for Large Language Models",
+                    "abstract": "Advancements in natural language generation (NLG) and large language models\n(LLMs) have led to proficient text generation in various tasks. However,\nintegrating intricate constraints into neural text generation, due to LLMs'\nopacity, remains challenging. This study investigates constrained text\ngeneration for LLMs, where predefined constraints are applied during LLM's\ngeneration process. Our research mainly focuses on mainstream open-source LLMs,\ncategorizing constraints into lexical, structural, and relation-based types. We\nalso present various benchmarks to facilitate fair evaluation. The study\naddresses some key research questions, including evaluating, understanding and\nimproving constrained text generation for LLMs. Results illuminate LLMs'\ncapacity and deficiency to incorporate constraints and provide insights for\nfuture developments in constrained text generation. Codes and datasets will be\nreleased upon acceptance.",
+                    "authors": "Xiang Chen, Xiaojun Wan",
+                    "published": "2023-10-25",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.14345v2",
+                    "title": "Bias Testing and Mitigation in LLM-based Code Generation",
+                    "abstract": "Utilizing state-of-the-art Large Language Models (LLMs), automatic code\ngeneration models play a pivotal role in enhancing the productivity of software\ndevelopment procedures. As the adoption of LLMs becomes more widespread in\nsoftware coding ecosystems, a pressing issue has emerged: does the generated\ncode contain social bias and unfairness, such as those related to age, gender,\nand race? This issue concerns the integrity, fairness, and ethical foundation\nof software applications that depend on the code generated by these models, yet\nis under-explored in the literature. This paper presents a novel bias testing\nframework that is specifically designed for code generation tasks. Based on\nthis framework, we conduct an extensive evaluation of the bias in code\ngenerated by five state-of-the-art LLMs. Our findings reveal that 20.29% to\n44.93% code functions generated by the models under study are biased when\nhandling bias sensitive tasks (i.e., tasks that involve sensitive attributes\nsuch as age and gender). This indicates that the existing LLMs can be unfair in\ncode generation, posing risks of unintended and harmful software behaviors. To\nmitigate bias for code generation models, we evaluate five bias mitigation\nprompt strategies, i.e., utilizing bias testing results to refine the code\n(zero-shot), one-, few-shot, and two Chain-of-Thought (CoT) prompts. Our\nevaluation results illustrate that these strategies are all effective in\nmitigating bias. Overall, one-shot and few-shot learning are the two most\neffective. For GPT-4, 80% to 90% code bias can be removed with one-shot\nlearning.",
+                    "authors": "Dong Huang, Qingwen Bu, Jie Zhang, Xiaofei Xie, Junjie Chen, Heming Cui",
+                    "published": "2023-09-03",
+                    "updated": "2024-01-09",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.03852v2",
+                    "title": "FLM-101B: An Open LLM and How to Train It with $100K Budget",
+                    "abstract": "Large language models (LLMs) have achieved remarkable success in NLP and\nmultimodal tasks, among others. Despite these successes, two main challenges\nremain in developing LLMs: (i) high computational cost, and (ii) fair and\nobjective evaluations. In this paper, we report a solution to significantly\nreduce LLM training cost through a growth strategy. We demonstrate that a\n101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US\ndollars. Inspired by IQ tests, we also consolidate an additional range of\nevaluations on top of existing evaluations that focus on knowledge-oriented\nabilities. These IQ evaluations include symbolic mapping, rule understanding,\npattern mining, and anti-interference. Such evaluations minimize the potential\nimpact of memorization. Experimental results show that our model, named\nFLM-101B, trained with a budget of 100K US dollars, achieves performance\ncomparable to powerful and well-known models, e.g., GPT-3 and GLM-130B,\nespecially on the additional range of IQ evaluations. The checkpoint of\nFLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.",
+                    "authors": "Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Xuying Meng, Siqi Fan, Peng Han, Jing Li, Li Du, Bowen Qin, Zheng Zhang, Aixin Sun, Yequan Wang",
+                    "published": "2023-09-07",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.12150v1",
+                    "title": "Your Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One",
+                    "abstract": "The widespread adoption of large language models (LLMs) underscores the\nurgent need to ensure their fairness. However, LLMs frequently present dominant\nviewpoints while ignoring alternative perspectives from minority parties,\nresulting in potential biases. We hypothesize that these fairness-violating\nbehaviors occur because LLMs express their viewpoints using a human personality\nthat represents the majority of training data. In response to this, we validate\nthat prompting LLMs with specific roles can allow LLMs to express diverse\nviewpoints. Building on this insight and observation, we develop FairThinking,\na pipeline designed to automatically generate roles that enable LLMs to\narticulate diverse perspectives for fair expressions. To evaluate FairThinking,\nwe create a dataset with a thousand items covering three fairness-related\ntopics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to\ndemonstrate its superior performance.",
+                    "authors": "Tianlin Li, Xiaoyu Zhang, Chao Du, Tianyu Pang, Qian Liu, Qing Guo, Chao Shen, Yang Liu",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "I.2; J.4"
+                    ],
+                    "category": "LLM Fairness"
+                }
+            ],
+            [
+                {
+                    "url": "http://arxiv.org/abs/2404.15420v1",
+                    "title": "XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference",
+                    "abstract": "In-context learning (ICL) approaches typically leverage prompting to\ncondition decoder-only language model generation on reference information.\nJust-in-time processing of a context is inefficient due to the quadratic cost\nof self-attention operations, and caching is desirable. However, caching\ntransformer states can easily require almost as much space as the model\nparameters. When the right context isn't known in advance, caching ICL can be\nchallenging. This work addresses these limitations by introducing models that,\ninspired by the encoder-decoder architecture, use cross-attention to condition\ngeneration on reference text without the prompt. More precisely, we leverage\npre-trained decoder-only models and only train a small number of added layers.\nWe use Question-Answering (QA) as a testbed to evaluate the ability of our\nmodels to perform conditional generation and observe that they outperform ICL,\nare comparable to fine-tuned prompted LLMs, and drastically reduce the space\nfootprint relative to standard KV caching by two orders of magnitude.",
+                    "authors": "Jo\u00e3o Monteiro, \u00c9tienne Marcotte, Pierre-Andr\u00e9 No\u00ebl, Valentina Zantedeschi, David V\u00e1zquez, Nicolas Chapados, Christopher Pal, Perouz Taslakian",
+                    "published": "2024-04-23",
+                    "updated": "2024-04-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Original Paper",
+                    "paper_cat": "LLM Fairness",
+                    "gt": "Decoders as encoders. Repurposing pre-trained decoders is becoming a popular approach to leverage powerful language models for applications other than generative modeling. For example, GRIT (Muennighoff et al., 2024) converts a pretrained causal decoder into a bi-directional encoder, yielding sentence-level embeddings while maintaining its ability to perform autoregressive generation of text. However, unlike the models we consider, this conversion requires fine-tuning all model parameters instead of additional ones. Parameterefficient approaches to turn decoders into encoders were also proposed, such as in (Meng et al., 2024) and (BehnamGhader et al., 2024), where a pretrained MISTRAL decoder (Jiang et al., 2023) is fine-tuned in a contrastive setting using LORA adapters to yield sentence level representations for retrieval. Closer to our method is CODET5+ (Wang et al., 2023d), which also defines encoder-decoder architectures rather than turning decoders into sentence encoders. Similar to one of our variants (XCLLAMA), it is assembled from two pre-trained decoders, one used as an encoder and the other as a decoder and linked with a few cross-attention operations. However, CODET5+ requires fine-tuning the entire and relatively large encoder. We show that a good pre-trained decoder has good enough representations, but one can improve upon it efficiently using just a very small trainable encoder. Conditioning without prompts. One recent line of work has focused on controlling a model\u2019s generation by intervening in its parameters directly (Wang et al., 2023c; Zhang et al., 2023; Wang et al., 2023b), in particular, to either introduce or erase knowledge post-training. Such approaches would typically require accessing and rewriting the internal parameters of a pre-trained language model and are not amenable to frequently changing contexts such as in a Question-Answer setting. Efficient inference. There exist several methods for improving inference speed and memory footprint. One approach is to lower the numerical precision or quantize the model weights and/or data, which has been shown to preserve the model accuracy with only 8 bits per weight (Dettmers et al., 2022), or even lower precision with 4, 3, or even 2 bits per weight (Frantar et al., 2023). Alternatively, the key-query cache can be compressed (Ainslie et al., 2023; Nawrot et al., 2024), although doing so requires fine-tuning. Finally, using Flash Attention (Dao et al., 2022) leads to significant savings for just-in-time processing of contexts. These methods orthogonally complement what we present in this paper and can be combined with XC-LLAMA.",
+                    "pre_questions": [],
+                    "main_content": "Introduction Large Language Models (LLMs) have propelled advances in language modeling and enabled automatic production of almost human-like text. Despite impressive progress, challenges persist in applying LLMs in practical settings such as the risk of hallucinations (or rather confabulatations (Bottou and Sch\u00f6lkopf, 2023; Millidge, 2023)) and of nonfactual (Li et al., 2023a; Xu et al., 2024) or toxic content (Zou et al., 2023; Xhonneux et al., 2024) in their generated text. Moreover, without fine-tuning, it is surprisingly difficult to adapt these models to incorporate new information not included in their training data (Luo et al., 2023; Kalajdzievski, 2024). Indeed, while LLMs can answer queries about their training data somewhat accurately in a zero-shot fashion (Petroni et al., 2019), queries 86 88 90 92 94 BERTScore ICL (Chat) ICL-JIT_KV ICL-LoRA ICL-LoRA-JIT_KV XC (Ours) XC-Enc (Ours) 0 128 256 384 512 Context cache memory footprint per token (kB) 30 40 50 60 70 F1 Question-Answer performance Figure 1: Average QA performance vs. caching memory footprint per context token. The closer to the top-right corner, the better. XC-LLAMA variants drastically reduce cache size at a small cost in accuracy. regarding information not available at training time often lead to inaccuracies (Maynez et al., 2020; Ji et al., 2023) as one would expect. This work focuses on grounding LLM generation on contextual information provided at inference time (Figure 2(a)). The most common approach for conditioning model generation is arguably In-Context Learning (ICL) (Radford et al., 2019; Brown et al., 2020): one prepends the relevant context to the prompt to generate an answer conditioned on the combined query and context. This technique is a core component of popular frameworks, such as Retrieval-Augmented Generation (RAG) (Lewis et al., 2020), with the specificity that the relevant context is not known a priori, but has to be retrieved from a corpus (Ram et al., 2023). While somewhat effective and straightforward, ICL, as typically performed with decoder-only architectures, has flaws. On the one hand, ICL-based generation is known to present high variance with respect to the prompt template so that equivalent valid-looking prompt formats produce drastically different results (Chen et al., 2023). On the other 1 arXiv:2404.15420v1  [cs.CL]  23 Apr 2024 Chapter 2: The Vanishing Easter Egg... Compose a short poem on the given theme. All rhymes... (b) (a) (c) (d) context context query answer answer answer answer query context query query cache cache What dream was Harry having when the Bunny ate him? The rise and fall of the Example Empire. (this work) Figure 2: Faster inference in context-conditional language modeling. (a) A use case where the user\u2019s query must be interpreted within some context to generate an answer. This work focuses on cases where the query and the answer are both small (light ), but the context is large (heavy ). The resulting LLM time complexity is thus O \u0000|context|2\u0001 (slow ). (b) In-context learning (ICL) and retrieval augmented generation (RAG) are two examples where the query is used to look up the context from a finite corpus. (c) In many cases, the context can be processed in advance to a cache enabling O \u0000|context|(|query| + |answer|) \u0001 (fast ) inference on a given query. (d) Finite context corpus may be processed offline, enabling fast execution at inference. Since cache size affects storage and communication costs, we search for models requiring smaller cache. hand, ICL is costly in terms of time and space. Just-in-time processing of the context suffers from quadratic complexity on the length due to selfattention operations (Vaswani et al., 2017). The alternative is pre-processing and caching the context internal states (the so-called key-value or KV states) to speed up inference (Figure 2(c)). However, this can require the same order of space as the model parameters themselves (we give details in Section 2). Recent work has reduced the space requirements of KV caching by sub-sampling states (Xiao et al., 2023; Adnan et al., 2024), although at the cost of ignoring relevant content. To overcome these limitations, we propose alternatives to ICL that perform conditional generation without injecting the relevant information in the prompt (Figure 2(a)), and seek to implement lightweight cache methods as illustrated in Figure 2(d). Our approach is reminiscent of the, now arguably legacy, encoder-decoder architectures, as it relies on cross-attention layers to condition generation on pre-computed context encodings. More precisely, we propose cross-context-cache (XCCACHE), which stores only the outputs of the encoder hidden states and relies on cross-attention to ingest the cache at inference time. We instantiate XC-CACHE via two parameter-efficient approaches that leverage pre-trained decoder-only models and extend them with a separate encoder to process the context: one approach uses the frozen decoder as an encoder (called XC-LLAMA), and the other uses a small bi-directional encoder (called XC-LLAMAENC). Crucially, our encoder-decoder architectures are more amenable to caching the context states, requiring orders of magnitude less space than their ICL counterparts. When context caching is enabled, fine-tuned models result in higher accuracy performance, but demand a large memory footprint (and consequently, higher latency and cost). In contrast, our XC-CACHE approach substantially reduces cache memory requirements by nearly 98%; and as other efficiency-improving techniques like quantization (Frantar et al., 2023), this reduction comes at a minor cost in accuracy, as illustrated in Figure 1. Nevertheless, our method consistently outperforms ICL alternatives based on LLAMA 2 or GPT-3.5, as detailed in Section 5. Overall, we advocate for a conceptual shift in architecture design for conditional generation, which should recenter on caching and make it integral to a model\u2019s operation rather than an afterthought. Our contributions are summarized as follows: 1. Cacheability: We provide evidence that encoder-decoder architectures are good candidates for conditional generation since our cache-friendly models enhance model performance compared to ICL, while reducing cache memory footprint by more than 98%. 2. Parameter efficiency: We show that training a few cross-attention layers (and optionally, a small encoder) suffices to convert decoders into encoder-decoder pairs. We contribute a mix of training tasks that enable contextconditional generation without costly ICL. 3. Decoder-as-encoder: We show that representations extracted from pre-trained causal decoders can be used as-is to replace an encoder. 2 Caching Representations Let context, query and answer denote sequences of tokens from a shared vocabulary V. We write |context|, |query| and |answer| the respective length of these sequences. Figure 2(a) illustrates an LLM 2 which, conditioned on context, produces an answer to a user-specified query. Assumptions. In what follows, we make three main assumptions: (A1) the context is not unique and depends on the query; (A2) there exists a manageable amount of contexts to condition on; and (A3) the context is large, i.e., |query| + |answer| \u226a |context|. In doing so, we restrict ourselves to the regimes where processing contexts offline is both compelling (A1 and A3) and viable (A2). Many ICL use cases satisfy these assumptions. For example, when facing a number of different tasks, we may craft task-specific instructions, each detailing how to obtain the answer to the query (Figure 2(b)). The same applies to RAG-based question-answering when retrieved documents are encoded independently: the retriever selects from a corpus the most relevant documents (context) to answer the user question (query). 2.1 Inference Time Complexity In general, the time complexity of LLM inference is dominated by its attention mechanism, which in the ICL setting is O \u0000(|context| + |query| + |answer|)2\u0001 . In the large context regime (A3), this simplifies to O \u0000|context|2\u0001 : we can thus expect tangible inference speedups by improving how we handle the context. One way to achieve such speedups is to pre-process the contexts offline to some intermediate states (cache) and provide it to the model at inference (Figure 2(c)). This way, the quadratic cost of processing the context is paid once, allowing inference to simply look up a ready-made cache (Fig. 2(d)). Note that the incurred storage and communication overheads are linear in the size of the cache, which is linear in context length. In this setup, inference time complexity becomes O \u0000|context|(|query| + |answer|) \u0001 , i.e. linear in context length (Fig. 3), a significant speedup. 2.2 Practical costs of caching Practical considerations might negatively affect a cached-enabled setup. Loading and communication overheads are both linear in cache size, motivating smaller cache. Any extra operations required at inference mitigates caching benefits. Caching methods may incur an (implicit or explicit) cost in the quality of the generated answer. See Appendix A for details. answer query context (a) Without cache. answer query context (b) KV caching. Figure 3: Stylized representation of attention execution time (area) for context-conditional language modeling. Dashed lines in (b) shows the savings when the context\u2019s (past) keys and values cache are provided. For causal models, the area below the diagonal represents execution time that could be saved by other means. 2.3 Approaches to Caching KV Caching. The default approach is to store the (past) key and value states generated while processing context, hereafter called KV caching. KV caching is commonly associated with a conversational setting, in which case the cache often remains on the GPU device between conversational rounds (Figure 2(d)). Other setups, such as serving multiple conversations, demand however that we move cache from storage (and/or CPU RAM) to the GPU, incurring further latency. As an example, for LLAMA 2-7B (Touvron et al., 2023) using 16 bits precision, we must move around a whopping 512 kB per token.1 Smaller per-token cache sizes are thus desirable. JIT-KV Caching. An alternative approach is to store the (past) hidden states of the model in the cache (in the case of LLAMA 2-7B, this would be half as big, i.e., 256 kB per token). At inference time, once these hidden states are loaded on GPU, we can recover the full keys and values in O \u0000|context| \u0001 well-parallelized operations (recall that a transformer\u2019s keys and values are obtained by a linear operation on the hidden states). We call this approach \u201cjust-in-time\u201d or JIT-KV caching. XC Caching (Ours). Both KV and JIT-KV caching perform the exact same computations as the original model without a cache. They both suffer from two types of costs while producing the same answer: the size of the cache and the oper132 layers, 32 heads per layer, 128 floating points per key and value (thus a factor 2), 2 bytes per 16 bit floating point 3 Self-Attn Self-Attn Self-Attn Self-Attn Cross-attn Cross-attn Context Context Embedding Prompt Predictions Small trainable  encoder Trainable Encoder (a) Trainable small bi-directional encoder. Self-Attn Self-Attn Self-Attn Self-Attn Self-Attn Self-Attn Self-Attn Self-Attn Cross-attn Cross-attn Context Context Embedding Prompt Predictions Base decoder as  encoder Only train  cross-attn  layers (b) Decoder as the (causal) encoder. Figure 4: XC-LLAMA\u2019s architectures. A decoder-only model implements encoder-decoder architectures. Finetuning out in a parameter-efficient fashion via training only a small number of cross-attention layers. ations to be performed at inference (e.g., convert hidden states to keys and values). In this work, we propose XC caching (read cross-context caching), which considers the quality loss of the generated answer as a third such cost and present two models balancing these three costs. Both models employ cross-attention layers integrated into the decoder architecture, enabling the model to attend on context cached with a compact memory footprint at inference time. Models are described in the next section. 3 XC: Cross Attending to Efficiently Cached Context To reduce the memory footprint of caching, our approach draws inspiration from the encoder-decoder architecture, which until recently was the go-to design for conditional generation. Recently, decoderonly models have become more popular in part due to their data efficiency: the entire parameter budget is allocated to the decoder and not \u201cwasted\u201d on an encoder, with the additional advantage that all parameters are trained against all data tokens. Even though maintaining an external encoder may seem wasteful, we favor an encoder-decoder architecture: it better lends itself to pre-computing and caching context representations. Indeed, only the encoder output vectors need to be stored, as opposed to intermediate states across all of the decoder\u2019s selfattention layers (KV and JIT-KV cache). In what follows, we refer to a model as composed of an encoder E : V|context| 7\u2192Rd|context|, which takes in a context and outputs token-level representations of size d, and of a decoder D : Vm \u00d7 Rd|context| 7\u2192\u2206|V|, which takes as input the query and the context encodings and outputs an answer in the simplex of size |V|. More precisely, the decoder D autoregressively outputs the parameters of a categorical distribution over the vocabulary. For the sake of parameter efficiency and to leverage state-of-the-art pre-trained LLMs, we start from an existing decoder-only model and augment it with new cross-attention layers interleaved between existing self-attention layers, as illustrated in Figure 4. We consider two strategies to implement the encoder E. The first one trains a small encoder, i.e., E is orders of magnitude smaller than D. The second one uses a decoder as encoder, i.e., the frozen decoder-only model is used out-of-the-box as the encoder (E := D). More precisely, we use as encodings the representations extracted from the pre-trained D at its last layer before the language modeling head. Choosing one approach over the other depends on practical considerations. If caching is possible and context representations can be computed offline, then using the decoder as an encoder is preferable for overall simplicity and parameter efficiency. Otherwise, a small encoder would make just-intime processing of contexts significantly less costly. Both approaches inherit the advantages of trained decoder-only models while benefiting from using an encoder during inference. In particular, contextual information can be efficiently cached since only the output at E\u2019s top layer must be stored instead of the entire set of D\u2019s intermediate states. Finally, to enable context conditioning, we train exclusively the newly added modules: the crossattention layers in both settings and the small encoder in the setting that requires it. The base de4 coder is kept frozen in both settings (even when it acts as an encoder). Consequently, our training procedure does not affect the original parameters of D, which can still be used as a general-purpose language model when the additional layers are removed. 4 Experimental Setup We focus on the question-answering task and train on a combination of datasets where context, query, and answer triplets are available. QA is an ideal testbed for our approaches as it requires efficiently looking up external information and accounting for it at generation time. 4.1 Training Dataset We build a training dataset by standardizing and pooling together the training partitions of five publicly available and diverse (open-domain, multihop, conversational) datasets: NATURAL QUESTIONS (NQ) (Kwiatkowski et al., 2019), HOTPOTQA (Yang et al., 2018), TOPIOCQA (Adlakha et al., 2022), MS MARCO (Bajaj et al., 2016), and SQUAD-V2 (Rajpurkar et al., 2018). Examples from the resulting training dataset contain a query (natural-language question), an answer (expected output), and one or more contexts (e.g., knowledge base articles), where at least one of them contains the answer to the query. We refer to this context as reference context. We make use of a validation set for model selection and hyperparameter tuning. For datasets with a validation set but not a test partition, we use the validation set for testing and create a validation set by holding out 10% of randomly selected training samples. We apply a further filtering step in our training data and remove examples with contexts longer than 6,000 tokens, corresponding to less than 5% of the samples. Evaluation is performed on the resulting test partitions of NATURAL QUESTIONS, HOTPOTQA, and TOPIOCQA. Further details and statistics of the datasets we considered can be found in Appendix B. 4.2 Auxiliary Tasks In addition to training on the primary QA tasks, we optimize our models on context repetition tasks, as described below. The advantage of defining such auxiliary tasks is two-fold. On the one hand, they allow us to optimize the likelihood of all available tokens, even those used as input to the encoder. Figure 5: Multitask training strategy. Within an epoch, each example is presented to the model twice. In the first round, the model predicts the answer conditionally based on the context and the query. In the second appearance of an example within the epoch, we train the model to repeat or infill the context. On the other hand, they help avoid sub-optimal solutions where the cross-attention layers behave as simple identity operators. Indeed, our models could learn to ignore the context but cannot do so when tasked to repeat the context. In practice, we use reserved tokens to instruct the model to repeat the context, either as-is or by infilling (Bavarian et al., 2022), i.e. returning it in the prefix-suffix-middle or the suffix-prefix-middle order, as done by Li et al. (2023b); Lozhkov et al. (2024). Such tasks introduce new variability, as the model learns to copy-paste the context and to find and replace its missing chunks, resulting in improved performance in the multi-epoch training setting. Note that we train our models on questionanswering and context repetition on every training sample. We set one training epoch to correspond to two passes over the training dataset, where we perform the primary or the secondary tasks in each pass, as illustrated in Figure 5. 4.3 Implementation Details We rely on the openly available pre-trained LLAMA 2 (Touvron et al., 2023) to define variations of XC-LLAMA. Specifically, all our empirical assessments use the 7 billion parameter version of LLAMA 2. For the variation of XC-LLAMA where a dedicated encoder is trained (referred to as XC-LLAMAENC from now on), we fine-tune as encoder a LONGFORMER (Beltagy et al., 2020), which is a BERT (Devlin et al., 2018) of approximately 125M parameters, pre-trained on relatively long documents using efficient self-attention. We note, however, that the longest publicly available LONGFORMER accepts inputs of at most 4,096 tokens, which is shorter than our maximum length of 6,000. We then increase the model\u2019s maximum 5 input length by repeating the starting position embeddings. Upon fine-tuning, the model does handle contexts longer than 4,096 tokens without noticeable problems. As per adding cross-attention layers to XCLLAMA, we introduce one cross-attention layer every few self-attention layers of the transformer. In particular, we found the 5-6 configuration (i.e., inserting five cross-attention layers while skipping six self-attention ones) to work consistently well. We thus use this configuration throughout our evaluations. We remark that we strive to be parameter efficient and keep the parameter count of added modules below 10% of D\u2019s parameter count. Training is carried out with the ADAMW optimizer (Loshchilov and Hutter, 2017) with a batch size of 256 and for 40,000 steps (amounting to 4 training epochs). We use a linear learning rate scheduler with a warm-up phase and such that the learning rate reaches 0 at the last training step. A comprehensive list of hyperparameter values is shown in Table 8 in Appendix C. 4.4 Metrics and Evaluation We compare our models against ICL methods for generating answers conditioned on context and query. We note that contexts in the cases we considered present a relatively low signal-to-noise ratio, as most of the tokens are related to the answer but are not at all relevant. In some more extreme situations, the posed question cannot be answered from the context, and models are expected (and trained or prompted to) indicate that it is not possible to answer based on the provided context. We use the same metrics and evaluation setup described by Adlakha et al. (2023) \u2013 such as F1 SCORE, RECALL, METEOR, and ROUGEL \u2013 but keep F1 as our metric of focus. In addition, we evaluate BERTSCORE (Zhang et al., 2019) measured between predictions and ground-truth answers. 5 Results 5.1 Comparison with existing methods We first compare our method to existing approaches for conditional generation. Our main baseline is ICL, i.e., providing the context as part of the prompt. More specifically, we report baseline results for LLAMA 2-CHAT, which we found to perform better than the base pre-trained LLAMA 2. We further report results for GPT-3.5-TURBO, OpenAI\u2019s black-box model, presumably conditioned by prompting. For these ICL baselines, we selected the prompt templates based on generated answer quality on sample validation data (refer to Appendix C for details). Finally, we report the results of FUSION-IN-DECODER (FiD) (Izacard and Grave, 2021), a T5-based (Raffel et al., 2020) conditional generative model, which consistently proved to be state-of-the-art on QA tasks (Borgeaud et al., 2022; Wang et al., 2023a; Adlakha et al., 2023). Unlike the decoder-only backbone of our models, FiD is arguably no longer a generalpurpose model, as all of its parameters are finetuned to perform QA. More importantly, as discussed in more depth in Section 5.2, pre-processing and caching context representations is not an option for FiD since it requires knowing the question at encoding time. Nevertheless, we introduce the FiD baseline to check where our models stand relative to established QA-specialized models. Results are presented in Table 1. On the considered datasets, cross-attending to the contexts (XC-LLAMA or XC-LLAMAENC) considerably improves performance w.r.t prompting (LLAMA 2CHAT). The gap varies depending on the dataset. We conjecture that this is due to the high variance induced by prompting, although it might exist an optimal prompt for each dataset to help close this gap. Thus, approaches that do not rely on the prompt offer the advantage of being more broadly applicable and, hence, more practical. We also note that even in the setting where the decoder is used as an encoder, cross-attending to contexts still yields better performance than ICL, no matter the base decoder we compare against. This suggests that the trained cross-attention layers compensate for the potential sub-optimality because the encoder representations are not explicitly trained for the task we evaluate. A broader set of results comprising more metrics and models can be found in Table 5. 5.2 Trading off accuracy against efficiency The results in the previous section show that adding and fine-tuning dedicated parameters for context-conditioning improves performance relative to prompting. Based on this observation, in this section, we expand our evaluation to consider alternative approaches that leverage a small number of additional parameters that enable models to condition generation on reference contexts. In addition to prediction performance, we now also focus on computational efficiency; namely, we assess how amenable different models are to pre-processing 6 Dataset Model F1 BERTSCORE GPT-3.5 TURBO 57.80 90.87 FID 59.05 91.75 NQ LLAMA 2-CHAT 41.26 87.43 XC-LLAMA (Ours) 59.95 92.87 XC-LLAMAENC (Ours) 63.12 93.30 GPT-3.5 TURBO 39.37 87.83 FID 45.60 90.56 HOTPOTQA LLAMA 2-CHAT 29.63 86.02 XC-LLAMA (Ours) 43.94 90.55 XC-LLAMAENC (Ours) 54.57 92.08 GPT-3.5 TURBO 40.18 87.52 FID 31.22 85.95 TOPIOCQA LLAMA 2-CHAT 33.45 86.33 XC-LLAMA (Ours) 45.47 89.16 XC-LLAMAENC (Ours) 47.73 89.40 Table 1: Question-Answer performance on three diverse information-seeking tasks. GPT-3.5 TURBO and LLAMA2-CHAT are given reference context through ICL (prompting), while our approach uses crossattention layers to ingest reference embeddings. and caching representations of contexts. We thus fine-tune LORA adapters (Hu et al., 2021) applied to the same LLAMA 2 decoder we use for XC-LLAMA. In other words, we fine-tune our ICL baseline from Section 5 to control for the effect of prompting in the model performance. As observed in Table 3, such finetuning drastically improves QA accuracy relative to the ICL baselines reported in Table 1. However, despite enabling improvements in prediction performance, relying on LORA-style model adaptation still necessitates storing all KV states throughout every layer to cache contextual information, incurring significant costs. In contrast, models with an encoder require caching the hidden states of only their last layer. This is more clearly depicted in Table 2, where we further consider a compromise scenario for ICL that caches hidden states at each layer rather than KV states. We refer to this variant of ICL as LLAMA 2-ICL-JIT-KV, since it performs just-in-time KV projection of cached hidden states, trading time for space. Note that XC-LLAMA variants greatly reduce the caching footprint simply because they require storing only the last hidden states of the encoder. XC-LLAMAENC reduces space requirements even further, as the representation of LONGFORMER are lower dimensional relative to that of LLAMA 2. Reducing cache size carries substantial practical implications, notably in reducing the memory footprint of a pre-processed corpus by orders of magnitude. This is particularly significant when storing Model Cache size (kB/token) LLAMA 2-ICL-KV 512 LLAMA 2-ICL-JIT-KV 256 LLAMA 2-ICL-LoRA 512 XC-LLAMA (Ours) 8 XC-LLAMAENC (Ours) 1.5 Table 2: Cache memory footprint per context token for different models. The encoder-decoder approach of XCLLAMA allows for huge savings as it requires storing only the last hidden states instead of KV states throughout layers. Cache sizes here assume 16-bit precision. pre-processed representations of vast datasets like the entire Wikipedia. Additionally, reducing the cache size leads to runtime savings by mitigating communication costs, as the volume of information transferred from disk to device is markedly reduced. Finally, scaling down the cache size frees up device memory at inference time, enabling longer generation or larger batch sizes for batched inference. See appendix A for details. We provide a joint view of those two performance components \u2013 prediction and compute \u2013 in Figure 1, where we plot aggregate F1 and BERTSCORE across datasets versus the amount of cache per context token required by models. Note that models closer to the top-right corner are preferred since they are highly accurate at a low caching cost. While no method is optimal for both criteria, the Pareto set consists of: on one end, ICL models fine-tuned with LORA which offer slightly higher aggregated BERTSCORE but require substantial caching space; on the other end, models with an encoder which make small sacrifices in prediction accuracy while significantly reducing the memory footprint. We also note that there is a gap between XC-LLAMA and XC-LLAMAENC, and the additional parameters introduced by XCLLAMAENC yield a boost in accuracy and improve space efficiency. Detailed QA results are reported in Table 3, which includes the GPT-3.5 TURBO ICL baseline and FiD, which is now fine-tuned on our training dataset. Again, we highlight that FiD, while included for performance reference, is not directly comparable to the other models and does not support caching in pre-processed contexts. Our reporting of FiD aims to provide insight into where our models stand relative to a specialized baseline. As previously mentioned, our models incur a slight reduction in prediction accuracy but achieve significant space savings, which proves advantageous in 7 Dataset Model F1 BERTSCORE GPT-3.5 TURBO 57.80 90.87 FID (fine-tuned) 72.67 94.76 NQ LLAMA 2 (LoRA) 67.38 93.27 XC-LLAMA (Ours) 59.95 92.87 XC-LLAMAENC (Ours) 63.12 93.30 GPT-3.5 TURBO 39.37 87.83 FID (fine-tuned) 53.54 89.64 HOTPOTQA LLAMA 2 (LoRA) 71.97 94.62 XC-LLAMA (Ours) 43.94 90.55 XC-LLAMAENC (Ours) 54.57 92.08 GPT-3.5 TURBO 40.18 87.52 FID (fine-tuned) 41.52 86.54 TOPIOCQA LLAMA 2 (LoRA) 55.41 90.80 XC-LLAMA (Ours) 45.47 89.16 XC-LLAMAENC (Ours) 47.73 89.40 Table 3: Question-Answer performance with ICL, carried out with a model specialized to a prompt template via LORA. Note that the FiD model was pre-trained on NQ, hence its high performance on that dataset. various practical scenarios. For a comprehensive overview, please refer to the full results in Table 5. We introduced XC-LLAMA as an approach to transform a pre-trained decoder-only language model into an encoder-decoder architecture that can condition generation on both the encoder inputs and the decoder query. This is achieved by integrating cross-attention layers interleaved in between existing self-attention layers of the pre-trained decoder. We describe two approaches for defining the encoder: using a copy of the decoder or introducing a trainable but small bi-directional encoder. The proposed architecture allows for a reduction in caching space by a factor exceeding 300. When evaluated in the QA setting, we observe a higher prediction accuracy than what standard ICL approaches achieve, either with LLAMA 2 or GPT3.5 TURBO. Additionally, we observe accuracy levels nearly on par with caching-intensive finetuned prompted models, providing a more cachingfriendly alternative to prompted language models that proves highly practical. Identified limitations are discussed in Section 8. 8 8 Limitations Most of the models discussed in this paper have impressive results in the QA setting. However, our experience working with these models reveals limitations primarily stemming from their reliance on an underlying language model. As such, these methods inherit potential flaws of the language model they build upon. For instance, typical large language models are trained on vast amounts of text, likely including information related to questions in publicly available QA benchmarks. While such training on related data may offer a shortcut to models for correctly answering questions in the context that they have \u201cmemorized\u201d during training, it can also introduce errors: the model may \u201cremember\u201d related but inaccurate information relative to a specific query. In simpler terms, prompting-based ICL approaches and potentially our models can bypass context and rely solely on memory to generate a continuation. This is undesirable as we aim to ensure that models accurately account for the reference context rather than pre-training\u2019s. Fine-tuning models address the abovementioned issue by encouraging reliance on the provided context, as we explicitly do with context repetition and filling auxiliary tasks. However, it also specializes in fine-tuning data to such an extent that it may hurt the generalization ability of datasets that deviate from what was observed during fine-tuning. We estimate to what extent models know the answer after pre-training in Table 6. Unseen Datasets. To test the generalization ability of the models on out-of-distribution data, nonidentically distributed with respect to fine-tuning one, we curated a test dataset containing 16,290 reference documents, with roughly 5 question-answer pairs per document. The documents were written by human annotators, who wrote about imaginary scenarios, each featuring several subsections with titles and subtitles. Importantly, given that the reference documents in this dataset contain fictional or untrue information, we can reasonably assume that none of the models can rely on or be misled by their memory of the training data. Additionally, samples in this dataset exhibit unique styles that may differ from the fine-tuning data. When evaluated in this more challenging outof-distribution scenario, the accuracy of all models experiences a significant decline, as reported in Figure 6, where we compare QA scores for all models across all datasets, including the curated one la82 84 86 88 90 92 94 96 BERTScore NQ HotpotQA T opiOCQA Unseen FiD FiD (finetuned) Llama 2-Chat Llama 2 (LoRA) XC-LlamaEnc XC-Llama 10 20 30 40 50 60 70 F1 Question-Answer performance Figure 6: QA performance of various models on our three test datasets against our curated unseen dataset. All models struggle to generalize in this out-ofdistribution scenario where the test dataset significantly differs from that used for training. beled as Unseen. We hypothesize that, apart from the inability of models to rely on memory to answer queries, the main factor driving the accuracy decline is the divergence between the test dataset and the fine-tuning data. This discrepancy may arise from variations in writing style, document lengths, signal-to-noise ratio, and the presence of distracting content related to the question but not useful for its answer, among others. Identifying the exact sources of errors and enhancing the robustness of context-conditional models are promising avenues for future research. Ethics Statement We acknowledge the environmental impact of computational resources, including the carbon footprint associated with data storage and processing. By focusing on conditional generation through caching with a small memory footprint, we aim to minimize energy consumption and promote sustainability in computational practices. While cache reduction can lead to speed improvements, we recognize the importance of additionally implementing safeguards to ensure responsible usage of computational models. This entails mitigating bias, promoting fairness, and preserving privacy, thus fostering ethical deployment of technology. 9 Acknowledgements We would like to thank Christian Hudon for the invaluable technical support and the securing of computational resources, which enabled the scaling up of our experiments. We also thank Siva Reddy for insightful discussions that helped shape this work. Contributions Each author played a vital role in shaping the research and conceptual development of the project reported in this paper. While certain contributions eventually formed the basis of the presented work, every individual\u2019s input was instrumental in refining and sculpting the final iteration we present here. Jo\u00e3o Monteiro proposed the initial idea of performing conditional generation with new crossattention layers; wrote the code for the two XC-LLAMA architectures and training strategies and ran the training and evaluation experiments for all large-scale experiments in the paper; wrote the code for the LoRA baselines that were explored in the project; processed the training dataset as well as other datasets used for ablations; wrote several key sections of the paper and shaped it to the current version; designed the architecture figures. \u00c9tienne Marcotte conducted the experiments with the LoRA baseline architectures, wrote code for evaluation and early versions of metric computation; helped with code debugging and cleanup; wrote code to cleanup and create variants of the HOTPOTQA dataset; wrote several parts of the paper; generated the charts presented in the paper. Pierre-Andr\u00e9 No\u00ebl developed all cachingrelated arguments and experiments; reviewed and debugged code for XC-LLAMA architectures; maintained health of the overall codebase by developing tests and maintaining the structure; wrote various dataset-related code; wrote parts of the paper and created some of the figures. Valentina Zantedeschi wrote the code for XCLLAMA\u2019s generation and evaluation pipeline; integrated the FiD baseline in our codebase; conducted training and finetuning of the FiD model and evaluating them on our test data; pushed the empirical estimation of caching gains and wrote code to analyze and compute memory footprint per token results; wrote various parts of the paper and shaped it to the current version. David V\u00e1zquez, Nicolas Chapados, Christopher Pal advised the project as a whole, provided feedback on the experiments and the paper, and contributed to the reviewing of the paper writeup and structure. Perouz Taslakian contributed to conceiving and pushing forward the research direction in the early stages of the project; wrote code for automatic metric computation, the sensitivity analysis results, and the GPT baseline; performed analysis and spotchecks of intermediate results; contributed to the acquisition of GPUs for training the models; wrote several main parts of the paper."
+                },
+                {
+                    "url": "http://arxiv.org/abs/2011.11233v2",
+                    "title": "ROME: Robustifying Memory-Efficient NAS via Topology Disentanglement and Gradient Accumulation",
+                    "abstract": "Albeit being a prevalent architecture searching approach, differentiable\narchitecture search (DARTS) is largely hindered by its substantial memory cost\nsince the entire supernet resides in the memory. This is where the single-path\nDARTS comes in, which only chooses a single-path submodel at each step. While\nbeing memory-friendly, it also comes with low computational costs. Nonetheless,\nwe discover a critical issue of single-path DARTS that has not been primarily\nnoticed. Namely, it also suffers from severe performance collapse since too\nmany parameter-free operations like skip connections are derived, just like\nDARTS does. In this paper, we propose a new algorithm called RObustifying\nMemory-Efficient NAS (ROME) to give a cure. First, we disentangle the topology\nsearch from the operation search to make searching and evaluation consistent.\nWe then adopt Gumbel-Top2 reparameterization and gradient accumulation to\nrobustify the unwieldy bi-level optimization. We verify ROME extensively across\n15 benchmarks to demonstrate its effectiveness and robustness.",
+                    "authors": "Xiaoxing Wang, Xiangxiang Chu, Yuda Fan, Zhexi Zhang, Bo Zhang, Xiaokang Yang, Junchi Yan",
+                    "published": "2020-11-23",
+                    "updated": "2023-08-03",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CV"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07922v2",
+                    "title": "CodeT5+: Open Code Large Language Models for Code Understanding and Generation",
+                    "abstract": "Large language models (LLMs) pretrained on vast source code have achieved\nprominent progress in code intelligence. However, existing code LLMs have two\nmain limitations in terms of architecture and pretraining tasks. First, they\noften adopt a specific architecture (encoder-only or decoder-only) or rely on a\nunified encoder-decoder network for different downstream tasks. The former\nparadigm is limited by inflexibility in applications while in the latter, the\nmodel is treated as a single system for all tasks, leading to suboptimal\nperformance on a subset of tasks. Secondly, they often employ a limited set of\npretraining objectives which might not be relevant to some downstream tasks and\nhence result in substantial performance degrade. To address these limitations,\nwe propose ``CodeT5+'', a family of encoder-decoder LLMs for code in which\ncomponent modules can be flexibly combined to suit a wide range of downstream\ncode tasks. Such flexibility is enabled by our proposed mixture of pretraining\nobjectives to mitigate the pretrain-finetune discrepancy. These objectives\ncover span denoising, contrastive learning, text-code matching, and causal LM\npretraining tasks, on both unimodal and bimodal multilingual code corpora.\nFurthermore, we propose to initialize CodeT5+ with frozen off-the-shelf LLMs\nwithout training from scratch to efficiently scale up our models, and explore\ninstruction-tuning to align with natural language instructions. We extensively\nevaluate CodeT5+ on over 20 code-related benchmarks in different settings,\nincluding zero-shot, finetuning, and instruction-tuning. We observe\nstate-of-the-art (SoTA) model performance on various code-related tasks, such\nas code generation and completion, math programming, and text-to-code retrieval\ntasks. Particularly, our instruction-tuned CodeT5+ 16B achieves new SoTA\nresults on HumanEval code generation task against other open code LLMs.",
+                    "authors": "Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi D. Q. Bui, Junnan Li, Steven C. H. Hoi",
+                    "published": "2023-05-13",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG",
+                        "cs.PL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.13040v1",
+                    "title": "Retrieval-augmented Multilingual Knowledge Editing",
+                    "abstract": "Knowledge represented in Large Language Models (LLMs) is quite often\nincorrect and can also become obsolete over time. Updating knowledge via\nfine-tuning is computationally resource-hungry and not reliable, and so\nknowledge editing (KE) has developed as an effective and economical alternative\nto inject new knowledge or to fix factual errors in LLMs. Although there has\nbeen considerable interest in this area, current KE research exclusively\nfocuses on the monolingual setting, typically in English. However, what happens\nif the new knowledge is supplied in one language, but we would like to query\nthe LLM in a different language? To address the problem of multilingual\nknowledge editing, we propose Retrieval-augmented Multilingual Knowledge Editor\n(ReMaKE) to update new knowledge in LLMs. ReMaKE can perform model-agnostic\nknowledge editing in multilingual settings. ReMaKE concatenates the new\nknowledge retrieved from a multilingual knowledge base with prompts. Our\nexperimental results show that ReMaKE outperforms baseline knowledge editing\nmethods by a significant margin and is the first KE method to work in a\nmultilingual setting. We provide our multilingual knowledge editing dataset\n(MzsRE) in 12 languages, which along with code, and additional project\ninformation is available at https://github.com/Vicky-Wil/ReMaKE.",
+                    "authors": "Weixuan Wang, Barry Haddow, Alexandra Birch",
+                    "published": "2023-12-20",
+                    "updated": "2023-12-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2210.17323v2",
+                    "title": "GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers",
+                    "abstract": "Generative Pre-trained Transformer models, known as GPT or OPT, set\nthemselves apart through breakthrough performance across complex language\nmodelling tasks, but also by their extremely high computational and storage\ncosts. Specifically, due to their massive size, even inference for large,\nhighly-accurate GPT models may require multiple performant GPUs, which limits\nthe usability of such models. While there is emerging work on relieving this\npressure via model compression, the applicability and performance of existing\ncompression techniques is limited by the scale and complexity of GPT models. In\nthis paper, we address this challenge, and propose GPTQ, a new one-shot weight\nquantization method based on approximate second-order information, that is both\nhighly-accurate and highly-efficient. Specifically, GPTQ can quantize GPT\nmodels with 175 billion parameters in approximately four GPU hours, reducing\nthe bitwidth down to 3 or 4 bits per weight, with negligible accuracy\ndegradation relative to the uncompressed baseline. Our method more than doubles\nthe compression gains relative to previously-proposed one-shot quantization\nmethods, preserving accuracy, allowing us for the first time to execute an 175\nbillion-parameter model inside a single GPU for generative inference. Moreover,\nwe also show that our method can still provide reasonable accuracy in the\nextreme quantization regime, in which weights are quantized to 2-bit or even\nternary quantization levels. We show experimentally that these improvements can\nbe leveraged for end-to-end inference speedups over FP16, of around 3.25x when\nusing high-end GPUs (NVIDIA A100) and 4.5x when using more cost-effective ones\n(NVIDIA A6000). The implementation is available at\nhttps://github.com/IST-DASLab/gptq.",
+                    "authors": "Elias Frantar, Saleh Ashkboos, Torsten Hoefler, Dan Alistarh",
+                    "published": "2022-10-31",
+                    "updated": "2023-03-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13245v3",
+                    "title": "GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints",
+                    "abstract": "Multi-query attention (MQA), which only uses a single key-value head,\ndrastically speeds up decoder inference. However, MQA can lead to quality\ndegradation, and moreover it may not be desirable to train a separate model\njust for faster inference. We (1) propose a recipe for uptraining existing\nmulti-head language model checkpoints into models with MQA using 5% of original\npre-training compute, and (2) introduce grouped-query attention (GQA), a\ngeneralization of multi-query attention which uses an intermediate (more than\none, less than number of query heads) number of key-value heads. We show that\nuptrained GQA achieves quality close to multi-head attention with comparable\nspeed to MQA.",
+                    "authors": "Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebr\u00f3n, Sumit Sanghai",
+                    "published": "2023-05-22",
+                    "updated": "2023-12-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09636v1",
+                    "title": "Dynamic Memory Compression: Retrofitting LLMs for Accelerated Inference",
+                    "abstract": "Transformers have emerged as the backbone of large language models (LLMs).\nHowever, generation remains inefficient due to the need to store in memory a\ncache of key-value representations for past tokens, whose size scales linearly\nwith the input sequence length and batch size. As a solution, we propose\nDynamic Memory Compression (DMC), a method for on-line key-value cache\ncompression at inference time. Most importantly, the model learns to apply\ndifferent compression rates in different heads and layers. We retrofit\npre-trained LLMs such as Llama 2 (7B, 13B and 70B) into DMC Transformers,\nachieving up to ~3.7x throughput increase in auto-regressive inference on a\nNVIDIA H100 GPU. DMC is applied via continued pre-training on a negligible\npercentage of the original data without adding any extra parameters. We find\nthat DMC preserves the original downstream performance with up to 4x cache\ncompression, outperforming up-trained grouped-query attention (GQA). GQA and\nDMC can be even combined to obtain compounded gains. As a result DMC fits\nlonger contexts and larger batches within any given memory budget.",
+                    "authors": "Piotr Nawrot, Adrian \u0141a\u0144cucki, Marcin Chochowski, David Tarjan, Edoardo M. Ponti",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2205.14135v2",
+                    "title": "FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness",
+                    "abstract": "Transformers are slow and memory-hungry on long sequences, since the time and\nmemory complexity of self-attention are quadratic in sequence length.\nApproximate attention methods have attempted to address this problem by trading\noff model quality to reduce the compute complexity, but often do not achieve\nwall-clock speedup. We argue that a missing principle is making attention\nalgorithms IO-aware -- accounting for reads and writes between levels of GPU\nmemory. We propose FlashAttention, an IO-aware exact attention algorithm that\nuses tiling to reduce the number of memory reads/writes between GPU high\nbandwidth memory (HBM) and GPU on-chip SRAM. We analyze the IO complexity of\nFlashAttention, showing that it requires fewer HBM accesses than standard\nattention, and is optimal for a range of SRAM sizes. We also extend\nFlashAttention to block-sparse attention, yielding an approximate attention\nalgorithm that is faster than any existing approximate attention method.\nFlashAttention trains Transformers faster than existing baselines: 15%\nend-to-end wall-clock speedup on BERT-large (seq. length 512) compared to the\nMLPerf 1.1 training speed record, 3$\\times$ speedup on GPT-2 (seq. length 1K),\nand 2.4$\\times$ speedup on long-range arena (seq. length 1K-4K). FlashAttention\nand block-sparse FlashAttention enable longer context in Transformers, yielding\nhigher quality models (0.7 better perplexity on GPT-2 and 6.4 points of lift on\nlong-document classification) and entirely new capabilities: the first\nTransformers to achieve better-than-chance performance on the Path-X challenge\n(seq. length 16K, 61.4% accuracy) and Path-256 (seq. length 64K, 63.1%\naccuracy).",
+                    "authors": "Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher R\u00e9",
+                    "published": "2022-05-27",
+                    "updated": "2022-06-23",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.09906v2",
+                    "title": "Generative Representational Instruction Tuning",
+                    "abstract": "All text-based language problems can be reduced to either generation or\nembedding. Current models only perform well at one or the other. We introduce\ngenerative representational instruction tuning (GRIT) whereby a large language\nmodel is trained to handle both generative and embedding tasks by\ndistinguishing between them through instructions. Compared to other open\nmodels, our resulting GritLM 7B sets a new state of the art on the Massive Text\nEmbedding Benchmark (MTEB) and outperforms all models up to its size on a range\nof generative tasks. By scaling up further, GritLM 8x7B outperforms all open\ngenerative language models that we tried while still being among the best\nembedding models. Notably, we find that GRIT matches training on only\ngenerative or embedding data, thus we can unify both at no performance loss.\nAmong other benefits, the unification via GRIT speeds up Retrieval-Augmented\nGeneration (RAG) by > 60% for long documents, by no longer requiring separate\nretrieval and generation models. Models, code, etc. are freely available at\nhttps://github.com/ContextualAI/gritlm.",
+                    "authors": "Niklas Muennighoff, Hongjin Su, Liang Wang, Nan Yang, Furu Wei, Tao Yu, Amanpreet Singh, Douwe Kiela",
+                    "published": "2024-02-15",
+                    "updated": "2024-04-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.06056v1",
+                    "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+                    "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+                    "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+                    "published": "2023-12-11",
+                    "updated": "2023-12-11",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11406v2",
+                    "title": "Don't Go To Extremes: Revealing the Excessive Sensitivity and Calibration Limitations of LLMs in Implicit Hate Speech Detection",
+                    "abstract": "The fairness and trustworthiness of Large Language Models (LLMs) are\nreceiving increasing attention. Implicit hate speech, which employs indirect\nlanguage to convey hateful intentions, occupies a significant portion of\npractice. However, the extent to which LLMs effectively address this issue\nremains insufficiently examined. This paper delves into the capability of LLMs\nto detect implicit hate speech (Classification Task) and express confidence in\ntheir responses (Calibration Task). Our evaluation meticulously considers\nvarious prompt patterns and mainstream uncertainty estimation methods. Our\nfindings highlight that LLMs exhibit two extremes: (1) LLMs display excessive\nsensitivity towards groups or topics that may cause fairness issues, resulting\nin misclassifying benign statements as hate speech. (2) LLMs' confidence scores\nfor each method excessively concentrate on a fixed range, remaining unchanged\nregardless of the dataset's complexity. Consequently, the calibration\nperformance is heavily reliant on primary classification accuracy. These\ndiscoveries unveil new limitations of LLMs, underscoring the need for caution\nwhen optimizing models to ensure they do not veer towards extremes. This serves\nas a reminder to carefully consider sensitivity and confidence in the pursuit\nof model fairness.",
+                    "authors": "Min Zhang, Jianfeng He, Taoran Ji, Chang-Tien Lu",
+                    "published": "2024-02-18",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2303.01248v3",
+                    "title": "Can ChatGPT Assess Human Personalities? A General Evaluation Framework",
+                    "abstract": "Large Language Models (LLMs) especially ChatGPT have produced impressive\nresults in various areas, but their potential human-like psychology is still\nlargely unexplored. Existing works study the virtual personalities of LLMs but\nrarely explore the possibility of analyzing human personalities via LLMs. This\npaper presents a generic evaluation framework for LLMs to assess human\npersonalities based on Myers Briggs Type Indicator (MBTI) tests. Specifically,\nwe first devise unbiased prompts by randomly permuting options in MBTI\nquestions and adopt the average testing result to encourage more impartial\nanswer generation. Then, we propose to replace the subject in question\nstatements to enable flexible queries and assessments on different subjects\nfrom LLMs. Finally, we re-formulate the question instructions in a manner of\ncorrectness evaluation to facilitate LLMs to generate clearer responses. The\nproposed framework enables LLMs to flexibly assess personalities of different\ngroups of people. We further propose three evaluation metrics to measure the\nconsistency, robustness, and fairness of assessment results from\nstate-of-the-art LLMs including ChatGPT and GPT-4. Our experiments reveal\nChatGPT's ability to assess human personalities, and the average results\ndemonstrate that it can achieve more consistent and fairer assessments in spite\nof lower robustness against prompt biases compared with InstructGPT.",
+                    "authors": "Haocong Rao, Cyril Leung, Chunyan Miao",
+                    "published": "2023-03-01",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.13840v1",
+                    "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+                    "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+                    "published": "2024-03-15",
+                    "updated": "2024-03-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.SI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.08517v1",
+                    "title": "Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forward",
+                    "abstract": "While Large Language Models (LLMs) have seen widespread applications across\nnumerous fields, their limited interpretability poses concerns regarding their\nsafe operations from multiple aspects, e.g., truthfulness, robustness, and\nfairness. Recent research has started developing quality assurance methods for\nLLMs, introducing techniques such as offline detector-based or uncertainty\nestimation methods. However, these approaches predominantly concentrate on\npost-generation analysis, leaving the online safety analysis for LLMs during\nthe generation phase an unexplored area. To bridge this gap, we conduct in this\nwork a comprehensive evaluation of the effectiveness of existing online safety\nanalysis methods on LLMs. We begin with a pilot study that validates the\nfeasibility of detecting unsafe outputs in the early generation process.\nFollowing this, we establish the first publicly available benchmark of online\nsafety analysis for LLMs, including a broad spectrum of methods, models, tasks,\ndatasets, and evaluation metrics. Utilizing this benchmark, we extensively\nanalyze the performance of state-of-the-art online safety analysis methods on\nboth open-source and closed-source LLMs. This analysis reveals the strengths\nand weaknesses of individual methods and offers valuable insights into\nselecting the most appropriate method based on specific application scenarios\nand task requirements. Furthermore, we also explore the potential of using\nhybridization methods, i.e., combining multiple methods to derive a collective\nsafety conclusion, to enhance the efficacy of online safety analysis for LLMs.\nOur findings indicate a promising direction for the development of innovative\nand trustworthy quality assurance methodologies for LLMs, facilitating their\nreliable deployments across diverse domains.",
+                    "authors": "Xuan Xie, Jiayang Song, Zhehua Zhou, Yuheng Huang, Da Song, Lei Ma",
+                    "published": "2024-04-12",
+                    "updated": "2024-04-12",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.04057v1",
+                    "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+                    "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+                    "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+                    "published": "2024-01-08",
+                    "updated": "2024-01-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05374v2",
+                    "title": "Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment",
+                    "abstract": "Ensuring alignment, which refers to making models behave in accordance with\nhuman intentions [1,2], has become a critical task before deploying large\nlanguage models (LLMs) in real-world applications. For instance, OpenAI devoted\nsix months to iteratively aligning GPT-4 before its release [3]. However, a\nmajor challenge faced by practitioners is the lack of clear guidance on\nevaluating whether LLM outputs align with social norms, values, and\nregulations. This obstacle hinders systematic iteration and deployment of LLMs.\nTo address this issue, this paper presents a comprehensive survey of key\ndimensions that are crucial to consider when assessing LLM trustworthiness. The\nsurvey covers seven major categories of LLM trustworthiness: reliability,\nsafety, fairness, resistance to misuse, explainability and reasoning, adherence\nto social norms, and robustness. Each major category is further divided into\nseveral sub-categories, resulting in a total of 29 sub-categories.\nAdditionally, a subset of 8 sub-categories is selected for further\ninvestigation, where corresponding measurement studies are designed and\nconducted on several widely-used LLMs. The measurement results indicate that,\nin general, more aligned models tend to perform better in terms of overall\ntrustworthiness. However, the effectiveness of alignment varies across the\ndifferent trustworthiness categories considered. This highlights the importance\nof conducting more fine-grained analyses, testing, and making continuous\nimprovements on LLM alignment. By shedding light on these key dimensions of LLM\ntrustworthiness, this paper aims to provide valuable insights and guidance to\npractitioners in the field. Understanding and addressing these concerns will be\ncrucial in achieving reliable and ethically sound deployment of LLMs in various\napplications.",
+                    "authors": "Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, Hang Li",
+                    "published": "2023-08-10",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.11483v1",
+                    "title": "Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in\nvarious NLP tasks. However, previous works have shown these models are\nsensitive towards prompt wording, and few-shot demonstrations and their order,\nposing challenges to fair assessment of these models. As these models become\nmore powerful, it becomes imperative to understand and address these\nlimitations. In this paper, we focus on LLMs robustness on the task of\nmultiple-choice questions -- commonly adopted task to study reasoning and\nfact-retrieving capability of LLMs. Investigating the sensitivity of LLMs\ntowards the order of options in multiple-choice questions, we demonstrate a\nconsiderable performance gap of approximately 13% to 75% in LLMs on different\nbenchmarks, when answer options are reordered, even when using demonstrations\nin a few-shot setting. Through a detailed analysis, we conjecture that this\nsensitivity arises when LLMs are uncertain about the prediction between the\ntop-2/3 choices, and specific options placements may favor certain prediction\nbetween those top choices depending on the question caused by positional bias.\nWe also identify patterns in top-2 choices that amplify or mitigate the model's\nbias toward option placement. We found that for amplifying bias, the optimal\nstrategy involves positioning the top two choices as the first and last\noptions. Conversely, to mitigate bias, we recommend placing these choices among\nthe adjacent options. To validate our conjecture, we conduct various\nexperiments and adopt two approaches to calibrate LLMs' predictions, leading to\nup to 8 percentage points improvement across different models and benchmarks.",
+                    "authors": "Pouya Pezeshkpour, Estevam Hruschka",
+                    "published": "2023-08-22",
+                    "updated": "2023-08-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09606v1",
+                    "title": "Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey",
+                    "abstract": "Causal inference has shown potential in enhancing the predictive accuracy,\nfairness, robustness, and explainability of Natural Language Processing (NLP)\nmodels by capturing causal relationships among variables. The emergence of\ngenerative Large Language Models (LLMs) has significantly impacted various NLP\ndomains, particularly through their advanced reasoning capabilities. This\nsurvey focuses on evaluating and improving LLMs from a causal view in the\nfollowing areas: understanding and improving the LLMs' reasoning capacity,\naddressing fairness and safety issues in LLMs, complementing LLMs with\nexplanations, and handling multimodality. Meanwhile, LLMs' strong reasoning\ncapacities can in turn contribute to the field of causal inference by aiding\ncausal relationship discovery and causal effect estimations. This review\nexplores the interplay between causal inference frameworks and LLMs from both\nperspectives, emphasizing their collective potential to further the development\nof more advanced and equitable artificial intelligence systems.",
+                    "authors": "Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.05668v1",
+                    "title": "CFaiRLLM: Consumer Fairness Evaluation in Large-Language Model Recommender System",
+                    "abstract": "In the evolving landscape of recommender systems, the integration of Large\nLanguage Models (LLMs) such as ChatGPT marks a new era, introducing the concept\nof Recommendation via LLM (RecLLM). While these advancements promise\nunprecedented personalization and efficiency, they also bring to the fore\ncritical concerns regarding fairness, particularly in how recommendations might\ninadvertently perpetuate or amplify biases associated with sensitive user\nattributes. In order to address these concerns, our study introduces a\ncomprehensive evaluation framework, CFaiRLLM, aimed at evaluating (and thereby\nmitigating) biases on the consumer side within RecLLMs.\n  Our research methodically assesses the fairness of RecLLMs by examining how\nrecommendations might vary with the inclusion of sensitive attributes such as\ngender, age, and their intersections, through both similarity alignment and\ntrue preference alignment. By analyzing recommendations generated under\ndifferent conditions-including the use of sensitive attributes in user\nprompts-our framework identifies potential biases in the recommendations\nprovided. A key part of our study involves exploring how different detailed\nstrategies for constructing user profiles (random, top-rated, recent) impact\nthe alignment between recommendations made without consideration of sensitive\nattributes and those that are sensitive-attribute-aware, highlighting the bias\nmechanisms within RecLLMs.\n  The findings in our study highlight notable disparities in the fairness of\nrecommendations, particularly when sensitive attributes are integrated into the\nrecommendation process, either individually or in combination. The analysis\ndemonstrates that the choice of user profile sampling strategy plays a\nsignificant role in affecting fairness outcomes, highlighting the complexity of\nachieving fair recommendations in the era of LLMs.",
+                    "authors": "Yashar Deldjoo, Tommaso di Noia",
+                    "published": "2024-03-08",
+                    "updated": "2024-03-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.12736v1",
+                    "title": "Large Language Model Supply Chain: A Research Agenda",
+                    "abstract": "The rapid advancements in pre-trained Large Language Models (LLMs) and Large\nMultimodal Models (LMMs) have ushered in a new era of intelligent applications,\ntransforming fields ranging from natural language processing to content\ngeneration. The LLM supply chain represents a crucial aspect of the\ncontemporary artificial intelligence landscape. It encompasses the entire\nlifecycle of pre-trained models, from its initial development and training to\nits final deployment and application in various domains. This paper presents a\ncomprehensive overview of the LLM supply chain, highlighting its three core\nelements: 1) the model infrastructure, encompassing datasets and toolchain for\ntraining, optimization, and deployment; 2) the model lifecycle, covering\ntraining, testing, releasing, and ongoing maintenance; and 3) the downstream\napplication ecosystem, enabling the integration of pre-trained models into a\nwide range of intelligent applications. However, this rapidly evolving field\nfaces numerous challenges across these key components, including data privacy\nand security, model interpretability and fairness, infrastructure scalability,\nand regulatory compliance. Addressing these challenges is essential for\nharnessing the full potential of LLMs and ensuring their ethical and\nresponsible use. This paper provides a future research agenda for the LLM\nsupply chain, aiming at driving the continued advancement and responsible\ndeployment of these transformative LLMs.",
+                    "authors": "Shenao Wang, Yanjie Zhao, Xinyi Hou, Haoyu Wang",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.03852v2",
+                    "title": "FLM-101B: An Open LLM and How to Train It with $100K Budget",
+                    "abstract": "Large language models (LLMs) have achieved remarkable success in NLP and\nmultimodal tasks, among others. Despite these successes, two main challenges\nremain in developing LLMs: (i) high computational cost, and (ii) fair and\nobjective evaluations. In this paper, we report a solution to significantly\nreduce LLM training cost through a growth strategy. We demonstrate that a\n101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US\ndollars. Inspired by IQ tests, we also consolidate an additional range of\nevaluations on top of existing evaluations that focus on knowledge-oriented\nabilities. These IQ evaluations include symbolic mapping, rule understanding,\npattern mining, and anti-interference. Such evaluations minimize the potential\nimpact of memorization. Experimental results show that our model, named\nFLM-101B, trained with a budget of 100K US dollars, achieves performance\ncomparable to powerful and well-known models, e.g., GPT-3 and GLM-130B,\nespecially on the additional range of IQ evaluations. The checkpoint of\nFLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.",
+                    "authors": "Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Xuying Meng, Siqi Fan, Peng Han, Jing Li, Li Du, Bowen Qin, Zheng Zhang, Aixin Sun, Yequan Wang",
+                    "published": "2023-09-07",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.09447v2",
+                    "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+                    "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+                    "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+                    "published": "2023-11-15",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.18569v1",
+                    "title": "Fairness of ChatGPT",
+                    "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+                    "authors": "Yunqi Li, Yongfeng Zhang",
+                    "published": "2023-05-22",
+                    "updated": "2023-05-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.17553v1",
+                    "title": "RuBia: A Russian Language Bias Detection Dataset",
+                    "abstract": "Warning: this work contains upsetting or disturbing content.\n  Large language models (LLMs) tend to learn the social and cultural biases\npresent in the raw pre-training data. To test if an LLM's behavior is fair,\nfunctional datasets are employed, and due to their purpose, these datasets are\nhighly language and culture-specific. In this paper, we address a gap in the\nscope of multilingual bias evaluation by presenting a bias detection dataset\nspecifically designed for the Russian language, dubbed as RuBia. The RuBia\ndataset is divided into 4 domains: gender, nationality, socio-economic status,\nand diverse, each of the domains is further divided into multiple fine-grained\nsubdomains. Every example in the dataset consists of two sentences with the\nfirst reinforcing a potentially harmful stereotype or trope and the second\ncontradicting it. These sentence pairs were first written by volunteers and\nthen validated by native-speaking crowdsourcing workers. Overall, there are\nnearly 2,000 unique sentence pairs spread over 19 subdomains in RuBia. To\nillustrate the dataset's purpose, we conduct a diagnostic evaluation of\nstate-of-the-art or near-state-of-the-art LLMs and discuss the LLMs'\npredisposition to social biases.",
+                    "authors": "Veronika Grigoreva, Anastasiia Ivanova, Ilseyar Alimova, Ekaterina Artemova",
+                    "published": "2024-03-26",
+                    "updated": "2024-03-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10397v2",
+                    "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+                    "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+                    "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+                    "published": "2023-08-21",
+                    "updated": "2023-10-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10149v2",
+                    "title": "A Survey on Fairness in Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have shown powerful performance and development\nprospects and are widely deployed in the real world. However, LLMs can capture\nsocial biases from unprocessed training data and propagate the biases to\ndownstream tasks. Unfair LLM systems have undesirable social impacts and\npotential harms. In this paper, we provide a comprehensive review of related\nresearch on fairness in LLMs. Considering the influence of parameter magnitude\nand training paradigm on research strategy, we divide existing fairness\nresearch into oriented to medium-sized LLMs under pre-training and fine-tuning\nparadigms and oriented to large-sized LLMs under prompting paradigms. First,\nfor medium-sized LLMs, we introduce evaluation metrics and debiasing methods\nfrom the perspectives of intrinsic bias and extrinsic bias, respectively. Then,\nfor large-sized LLMs, we introduce recent fairness research, including fairness\nevaluation, reasons for bias, and debiasing methods. Finally, we discuss and\nprovide insight on the challenges and future directions for the development of\nfairness in LLMs.",
+                    "authors": "Yingji Li, Mengnan Du, Rui Song, Xin Wang, Ying Wang",
+                    "published": "2023-08-20",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00306v1",
+                    "title": "Probing Explicit and Implicit Gender Bias through LLM Conditional Text Generation",
+                    "abstract": "Large Language Models (LLMs) can generate biased and toxic responses. Yet\nmost prior work on LLM gender bias evaluation requires predefined\ngender-related phrases or gender stereotypes, which are challenging to be\ncomprehensively collected and are limited to explicit bias evaluation. In\naddition, we believe that instances devoid of gender-related language or\nexplicit stereotypes in inputs can still induce gender bias in LLMs. Thus, in\nthis work, we propose a conditional text generation mechanism without the need\nfor predefined gender phrases and stereotypes. This approach employs three\ntypes of inputs generated through three distinct strategies to probe LLMs,\naiming to show evidence of explicit and implicit gender biases in LLMs. We also\nutilize explicit and implicit evaluation metrics to evaluate gender bias in\nLLMs under different strategies. Our experiments demonstrate that an increased\nmodel size does not consistently lead to enhanced fairness and all tested LLMs\nexhibit explicit and/or implicit gender bias, even when explicit gender\nstereotypes are absent in the inputs.",
+                    "authors": "Xiangjue Dong, Yibo Wang, Philip S. Yu, James Caverlee",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.11761v1",
+                    "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+                    "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2023-07-14",
+                    "updated": "2023-07-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00625v2",
+                    "title": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models",
+                    "abstract": "The burgeoning field of Large Language Models (LLMs), exemplified by\nsophisticated models like OpenAI's ChatGPT, represents a significant\nadvancement in artificial intelligence. These models, however, bring forth\nsubstantial challenges in the high consumption of computational, memory,\nenergy, and financial resources, especially in environments with limited\nresource capabilities. This survey aims to systematically address these\nchallenges by reviewing a broad spectrum of techniques designed to enhance the\nresource efficiency of LLMs. We categorize methods based on their optimization\nfocus: computational, memory, energy, financial, and network resources and\ntheir applicability across various stages of an LLM's lifecycle, including\narchitecture design, pretraining, finetuning, and system design. Additionally,\nthe survey introduces a nuanced categorization of resource efficiency\ntechniques by their specific resource types, which uncovers the intricate\nrelationships and mappings between various resources and corresponding\noptimization techniques. A standardized set of evaluation metrics and datasets\nis also presented to facilitate consistent and fair comparisons across\ndifferent models and techniques. By offering a comprehensive overview of the\ncurrent sota and identifying open research avenues, this survey serves as a\nfoundational reference for researchers and practitioners, aiding them in\ndeveloping more sustainable and efficient LLMs in a rapidly evolving landscape.",
+                    "authors": "Guangji Bai, Zheng Chai, Chen Ling, Shiyu Wang, Jiaying Lu, Nan Zhang, Tingwei Shi, Ziyang Yu, Mengdan Zhu, Yifei Zhang, Carl Yang, Yue Cheng, Liang Zhao",
+                    "published": "2024-01-01",
+                    "updated": "2024-01-04",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00811v1",
+                    "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+                    "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+                    "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+                    "published": "2024-02-25",
+                    "updated": "2024-02-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.15585v1",
+                    "title": "Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting",
+                    "abstract": "There exist both scalable tasks, like reading comprehension and\nfact-checking, where model performance improves with model size, and unscalable\ntasks, like arithmetic reasoning and symbolic reasoning, where model\nperformance does not necessarily improve with model size. Large language models\n(LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate\nincremental predictions even on unscalable tasks. Unfortunately, despite their\nexceptional reasoning abilities, LLMs tend to internalize and reproduce\ndiscriminatory societal biases. Whether CoT can provide discriminatory or\negalitarian rationalizations for the implicit information in unscalable tasks\nremains an open question.\n  In this study, we examine the impact of LLMs' step-by-step predictions on\ngender bias in unscalable tasks. For this purpose, we construct a benchmark for\nan unscalable task where the LLM is given a list of words comprising feminine,\nmasculine, and gendered occupational words, and is required to count the number\nof feminine and masculine words. In our CoT prompts, we require the LLM to\nexplicitly indicate whether each word in the word list is a feminine or\nmasculine before making the final predictions. With counting and handling the\nmeaning of words, this benchmark has characteristics of both arithmetic\nreasoning and symbolic reasoning. Experimental results in English show that\nwithout step-by-step prediction, most LLMs make socially biased predictions,\ndespite the task being as simple as counting words. Interestingly, CoT\nprompting reduces this unconscious social bias in LLMs and encourages fair\npredictions.",
+                    "authors": "Masahiro Kaneko, Danushka Bollegala, Naoaki Okazaki, Timothy Baldwin",
+                    "published": "2024-01-28",
+                    "updated": "2024-01-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00588v1",
+                    "title": "Fairness in Serving Large Language Models",
+                    "abstract": "High-demand LLM inference services (e.g., ChatGPT and BARD) support a wide\nrange of requests from short chat conversations to long document reading. To\nensure that all client requests are processed fairly, most major LLM inference\nservices have request rate limits, to ensure that no client can dominate the\nrequest queue. However, this rudimentary notion of fairness also results in\nunder-utilization of the resources and poor client experience when there is\nspare capacity. While there is a rich literature on fair scheduling, serving\nLLMs presents new challenges due to their unpredictable request lengths and\ntheir unique batching characteristics on parallel accelerators. This paper\nintroduces the definition of LLM serving fairness based on a cost function that\naccounts for the number of input and output tokens processed. To achieve\nfairness in serving, we propose a novel scheduling algorithm, the Virtual Token\nCounter (VTC), a fair scheduler based on the continuous batching mechanism. We\nprove a 2x tight upper bound on the service difference between two backlogged\nclients, adhering to the requirement of work-conserving. Through extensive\nexperiments, we demonstrate the superior performance of VTC in ensuring\nfairness, especially in contrast to other baseline methods, which exhibit\nshortcomings under various conditions.",
+                    "authors": "Ying Sheng, Shiyi Cao, Dacheng Li, Banghua Zhu, Zhuohan Li, Danyang Zhuo, Joseph E. Gonzalez, Ion Stoica",
+                    "published": "2023-12-31",
+                    "updated": "2023-12-31",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.PF"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.06899v4",
+                    "title": "Flames: Benchmarking Value Alignment of LLMs in Chinese",
+                    "abstract": "The widespread adoption of large language models (LLMs) across various\nregions underscores the urgent need to evaluate their alignment with human\nvalues. Current benchmarks, however, fall short of effectively uncovering\nsafety vulnerabilities in LLMs. Despite numerous models achieving high scores\nand 'topping the chart' in these evaluations, there is still a significant gap\nin LLMs' deeper alignment with human values and achieving genuine harmlessness.\nTo this end, this paper proposes a value alignment benchmark named Flames,\nwhich encompasses both common harmlessness principles and a unique morality\ndimension that integrates specific Chinese values such as harmony. Accordingly,\nwe carefully design adversarial prompts that incorporate complex scenarios and\njailbreaking methods, mostly with implicit malice. By prompting 17 mainstream\nLLMs, we obtain model responses and rigorously annotate them for detailed\nevaluation. Our findings indicate that all the evaluated LLMs demonstrate\nrelatively poor performance on Flames, particularly in the safety and fairness\ndimensions. We also develop a lightweight specified scorer capable of scoring\nLLMs across multiple dimensions to efficiently evaluate new models on the\nbenchmark. The complexity of Flames has far exceeded existing benchmarks,\nsetting a new challenge for contemporary LLMs and highlighting the need for\nfurther alignment of LLMs. Our benchmark is publicly available at\nhttps://github.com/AIFlames/Flames.",
+                    "authors": "Kexin Huang, Xiangyang Liu, Qianyu Guo, Tianxiang Sun, Jiawei Sun, Yaru Wang, Zeyang Zhou, Yixu Wang, Yan Teng, Xipeng Qiu, Yingchun Wang, Dahua Lin",
+                    "published": "2023-11-12",
+                    "updated": "2024-04-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2304.03728v1",
+                    "title": "Interpretable Unified Language Checking",
+                    "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+                    "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+                    "published": "2023-04-07",
+                    "updated": "2023-04-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18580v1",
+                    "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+                    "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+                    "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+                    "published": "2023-11-30",
+                    "updated": "2023-11-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.14607v2",
+                    "title": "Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications",
+                    "abstract": "Recent literature has suggested the potential of using large language models\n(LLMs) to make classifications for tabular tasks. However, LLMs have been shown\nto exhibit harmful social biases that reflect the stereotypes and inequalities\npresent in society. To this end, as well as the widespread use of tabular data\nin many high-stake applications, it is important to explore the following\nquestions: what sources of information do LLMs draw upon when making\nclassifications for tabular tasks; whether and to what extent are LLM\nclassifications for tabular data influenced by social biases and stereotypes;\nand what are the consequential implications for fairness?\n  Through a series of experiments, we delve into these questions and show that\nLLMs tend to inherit social biases from their training data which significantly\nimpact their fairness in tabular classification tasks. Furthermore, our\ninvestigations show that in the context of bias mitigation, though in-context\nlearning and finetuning have a moderate effect, the fairness metric gap between\ndifferent subgroups is still larger than that in traditional machine learning\nmodels, such as Random Forest and shallow Neural Networks. This observation\nemphasizes that the social biases are inherent within the LLMs themselves and\ninherited from their pretraining corpus, not only from the downstream task\ndatasets. Besides, we demonstrate that label-flipping of in-context examples\ncan significantly reduce biases, further highlighting the presence of inherent\nbias within LLMs.",
+                    "authors": "Yanchen Liu, Srishti Gautam, Jiaqi Ma, Himabindu Lakkaraju",
+                    "published": "2023-10-23",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.07688v1",
+                    "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+                    "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+                    "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+                    "published": "2024-02-12",
+                    "updated": "2024-02-12",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15398v1",
+                    "title": "Fairness-Aware Structured Pruning in Transformers",
+                    "abstract": "The increasing size of large language models (LLMs) has introduced challenges\nin their training and inference. Removing model components is perceived as a\nsolution to tackle the large model sizes, however, existing pruning methods\nsolely focus on performance, without considering an essential aspect for the\nresponsible use of LLMs: model fairness. It is crucial to address the fairness\nof LLMs towards diverse groups, such as women, Black people, LGBTQ+, Jewish\ncommunities, among others, as they are being deployed and available to a wide\naudience. In this work, first, we investigate how attention heads impact\nfairness and performance in pre-trained transformer-based language models. We\nthen propose a novel method to prune the attention heads that negatively impact\nfairness while retaining the heads critical for performance, i.e. language\nmodeling capabilities. Our approach is practical in terms of time and\nresources, as it does not require fine-tuning the final pruned, and fairer,\nmodel. Our findings demonstrate a reduction in gender bias by 19%, 19.5%,\n39.5%, 34.7%, 23%, and 8% for DistilGPT-2, GPT-2, GPT-Neo of two different\nsizes, GPT-J, and Llama 2 models, respectively, in comparison to the biased\nmodel, with only a slight decrease in performance.",
+                    "authors": "Abdelrahman Zayed, Goncalo Mordido, Samira Shabanian, Ioana Baldini, Sarath Chandar",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02294v1",
+                    "title": "LLMs grasp morality in concept",
+                    "abstract": "Work in AI ethics and fairness has made much progress in regulating LLMs to\nreflect certain values, such as fairness, truth, and diversity. However, it has\ntaken the problem of how LLMs might 'mean' anything at all for granted. Without\naddressing this, it is not clear what imbuing LLMs with such values even means.\nIn response, we provide a general theory of meaning that extends beyond humans.\nWe use this theory to explicate the precise nature of LLMs as meaning-agents.\nWe suggest that the LLM, by virtue of its position as a meaning-agent, already\ngrasps the constructions of human society (e.g. morality, gender, and race) in\nconcept. Consequently, under certain ethical frameworks, currently popular\nmethods for model alignment are limited at best and counterproductive at worst.\nMoreover, unaligned models may help us better develop our moral and social\nphilosophy.",
+                    "authors": "Mark Pock, Andre Ye, Jared Moore",
+                    "published": "2023-11-04",
+                    "updated": "2023-11-04",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15478v1",
+                    "title": "A Group Fairness Lens for Large Language Models",
+                    "abstract": "The rapid advancement of large language models has revolutionized various\napplications but also raised crucial concerns about their potential to\nperpetuate biases and unfairness when deployed in social media contexts.\nEvaluating LLMs' potential biases and fairness has become crucial, as existing\nmethods rely on limited prompts focusing on just a few groups, lacking a\ncomprehensive categorical perspective. In this paper, we propose evaluating LLM\nbiases from a group fairness lens using a novel hierarchical schema\ncharacterizing diverse social groups. Specifically, we construct a dataset,\nGFair, encapsulating target-attribute combinations across multiple dimensions.\nIn addition, we introduce statement organization, a new open-ended text\ngeneration task, to uncover complex biases in LLMs. Extensive evaluations of\npopular LLMs reveal inherent safety concerns. To mitigate the biases of LLM\nfrom a group fairness perspective, we pioneer a novel chain-of-thought method\nGF-Think to mitigate biases of LLMs from a group fairness perspective.\nExperimental results demonstrate its efficacy in mitigating bias in LLMs to\nachieve fairness.",
+                    "authors": "Guanqun Bi, Lei Shen, Yuqiang Xie, Yanan Cao, Tiangang Zhu, Xiaodong He",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.13862v2",
+                    "title": "A Trip Towards Fairness: Bias and De-Biasing in Large Language Models",
+                    "abstract": "Cheap-to-Build Very Large-Language Models (CtB-LLMs) with affordable training\nare emerging as the next big revolution in natural language processing and\nunderstanding. These CtB-LLMs are democratizing access to trainable Very\nLarge-Language Models (VLLMs) and, thus, may represent the building blocks of\nmany NLP systems solving downstream tasks. Hence, a little or a large bias in\nCtB-LLMs may cause huge harm. In this paper, we performed a large investigation\nof the bias of three families of CtB-LLMs, and we showed that debiasing\ntechniques are effective and usable. Indeed, according to current tests, the\nLLaMA and the OPT families have an important bias in gender, race, religion,\nand profession. In contrast to the analysis for other LLMs, we discovered that\nbias depends not on the number of parameters but on the perplexity. Finally,\nthe debiasing of OPT using LoRA reduces bias up to 4.12 points in the\nnormalized stereotype score.",
+                    "authors": "Leonardo Ranaldi, Elena Sofia Ruzzetti, Davide Venditti, Dario Onorati, Fabio Massimo Zanzotto",
+                    "published": "2023-05-23",
+                    "updated": "2023-08-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.12150v1",
+                    "title": "Your Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One",
+                    "abstract": "The widespread adoption of large language models (LLMs) underscores the\nurgent need to ensure their fairness. However, LLMs frequently present dominant\nviewpoints while ignoring alternative perspectives from minority parties,\nresulting in potential biases. We hypothesize that these fairness-violating\nbehaviors occur because LLMs express their viewpoints using a human personality\nthat represents the majority of training data. In response to this, we validate\nthat prompting LLMs with specific roles can allow LLMs to express diverse\nviewpoints. Building on this insight and observation, we develop FairThinking,\na pipeline designed to automatically generate roles that enable LLMs to\narticulate diverse perspectives for fair expressions. To evaluate FairThinking,\nwe create a dataset with a thousand items covering three fairness-related\ntopics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to\ndemonstrate its superior performance.",
+                    "authors": "Tianlin Li, Xiaoyu Zhang, Chao Du, Tianyu Pang, Qian Liu, Qing Guo, Chao Shen, Yang Liu",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "I.2; J.4"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08472v1",
+                    "title": "Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models",
+                    "abstract": "Recently, work in NLP has shifted to few-shot (in-context) learning, with\nlarge language models (LLMs) performing well across a range of tasks. However,\nwhile fairness evaluations have become a standard for supervised methods,\nlittle is known about the fairness of LLMs as prediction systems. Further,\ncommon standard methods for fairness involve access to models weights or are\napplied during finetuning, which are not applicable in few-shot learning. Do\nLLMs exhibit prediction biases when used for standard NLP tasks? In this work,\nwe explore the effect of shots, which directly affect the performance of\nmodels, on the fairness of LLMs as NLP classification systems. We consider how\ndifferent shot selection strategies, both existing and new demographically\nsensitive methods, affect model fairness across three standard fairness\ndatasets. We discuss how future work can include LLM fairness evaluations.",
+                    "authors": "Carlos Aguirre, Kuleen Sasse, Isabel Cachola, Mark Dredze",
+                    "published": "2023-11-14",
+                    "updated": "2023-11-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.02650v1",
+                    "title": "Towards detecting unanticipated bias in Large Language Models",
+                    "abstract": "Over the last year, Large Language Models (LLMs) like ChatGPT have become\nwidely available and have exhibited fairness issues similar to those in\nprevious machine learning systems. Current research is primarily focused on\nanalyzing and quantifying these biases in training data and their impact on the\ndecisions of these models, alongside developing mitigation strategies. This\nresearch largely targets well-known biases related to gender, race, ethnicity,\nand language. However, it is clear that LLMs are also affected by other, less\nobvious implicit biases. The complex and often opaque nature of these models\nmakes detecting such biases challenging, yet this is crucial due to their\npotential negative impact in various applications. In this paper, we explore\nnew avenues for detecting these unanticipated biases in LLMs, focusing\nspecifically on Uncertainty Quantification and Explainable AI methods. These\napproaches aim to assess the certainty of model decisions and to make the\ninternal decision-making processes of LLMs more transparent, thereby\nidentifying and understanding biases that are not immediately apparent. Through\nthis research, we aim to contribute to the development of fairer and more\ntransparent AI systems.",
+                    "authors": "Anna Kruspe",
+                    "published": "2024-04-03",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.02839v1",
+                    "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+                    "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+                    "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+                    "published": "2024-03-05",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14804v1",
+                    "title": "Use large language models to promote equity",
+                    "abstract": "Advances in large language models (LLMs) have driven an explosion of interest\nabout their societal impacts. Much of the discourse around how they will impact\nsocial equity has been cautionary or negative, focusing on questions like \"how\nmight LLMs be biased and how would we mitigate those biases?\" This is a vital\ndiscussion: the ways in which AI generally, and LLMs specifically, can entrench\nbiases have been well-documented. But equally vital, and much less discussed,\nis the more opportunity-focused counterpoint: \"what promising applications do\nLLMs enable that could promote equity?\" If LLMs are to enable a more equitable\nworld, it is not enough just to play defense against their biases and failure\nmodes. We must also go on offense, applying them positively to equity-enhancing\nuse cases to increase opportunities for underserved groups and reduce societal\ndiscrimination. There are many choices which determine the impact of AI, and a\nfundamental choice very early in the pipeline is the problems we choose to\napply it to. If we focus only later in the pipeline -- making LLMs marginally\nmore fair as they facilitate use cases which intrinsically entrench power -- we\nwill miss an important opportunity to guide them to equitable impacts. Here, we\nhighlight the emerging potential of LLMs to promote equity by presenting four\nnewly possible, promising research directions, while keeping risks and\ncautionary points in clear view.",
+                    "authors": "Emma Pierson, Divya Shanmugam, Rajiv Movva, Jon Kleinberg, Monica Agrawal, Mark Dredze, Kadija Ferryman, Judy Wawira Gichoya, Dan Jurafsky, Pang Wei Koh, Karen Levy, Sendhil Mullainathan, Ziad Obermeyer, Harini Suresh, Keyon Vafa",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-22",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18140v1",
+                    "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+                    "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+                    "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+                    "published": "2023-11-29",
+                    "updated": "2023-11-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18333v3",
+                    "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+                    "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+                    "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+                    "published": "2023-10-20",
+                    "updated": "2023-12-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.01937v1",
+                    "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+                    "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+                    "authors": "Cheng-Han Chiang, Hung-yi Lee",
+                    "published": "2023-05-03",
+                    "updated": "2023-05-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.01769v1",
+                    "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+                    "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+                    "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+                    "published": "2024-05-02",
+                    "updated": "2024-05-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.01349v1",
+                    "title": "Fairness in Large Language Models: A Taxonomic Survey",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+                    "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+                    "published": "2024-03-31",
+                    "updated": "2024-03-31",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.11653v2",
+                    "title": "\"It's a Fair Game\", or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents",
+                    "abstract": "The widespread use of Large Language Model (LLM)-based conversational agents\n(CAs), especially in high-stakes domains, raises many privacy concerns.\nBuilding ethical LLM-based CAs that respect user privacy requires an in-depth\nunderstanding of the privacy risks that concern users the most. However,\nexisting research, primarily model-centered, does not provide insight into\nusers' perspectives. To bridge this gap, we analyzed sensitive disclosures in\nreal-world ChatGPT conversations and conducted semi-structured interviews with\n19 LLM-based CA users. We found that users are constantly faced with trade-offs\nbetween privacy, utility, and convenience when using LLM-based CAs. However,\nusers' erroneous mental models and the dark patterns in system design limited\ntheir awareness and comprehension of the privacy risks. Additionally, the\nhuman-like interactions encouraged more sensitive disclosures, which\ncomplicated users' ability to navigate the trade-offs. We discuss practical\ndesign guidelines and the needs for paradigm shifts to protect the privacy of\nLLM-based CA users.",
+                    "authors": "Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, Tianshi Li",
+                    "published": "2023-09-20",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC",
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.10199v3",
+                    "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+                    "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+                    "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+                    "published": "2024-04-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.02219v1",
+                    "title": "FairEvalLLM. A Comprehensive Framework for Benchmarking Fairness in Large Language Model Recommender Systems",
+                    "abstract": "This paper presents a framework for evaluating fairness in recommender\nsystems powered by Large Language Models (RecLLMs), addressing the need for a\nunified approach that spans various fairness dimensions including sensitivity\nto user attributes, intrinsic fairness, and discussions of fairness based on\nunderlying benefits. In addition, our framework introduces counterfactual\nevaluations and integrates diverse user group considerations to enhance the\ndiscourse on fairness evaluation for RecLLMs.\n  Our key contributions include the development of a robust framework for\nfairness evaluation in LLM-based recommendations and a structured method to\ncreate \\textit{informative user profiles} from demographic data, historical\nuser preferences, and recent interactions. We argue that the latter is\nessential for enhancing personalization in such systems, especially in\ntemporal-driven scenarios. We demonstrate the utility of our framework through\npractical applications on two datasets, LastFM-1K and ML-1M. We conduct\nexperiments on a subsample of 80 users from each dataset, testing and assessing\nthe effectiveness of various prompt construction scenarios and in-context\nlearning, comprising more than 50 scenarios. This results in more than 4000\nrecommendations (80 * 50 = 4000). Our study reveals that while there are no\nsignificant unfairness issues in scenarios involving sensitive attributes, some\nconcerns remain. However, in terms of intrinsic fairness, which does not\ninvolve direct sensitivity, unfairness across demographic groups remains\nsignificant. The code and data used for this paper are available at:\n\\url{https://shorturl.at/awBFM}.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2024-05-03",
+                    "updated": "2024-05-03",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.13095v1",
+                    "title": "Enhancing Logical Reasoning in Large Language Models to Facilitate Legal Applications",
+                    "abstract": "Language serves as a vehicle for conveying thought, enabling communication\namong individuals. The ability to distinguish between diverse concepts,\nidentify fairness and injustice, and comprehend a range of legal notions\nfundamentally relies on logical reasoning. Large Language Models (LLMs) attempt\nto emulate human language understanding and generation, but their competency in\nlogical reasoning remains limited. This paper seeks to address the\nphilosophical question: How can we effectively teach logical reasoning to LLMs\nwhile maintaining a deep understanding of the intricate relationship between\nlanguage and logic? By focusing on bolstering LLMs' capabilities in logical\nreasoning, we aim to expand their applicability in law and other\nlogic-intensive disciplines. To this end, we propose a Reinforcement Learning\nfrom Logical Feedback (RLLF) approach, which serves as a potential framework\nfor refining LLMs' reasoning capacities. Through RLLF and a revised evaluation\nmethodology, we explore new avenues for research in this domain and contribute\nto the development of LLMs capable of handling complex legal reasoning tasks\nwhile acknowledging the fundamental connection between language and logic.",
+                    "authors": "Ha-Thanh Nguyen, Wachara Fungwacharakorn, Ken Satoh",
+                    "published": "2023-11-22",
+                    "updated": "2023-11-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.03514v3",
+                    "title": "Can Large Language Models Transform Computational Social Science?",
+                    "abstract": "Large Language Models (LLMs) are capable of successfully performing many\nlanguage processing tasks zero-shot (without training data). If zero-shot LLMs\ncan also reliably classify and explain social phenomena like persuasiveness and\npolitical ideology, then LLMs could augment the Computational Social Science\n(CSS) pipeline in important ways. This work provides a road map for using LLMs\nas CSS tools. Towards this end, we contribute a set of prompting best practices\nand an extensive evaluation pipeline to measure the zero-shot performance of 13\nlanguage models on 25 representative English CSS benchmarks. On taxonomic\nlabeling tasks (classification), LLMs fail to outperform the best fine-tuned\nmodels but still achieve fair levels of agreement with humans. On free-form\ncoding tasks (generation), LLMs produce explanations that often exceed the\nquality of crowdworkers' gold references. We conclude that the performance of\ntoday's LLMs can augment the CSS research pipeline in two ways: (1) serving as\nzero-shot data annotators on human annotation teams, and (2) bootstrapping\nchallenging creative generation tasks (e.g., explaining the underlying\nattributes of a text). In summary, LLMs are posed to meaningfully participate\nin social science analysis in partnership with humans.",
+                    "authors": "Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, Diyi Yang",
+                    "published": "2023-04-12",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.15215v1",
+                    "title": "Item-side Fairness of Large Language Model-based Recommendation System",
+                    "abstract": "Recommendation systems for Web content distribution intricately connect to\nthe information access and exposure opportunities for vulnerable populations.\nThe emergence of Large Language Models-based Recommendation System (LRS) may\nintroduce additional societal challenges to recommendation systems due to the\ninherent biases in Large Language Models (LLMs). From the perspective of\nitem-side fairness, there remains a lack of comprehensive investigation into\nthe item-side fairness of LRS given the unique characteristics of LRS compared\nto conventional recommendation systems. To bridge this gap, this study examines\nthe property of LRS with respect to item-side fairness and reveals the\ninfluencing factors of both historical users' interactions and inherent\nsemantic biases of LLMs, shedding light on the need to extend conventional\nitem-side fairness methods for LRS. Towards this goal, we develop a concise and\neffective framework called IFairLRS to enhance the item-side fairness of an\nLRS. IFairLRS covers the main stages of building an LRS with specifically\nadapted strategies to calibrate the recommendations of LRS. We utilize IFairLRS\nto fine-tune LLaMA, a representative LLM, on \\textit{MovieLens} and\n\\textit{Steam} datasets, and observe significant item-side fairness\nimprovements. The code can be found in\nhttps://github.com/JiangM-C/IFairLRS.git.",
+                    "authors": "Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, Xiangnan He",
+                    "published": "2024-02-23",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.05694v1",
+                    "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+                    "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+                    "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+                    "published": "2023-10-09",
+                    "updated": "2023-10-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00884v2",
+                    "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+                    "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+                    "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+                    "published": "2024-03-01",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.DB",
+                    "cats": [
+                        "cs.DB",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11764v1",
+                    "title": "ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs",
+                    "abstract": "Large Language models (LLMs), while powerful, exhibit harmful social biases.\nDebiasing is often challenging due to computational costs, data constraints,\nand potential degradation of multi-task language capabilities. This work\nintroduces a novel approach utilizing ChatGPT to generate synthetic training\ndata, aiming to enhance the debiasing of LLMs. We propose two strategies:\nTargeted Prompting, which provides effective debiasing for known biases but\nnecessitates prior specification of bias in question; and General Prompting,\nwhich, while slightly less effective, offers debiasing across various\ncategories. We leverage resource-efficient LLM debiasing using adapter tuning\nand compare the effectiveness of our synthetic data to existing debiasing\ndatasets. Our results reveal that: (1) ChatGPT can efficiently produce\nhigh-quality training data for debiasing other LLMs; (2) data produced via our\napproach surpasses existing datasets in debiasing performance while also\npreserving internal knowledge of a pre-trained LLM; and (3) synthetic data\nexhibits generalizability across categories, effectively mitigating various\nbiases, including intersectional ones. These findings underscore the potential\nof synthetic data in advancing the fairness of LLMs with minimal retraining\ncost.",
+                    "authors": "Pengrui Han, Rafal Kocielnik, Adhithya Saravanan, Roy Jiang, Or Sharir, Anima Anandkumar",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "68T50",
+                        "I.2.7; K.4.1"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.06852v2",
+                    "title": "ChemLLM: A Chemical Large Language Model",
+                    "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+                    "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+                    "published": "2024-02-10",
+                    "updated": "2024-04-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15491v1",
+                    "title": "Open Source Conversational LLMs do not know most Spanish words",
+                    "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+                    "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.19465v1",
+                    "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+                    "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+                    "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+                    "published": "2024-02-29",
+                    "updated": "2024-02-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.08780v1",
+                    "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+                    "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+                    "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+                    "published": "2023-10-13",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.07420v1",
+                    "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+                    "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+                    "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+                    "published": "2023-12-12",
+                    "updated": "2023-12-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.09397v1",
+                    "title": "Do Large GPT Models Discover Moral Dimensions in Language Representations? A Topological Study Of Sentence Embeddings",
+                    "abstract": "As Large Language Models are deployed within Artificial Intelligence systems,\nthat are increasingly integrated with human society, it becomes more important\nthan ever to study their internal structures. Higher level abilities of LLMs\nsuch as GPT-3.5 emerge in large part due to informative language\nrepresentations they induce from raw text data during pre-training on trillions\nof words. These embeddings exist in vector spaces of several thousand\ndimensions, and their processing involves mapping between multiple vector\nspaces, with total number of parameters on the order of trillions. Furthermore,\nthese language representations are induced by gradient optimization, resulting\nin a black box system that is hard to interpret. In this paper, we take a look\nat the topological structure of neuronal activity in the \"brain\" of Chat-GPT's\nfoundation language model, and analyze it with respect to a metric representing\nthe notion of fairness. We develop a novel approach to visualize GPT's moral\ndimensions. We first compute a fairness metric, inspired by social psychology\nliterature, to identify factors that typically influence fairness assessments\nin humans, such as legitimacy, need, and responsibility. Subsequently, we\nsummarize the manifold's shape using a lower-dimensional simplicial complex,\nwhose topology is derived from this metric. We color it with a heat map\nassociated with this fairness metric, producing human-readable visualizations\nof the high-dimensional sentence manifold. Our results show that sentence\nembeddings based on GPT-3.5 can be decomposed into two submanifolds\ncorresponding to fair and unfair moral judgments. This indicates that GPT-based\nlanguage models develop a moral dimension within their representation spaces\nand induce an understanding of fairness during their training process.",
+                    "authors": "Stephen Fitz",
+                    "published": "2023-09-17",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.04814v2",
+                    "title": "Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks",
+                    "abstract": "We introduce Syntax-Aware Fill-In-the-Middle (SAFIM), a new benchmark for\nevaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM)\ntask. This benchmark focuses on syntax-aware completions of program structures\nsuch as code blocks and conditional expressions, and includes 17,720 examples\nfrom multiple programming languages, sourced from recent code submissions after\nApril 2022 to minimize data contamination. SAFIM provides a robust framework\nwith various prompt designs and novel syntax-aware post-processing techniques,\nfacilitating accurate and fair comparisons across LLMs. Our comprehensive\nevaluation of 15 LLMs shows that FIM pretraining not only enhances FIM\nproficiency but also improves Left-to-Right (L2R) inference using LLMs. Our\nfindings challenge conventional beliefs and suggest that pretraining methods\nand data quality have more impact than model size. SAFIM thus serves as a\nfoundational platform for future research in effective pretraining strategies\nfor code LLMs. The evaluation toolkit and dataset are available at\nhttps://github.com/gonglinyuan/safim, and the leaderboard is available at\nhttps://safimbenchmark.com.",
+                    "authors": "Linyuan Gong, Sida Wang, Mostafa Elhoushi, Alvin Cheung",
+                    "published": "2024-03-07",
+                    "updated": "2024-04-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07609v3",
+                    "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+                    "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+                    "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+                    "published": "2023-05-12",
+                    "updated": "2023-10-17",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.18276v1",
+                    "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+                    "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+                    "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+                    "published": "2024-04-28",
+                    "updated": "2024-04-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "D.1; I.2"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.02680v1",
+                    "title": "Large Language Models are Geographically Biased",
+                    "abstract": "Large Language Models (LLMs) inherently carry the biases contained in their\ntraining corpora, which can lead to the perpetuation of societal harm. As the\nimpact of these foundation models grows, understanding and evaluating their\nbiases becomes crucial to achieving fairness and accuracy. We propose to study\nwhat LLMs know about the world we live in through the lens of geography. This\napproach is particularly powerful as there is ground truth for the numerous\naspects of human life that are meaningfully projected onto geographic space\nsuch as culture, race, language, politics, and religion. We show various\nproblematic geographic biases, which we define as systemic errors in geospatial\npredictions. Initially, we demonstrate that LLMs are capable of making accurate\nzero-shot geospatial predictions in the form of ratings that show strong\nmonotonic correlation with ground truth (Spearman's $\\rho$ of up to 0.89). We\nthen show that LLMs exhibit common biases across a range of objective and\nsubjective topics. In particular, LLMs are clearly biased against locations\nwith lower socioeconomic conditions (e.g. most of Africa) on a variety of\nsensitive subjective topics such as attractiveness, morality, and intelligence\n(Spearman's $\\rho$ of up to 0.70). Finally, we introduce a bias score to\nquantify this and find that there is significant variation in the magnitude of\nbias across existing LLMs.",
+                    "authors": "Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon",
+                    "published": "2024-02-05",
+                    "updated": "2024-02-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.19118v1",
+                    "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate",
+                    "abstract": "Modern large language models (LLMs) like ChatGPT have shown remarkable\nperformance on general language tasks but still struggle on complex reasoning\ntasks, which drives the research on cognitive behaviors of LLMs to explore\nhuman-like problem-solving strategies. Along this direction, one representative\nstrategy is self-reflection, which asks an LLM to refine the solution with the\nfeedback generated by itself iteratively. However, our study shows that such\nreflection-style methods suffer from the Degeneration-of-Thought (DoT) problem:\nonce the LLM has established confidence in its solutions, it is unable to\ngenerate novel thoughts later through reflection even if its initial stance is\nincorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD)\nframework, in which multiple agents express their arguments in the state of\n\"tit for tat\" and a judge manages the debate process to obtain a final\nsolution. Clearly, our MAD framework encourages divergent thinking in LLMs\nwhich would be helpful for tasks that require deep levels of contemplation.\nExperiment results on two challenging datasets, commonsense machine translation\nand counter-intuitive arithmetic reasoning, demonstrate the effectiveness of\nour MAD framework. Extensive analyses suggest that the adaptive break of debate\nand the modest level of \"tit for tat\" state are required for MAD to obtain good\nperformance. Moreover, we find that LLMs might not be a fair judge if different\nLLMs are used for agents. Codes:\nhttps://github.com/Skytliang/Multi-Agents-Debate",
+                    "authors": "Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi",
+                    "published": "2023-05-30",
+                    "updated": "2023-05-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.15997v1",
+                    "title": "RoCar: A Relationship Network-based Evaluation Method to Large Language Models",
+                    "abstract": "Large language models (LLMs) have received increasing attention. However, due\nto the complexity of its capabilities, how to rationally evaluate the\ncapabilities of LLMs is still a task to be solved. We propose the RoCar method,\nwhich utilizes the defined basic schemas to randomly construct a task graph and\ngenerates natural language evaluation tasks based on the task graph to evaluate\nthe reasoning and memory abilities of LLMs respectively. Due to the very large\nrandomness of the task construction process, it is possible to ensure that none\nof the LLMs to be tested has directly learned the evaluation tasks,\nguaranteeing the fairness of the evaluation method.",
+                    "authors": "Ming Wang, Wenfang Wu, Chongyun Gao, Daling Wang, Shi Feng, Yifei Zhang",
+                    "published": "2023-07-29",
+                    "updated": "2023-07-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.17916v2",
+                    "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+                    "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+                    "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+                    "published": "2024-02-27",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.03838v2",
+                    "title": "RADAR: Robust AI-Text Detection via Adversarial Learning",
+                    "abstract": "Recent advances in large language models (LLMs) and the intensifying\npopularity of ChatGPT-like applications have blurred the boundary of\nhigh-quality text generation between humans and machines. However, in addition\nto the anticipated revolutionary changes to our technology and society, the\ndifficulty of distinguishing LLM-generated texts (AI-text) from human-generated\ntexts poses new challenges of misuse and fairness, such as fake content\ngeneration, plagiarism, and false accusations of innocent writers. While\nexisting works show that current AI-text detectors are not robust to LLM-based\nparaphrasing, this paper aims to bridge this gap by proposing a new framework\ncalled RADAR, which jointly trains a robust AI-text detector via adversarial\nlearning. RADAR is based on adversarial training of a paraphraser and a\ndetector. The paraphraser's goal is to generate realistic content to evade\nAI-text detection. RADAR uses the feedback from the detector to update the\nparaphraser, and vice versa. Evaluated with 8 different LLMs (Pythia, Dolly\n2.0, Palmyra, Camel, GPT-J, Dolly 1.0, LLaMA, and Vicuna) across 4 datasets,\nexperimental results show that RADAR significantly outperforms existing AI-text\ndetection methods, especially when paraphrasing is in place. We also identify\nthe strong transferability of RADAR from instruction-tuned LLMs to other LLMs,\nand evaluate the improved capability of RADAR via GPT-3.5-Turbo.",
+                    "authors": "Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho",
+                    "published": "2023-07-07",
+                    "updated": "2023-10-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.11033v4",
+                    "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+                    "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+                    "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+                    "published": "2024-01-19",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.14345v2",
+                    "title": "Bias Testing and Mitigation in LLM-based Code Generation",
+                    "abstract": "Utilizing state-of-the-art Large Language Models (LLMs), automatic code\ngeneration models play a pivotal role in enhancing the productivity of software\ndevelopment procedures. As the adoption of LLMs becomes more widespread in\nsoftware coding ecosystems, a pressing issue has emerged: does the generated\ncode contain social bias and unfairness, such as those related to age, gender,\nand race? This issue concerns the integrity, fairness, and ethical foundation\nof software applications that depend on the code generated by these models, yet\nis under-explored in the literature. This paper presents a novel bias testing\nframework that is specifically designed for code generation tasks. Based on\nthis framework, we conduct an extensive evaluation of the bias in code\ngenerated by five state-of-the-art LLMs. Our findings reveal that 20.29% to\n44.93% code functions generated by the models under study are biased when\nhandling bias sensitive tasks (i.e., tasks that involve sensitive attributes\nsuch as age and gender). This indicates that the existing LLMs can be unfair in\ncode generation, posing risks of unintended and harmful software behaviors. To\nmitigate bias for code generation models, we evaluate five bias mitigation\nprompt strategies, i.e., utilizing bias testing results to refine the code\n(zero-shot), one-, few-shot, and two Chain-of-Thought (CoT) prompts. Our\nevaluation results illustrate that these strategies are all effective in\nmitigating bias. Overall, one-shot and few-shot learning are the two most\neffective. For GPT-4, 80% to 90% code bias can be removed with one-shot\nlearning.",
+                    "authors": "Dong Huang, Qingwen Bu, Jie Zhang, Xiaofei Xie, Junjie Chen, Heming Cui",
+                    "published": "2023-09-03",
+                    "updated": "2024-01-09",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.07981v1",
+                    "title": "Manipulating Large Language Models to Increase Product Visibility",
+                    "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+                    "authors": "Aounon Kumar, Himabindu Lakkaraju",
+                    "published": "2024-04-11",
+                    "updated": "2024-04-11",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.14473v1",
+                    "title": "The Ethics of ChatGPT in Medicine and Healthcare: A Systematic Review on Large Language Models (LLMs)",
+                    "abstract": "With the introduction of ChatGPT, Large Language Models (LLMs) have received\nenormous attention in healthcare. Despite their potential benefits, researchers\nhave underscored various ethical implications. While individual instances have\ndrawn much attention, the debate lacks a systematic overview of practical\napplications currently researched and ethical issues connected to them. Against\nthis background, this work aims to map the ethical landscape surrounding the\ncurrent stage of deployment of LLMs in medicine and healthcare. Electronic\ndatabases and preprint servers were queried using a comprehensive search\nstrategy. Studies were screened and extracted following a modified rapid review\napproach. Methodological quality was assessed using a hybrid approach. For 53\nrecords, a meta-aggregative synthesis was performed. Four fields of\napplications emerged and testify to a vivid exploration phase. Advantages of\nusing LLMs are attributed to their capacity in data analysis, personalized\ninformation provisioning, support in decision-making, mitigating information\nloss and enhancing information accessibility. However, we also identifies\nrecurrent ethical concerns connected to fairness, bias, non-maleficence,\ntransparency, and privacy. A distinctive concern is the tendency to produce\nharmful misinformation or convincingly but inaccurate content. A recurrent plea\nfor ethical guidance and human oversight is evident. Given the variety of use\ncases, it is suggested that the ethical guidance debate be reframed to focus on\ndefining what constitutes acceptable human oversight across the spectrum of\napplications. This involves considering diverse settings, varying potentials\nfor harm, and different acceptable thresholds for performance and certainty in\nhealthcare. In addition, a critical inquiry is necessary to determine the\nextent to which the current experimental use of LLMs is necessary and\njustified.",
+                    "authors": "Joschka Haltaufderheide, Robert Ranisch",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06003v1",
+                    "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+                    "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+                    "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.01262v2",
+                    "title": "Fairness Certification for Natural Language Processing and Large Language Models",
+                    "abstract": "Natural Language Processing (NLP) plays an important role in our daily lives,\nparticularly due to the enormous progress of Large Language Models (LLM).\nHowever, NLP has many fairness-critical use cases, e.g., as an expert system in\nrecruitment or as an LLM-based tutor in education. Since NLP is based on human\nlanguage, potentially harmful biases can diffuse into NLP systems and produce\nunfair results, discriminate against minorities or generate legal issues.\nHence, it is important to develop a fairness certification for NLP approaches.\nWe follow a qualitative research approach towards a fairness certification for\nNLP. In particular, we have reviewed a large body of literature on algorithmic\nfairness, and we have conducted semi-structured expert interviews with a wide\nrange of experts from that area. We have systematically devised six fairness\ncriteria for NLP, which can be further refined into 18 sub-categories. Our\ncriteria offer a foundation for operationalizing and testing processes to\ncertify fairness, both from the perspective of the auditor and the audited\norganization.",
+                    "authors": "Vincent Freiberger, Erik Buchmann",
+                    "published": "2024-01-02",
+                    "updated": "2024-01-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.08656v1",
+                    "title": "Linear Cross-document Event Coreference Resolution with X-AMR",
+                    "abstract": "Event Coreference Resolution (ECR) as a pairwise mention classification task\nis expensive both for automated systems and manual annotations. The task's\nquadratic difficulty is exacerbated when using Large Language Models (LLMs),\nmaking prompt engineering for ECR prohibitively costly. In this work, we\npropose a graphical representation of events, X-AMR, anchored around individual\nmentions using a \\textbf{cross}-document version of \\textbf{A}bstract\n\\textbf{M}eaning \\textbf{R}epresentation. We then linearize the ECR with a\nnovel multi-hop coreference algorithm over the event graphs. The event graphs\nsimplify ECR, making it a) LLM cost-effective, b) compositional and\ninterpretable, and c) easily annotated. For a fair assessment, we first enrich\nan existing ECR benchmark dataset with these event graphs using an\nannotator-friendly tool we introduce. Then, we employ GPT-4, the newest LLM by\nOpenAI, for these annotations. Finally, using the ECR algorithm, we assess\nGPT-4 against humans and analyze its limitations. Through this research, we aim\nto advance the state-of-the-art for efficient ECR and shed light on the\npotential shortcomings of current LLMs at this task. Code and annotations:\n\\url{https://github.com/ahmeshaf/gpt_coref}",
+                    "authors": "Shafiuddin Rehan Ahmed, George Arthur Baker, Evi Judge, Michael Regan, Kristin Wright-Bettner, Martha Palmer, James H. Martin",
+                    "published": "2024-03-25",
+                    "updated": "2024-03-25",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04205v2",
+                    "title": "Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves",
+                    "abstract": "Misunderstandings arise not only in interpersonal communication but also\nbetween humans and Large Language Models (LLMs). Such discrepancies can make\nLLMs interpret seemingly unambiguous questions in unexpected ways, yielding\nincorrect responses. While it is widely acknowledged that the quality of a\nprompt, such as a question, significantly impacts the quality of the response\nprovided by LLMs, a systematic method for crafting questions that LLMs can\nbetter comprehend is still underdeveloped. In this paper, we present a method\nnamed `Rephrase and Respond' (RaR), which allows LLMs to rephrase and expand\nquestions posed by humans and provide responses in a single prompt. This\napproach serves as a simple yet effective prompting method for improving\nperformance. We also introduce a two-step variant of RaR, where a rephrasing\nLLM first rephrases the question and then passes the original and rephrased\nquestions together to a different responding LLM. This facilitates the\neffective utilization of rephrased questions generated by one LLM with another.\nOur experiments demonstrate that our methods significantly improve the\nperformance of different models across a wide range to tasks. We further\nprovide a comprehensive comparison between RaR and the popular Chain-of-Thought\n(CoT) methods, both theoretically and empirically. We show that RaR is\ncomplementary to CoT and can be combined with CoT to achieve even better\nperformance. Our work not only contributes to enhancing LLM performance\nefficiently and effectively but also sheds light on a fair evaluation of LLM\ncapabilities. Data and codes are available at\nhttps://github.com/uclaml/Rephrase-and-Respond.",
+                    "authors": "Yihe Deng, Weitong Zhang, Zixiang Chen, Quanquan Gu",
+                    "published": "2023-11-07",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.03033v1",
+                    "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+                    "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+                    "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+                    "published": "2023-11-06",
+                    "updated": "2023-11-06",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04892v2",
+                    "title": "Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs",
+                    "abstract": "Recent works have showcased the ability of LLMs to embody diverse personas in\ntheir responses, exemplified by prompts like 'You are Yoda. Explain the Theory\nof Relativity.' While this ability allows personalization of LLMs and enables\nhuman behavior simulation, its effect on LLMs' capabilities remains unclear. To\nfill this gap, we present the first extensive study of the unintended\nside-effects of persona assignment on the ability of LLMs to perform basic\nreasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse\npersonas (e.g. an Asian person) spanning 5 socio-demographic groups. Our\nexperiments unveil that LLMs harbor deep rooted bias against various\nsocio-demographics underneath a veneer of fairness. While they overtly reject\nstereotypes when explicitly asked ('Are Black people less skilled at\nmathematics?'), they manifest stereotypical and erroneous presumptions when\nasked to answer questions while adopting a persona. These can be observed as\nabstentions in responses, e.g., 'As a Black person, I can't answer this\nquestion as it requires math knowledge', and generally result in a substantial\nperformance drop. Our experiments with ChatGPT-3.5 show that this bias is\nubiquitous - 80% of our personas demonstrate bias; it is significant - some\ndatasets show performance drops of 70%+; and can be especially harmful for\ncertain groups - some personas suffer statistically significant drops on 80%+\nof the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with\nGPT-4-Turbo showing the least but still a problematic amount of bias (evident\nin 42% of the personas). Further analysis shows that these persona-induced\nerrors can be hard-to-discern and hard-to-avoid. Our findings serve as a\ncautionary tale that the practice of assigning personas to LLMs - a trend on\nthe rise - can surface their deep-rooted biases and have unforeseeable and\ndetrimental side-effects.",
+                    "authors": "Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot",
+                    "published": "2023-11-08",
+                    "updated": "2024-01-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.14208v2",
+                    "title": "Content Conditional Debiasing for Fair Text Embedding",
+                    "abstract": "Mitigating biases in machine learning models has gained increasing attention\nin Natural Language Processing (NLP). Yet, only a few studies focus on fair\ntext embeddings, which are crucial yet challenging for real-world applications.\nIn this paper, we propose a novel method for learning fair text embeddings. We\nachieve fairness while maintaining utility trade-off by ensuring conditional\nindependence between sensitive attributes and text embeddings conditioned on\nthe content. Specifically, we enforce that embeddings of texts with different\nsensitive attributes but identical content maintain the same distance toward\nthe embedding of their corresponding neutral text. Furthermore, we address the\nissue of lacking proper training data by using Large Language Models (LLMs) to\naugment texts into different sensitive groups. Our extensive evaluations\ndemonstrate that our approach effectively improves fairness while preserving\nthe utility of embeddings, representing a pioneering effort in achieving\nconditional independence for fair text embeddings.",
+                    "authors": "Wenlong Deng, Blair Chen, Xiaoxiao Li, Christos Thrampoulidis",
+                    "published": "2024-02-22",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.08836v2",
+                    "title": "Bias and Fairness in Chatbots: An Overview",
+                    "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+                    "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+                    "published": "2023-09-16",
+                    "updated": "2023-12-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.01964v1",
+                    "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+                    "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+                    "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.06500v1",
+                    "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+                    "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+                    "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+                    "published": "2023-10-10",
+                    "updated": "2023-10-10",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.18502v1",
+                    "title": "Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification",
+                    "abstract": "Employing Large Language Models (LLM) in various downstream applications such\nas classification is crucial, especially for smaller companies lacking the\nexpertise and resources required for fine-tuning a model. Fairness in LLMs\nhelps ensure inclusivity, equal representation based on factors such as race,\ngender and promotes responsible AI deployment. As the use of LLMs has become\nincreasingly prevalent, it is essential to assess whether LLMs can generate\nfair outcomes when subjected to considerations of fairness. In this study, we\nintroduce a framework outlining fairness regulations aligned with various\nfairness definitions, with each definition being modulated by varying degrees\nof abstraction. We explore the configuration for in-context learning and the\nprocedure for selecting in-context demonstrations using RAG, while\nincorporating fairness rules into the process. Experiments conducted with\ndifferent LLMs indicate that GPT-4 delivers superior results in terms of both\naccuracy and fairness compared to other models. This work is one of the early\nattempts to achieve fairness in prediction tasks by utilizing LLMs through\nin-context learning.",
+                    "authors": "Garima Chhikara, Anurag Sharma, Kripabandhu Ghosh, Abhijnan Chakraborty",
+                    "published": "2024-02-28",
+                    "updated": "2024-02-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05345v3",
+                    "title": "RTLLM: An Open-Source Benchmark for Design RTL Generation with Large Language Model",
+                    "abstract": "Inspired by the recent success of large language models (LLMs) like ChatGPT,\nresearchers start to explore the adoption of LLMs for agile hardware design,\nsuch as generating design RTL based on natural-language instructions. However,\nin existing works, their target designs are all relatively simple and in a\nsmall scale, and proposed by the authors themselves, making a fair comparison\namong different LLM solutions challenging. In addition, many prior works only\nfocus on the design correctness, without evaluating the design qualities of\ngenerated design RTL. In this work, we propose an open-source benchmark named\nRTLLM, for generating design RTL with natural language instructions. To\nsystematically evaluate the auto-generated design RTL, we summarized three\nprogressive goals, named syntax goal, functionality goal, and design quality\ngoal. This benchmark can automatically provide a quantitative evaluation of any\ngiven LLM-based solution. Furthermore, we propose an easy-to-use yet\nsurprisingly effective prompt engineering technique named self-planning, which\nproves to significantly boost the performance of GPT-3.5 in our proposed\nbenchmark.",
+                    "authors": "Yao Lu, Shang Liu, Qijun Zhang, Zhiyao Xie",
+                    "published": "2023-08-10",
+                    "updated": "2023-11-11",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.12090v1",
+                    "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+                    "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+                    "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+                    "published": "2023-05-20",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.15007v1",
+                    "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+                    "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+                    "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+                    "published": "2023-10-23",
+                    "updated": "2023-10-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.09219v5",
+                    "title": "\"Kelly is a Warm Person, Joseph is a Role Model\": Gender Biases in LLM-Generated Reference Letters",
+                    "abstract": "Large Language Models (LLMs) have recently emerged as an effective tool to\nassist individuals in writing various types of content, including professional\ndocuments such as recommendation letters. Though bringing convenience, this\napplication also introduces unprecedented fairness concerns. Model-generated\nreference letters might be directly used by users in professional scenarios. If\nunderlying biases exist in these model-constructed letters, using them without\nscrutinization could lead to direct societal harms, such as sabotaging\napplication success rates for female applicants. In light of this pressing\nissue, it is imminent and necessary to comprehensively study fairness issues\nand associated harms in this real-world use case. In this paper, we critically\nexamine gender biases in LLM-generated reference letters. Drawing inspiration\nfrom social science findings, we design evaluation methods to manifest biases\nthrough 2 dimensions: (1) biases in language style and (2) biases in lexical\ncontent. We further investigate the extent of bias propagation by analyzing the\nhallucination bias of models, a term that we define to be bias exacerbation in\nmodel-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs-\nChatGPT and Alpaca, we reveal significant gender biases in LLM-generated\nrecommendation letters. Our findings not only warn against using LLMs for this\napplication without scrutinization, but also illuminate the importance of\nthoroughly studying hidden biases and harms in LLM-generated professional\ndocuments.",
+                    "authors": "Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng",
+                    "published": "2023-10-13",
+                    "updated": "2023-12-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.11595v3",
+                    "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+                    "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+                    "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+                    "published": "2023-05-19",
+                    "updated": "2023-10-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.03192v1",
+                    "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+                    "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+                    "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+                    "published": "2024-04-04",
+                    "updated": "2024-04-04",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15451v1",
+                    "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+                    "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+                    "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+                    "published": "2024-03-18",
+                    "updated": "2024-03-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.13925v1",
+                    "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+                    "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+                    "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.08189v1",
+                    "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+                    "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+                    "authors": "Karthik Sreedhar, Lydia Chilton",
+                    "published": "2024-02-13",
+                    "updated": "2024-02-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.07884v2",
+                    "title": "Fair Abstractive Summarization of Diverse Perspectives",
+                    "abstract": "People from different social and demographic groups express diverse\nperspectives and conflicting opinions on a broad set of topics such as product\nreviews, healthcare, law, and politics. A fair summary should provide a\ncomprehensive coverage of diverse perspectives without underrepresenting\ncertain groups. However, current work in summarization metrics and Large\nLanguage Models (LLMs) evaluation has not explored fair abstractive\nsummarization. In this paper, we systematically investigate fair abstractive\nsummarization for user-generated data. We first formally define fairness in\nabstractive summarization as not underrepresenting perspectives of any groups\nof people, and we propose four reference-free automatic metrics by measuring\nthe differences between target and source perspectives. We evaluate nine LLMs,\nincluding three GPT models, four LLaMA models, PaLM 2, and Claude, on six\ndatasets collected from social media, online reviews, and recorded transcripts.\nExperiments show that both the model-generated and the human-written reference\nsummaries suffer from low fairness. We conduct a comprehensive analysis of the\ncommon factors influencing fairness and propose three simple but effective\nmethods to alleviate unfair summarization. Our dataset and code are available\nat https://github.com/psunlpgroup/FairSumm.",
+                    "authors": "Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang",
+                    "published": "2023-11-14",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18130v2",
+                    "title": "DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues",
+                    "abstract": "Controversy is a reflection of our zeitgeist, and an important aspect to any\ndiscourse. The rise of large language models (LLMs) as conversational systems\nhas increased public reliance on these systems for answers to their various\nquestions. Consequently, it is crucial to systematically examine how these\nmodels respond to questions that pertaining to ongoing debates. However, few\nsuch datasets exist in providing human-annotated labels reflecting the\ncontemporary discussions. To foster research in this area, we propose a novel\nconstruction of a controversial questions dataset, expanding upon the publicly\nreleased Quora Question Pairs Dataset. This dataset presents challenges\nconcerning knowledge recency, safety, fairness, and bias. We evaluate different\nLLMs using a subset of this dataset, illuminating how they handle controversial\nissues and the stances they adopt. This research ultimately contributes to our\nunderstanding of LLMs' interaction with controversial issues, paving the way\nfor improvements in their comprehension and handling of complex societal\ndebates.",
+                    "authors": "David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein, Jason D. Williams",
+                    "published": "2023-10-27",
+                    "updated": "2023-11-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15198v2",
+                    "title": "Do LLM Agents Exhibit Social Behavior?",
+                    "abstract": "The advances of Large Language Models (LLMs) are expanding their utility in\nboth academic research and practical applications. Recent social science\nresearch has explored the use of these ``black-box'' LLM agents for simulating\ncomplex social systems and potentially substituting human subjects in\nexperiments. Our study delves into this emerging domain, investigating the\nextent to which LLMs exhibit key social interaction principles, such as social\nlearning, social preference, and cooperative behavior (indirect reciprocity),\nin their interactions with humans and other agents. We develop a framework for\nour study, wherein classical laboratory experiments involving human subjects\nare adapted to use LLM agents. This approach involves step-by-step reasoning\nthat mirrors human cognitive processes and zero-shot learning to assess the\ninnate preferences of LLMs. Our analysis of LLM agents' behavior includes both\nthe primary effects and an in-depth examination of the underlying mechanisms.\nFocusing on GPT-4, our analyses suggest that LLM agents appear to exhibit a\nrange of human-like social behaviors such as distributional and reciprocity\npreferences, responsiveness to group identity cues, engagement in indirect\nreciprocity, and social learning capabilities. However, our analysis also\nreveals notable differences: LLMs demonstrate a pronounced fairness preference,\nweaker positive reciprocity, and a more calculating approach in social learning\ncompared to humans. These insights indicate that while LLMs hold great promise\nfor applications in social science research, such as in laboratory experiments\nand agent-based modeling, the subtle behavioral differences between LLM agents\nand humans warrant further investigation. Careful examination and development\nof protocols in evaluating the social behaviors of LLMs are necessary before\ndirectly applying these models to emulate human behavior.",
+                    "authors": "Yan Leng, Yuan Yuan",
+                    "published": "2023-12-23",
+                    "updated": "2024-02-22",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.SI",
+                        "econ.GN",
+                        "q-fin.EC"
+                    ],
+                    "category": "LLM Fairness"
+                }
+            ],
+            [
+                {
+                    "url": "http://arxiv.org/abs/2404.12535v1",
+                    "title": "HalluciBot: Is There No Such Thing as a Bad Question?",
+                    "abstract": "Hallucination continues to be one of the most critical challenges in the\ninstitutional adoption journey of Large Language Models (LLMs). In this\ncontext, an overwhelming number of studies have focused on analyzing the\npost-generation phase - refining outputs via feedback, analyzing logit output\nvalues, or deriving clues via the outputs' artifacts. We propose HalluciBot, a\nmodel that predicts the probability of hallucination $\\textbf{before\ngeneration}$, for any query imposed to an LLM. In essence, HalluciBot does not\ninvoke any generation during inference. To derive empirical evidence for\nHalluciBot, we employ a Multi-Agent Monte Carlo Simulation using a Query\nPerturbator to craft $n$ variations per query at train time. The construction\nof our Query Perturbator is motivated by our introduction of a new definition\nof hallucination - $\\textit{truthful hallucination}$. Our training methodology\ngenerated 2,219,022 estimates for a training corpus of 369,837 queries,\nspanning 13 diverse datasets and 3 question-answering scenarios. HalluciBot\npredicts both binary and multi-class probabilities of hallucination, enabling a\nmeans to judge the query's quality with regards to its propensity to\nhallucinate. Therefore, HalluciBot paves the way to revise or cancel a query\nbefore generation and the ensuing computational waste. Moreover, it provides a\nlucid means to measure user accountability for hallucinatory queries.",
+                    "authors": "William Watson, Nicole Cho",
+                    "published": "2024-04-18",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "label": "Original Paper",
+                    "paper_cat": "LLM Fairness",
+                    "gt": "With regards to hallucination mitigation studies, an overwhelming majority focuses on the postgeneration stage of analyzing outputs. A minority concentrates on the pre-generation phase and even amongst those, the focus lies in incorporating recent knowledge into LLMs. In detail, many expand on the universally utilized method of context-based retrieval systems (Reimers and Gurevych, 2019; Johnson et al., 2019; Nogueira and Cho, 2020; Karpukhin et al., 2020; Lewis et al., 2020; Izacard and Grave, 2021). Other methods include relying on the model\u2019s general knowledge (Khashabi et al., 2020) or conditioning the QA model on context generated by the LLM itself (Yu et al., 2023). Certain work has focused on mitigating hallucinations by augmenting the way LLMs generate their answers. One of the more popular techniques is to have the model enumerate its chain-of-thought (Wei et al., 2023) and think step by step (Nye et al., 2021), while building context. Another method to augment generation with context is by semantic retrieval (Lewis et al., 2021; Liu et al., 2021), handling hallucinations as they arise (Varshney et al., 2023), or using LLMs to generate context rather than retrieve (Yu et al., 2023). PromptChainer (Wu et al., 2022) profiled techniques to craft LLM chains, in which the output of one LLM\u2019s generation process, when fed into the next LLM, can 2 HalluciBot Inference HalluciBot Training Methodology  q0 LLM q1 q2 q3 qn qn-1 LLM1 LLM2 LLM3 LLMn-1 LLMn a2 a3 an-1 a1 an LLM0 a0 Multi-Agent Simulation  Estimates Hallucination Probability Before Generation    Empirical  Estimate  \ud835\udfcf \ud835\udc8f+ \ud835\udfcf\u0dcd \ud835\udc8a=\ud835\udfce \ud835\udc8f+\ud835\udfcf \ud835\udd40(\ud835\udc82\ud835\udc8a\u2260\ud835\udc9a) \u2248\ud835\udc91\ud835\udc89q0 Monte Carlo  Simulation Encoder-based  Classifier Query Perturbator \u2026 \u2026 \u2026 q0 \ud835\udc78\ud835\udc96\ud835\udc86\ud835\udc93\ud835\udc9a Figure 2: System Overview. A single query q0, supplied by the user, is perturbed in n different ways. Next, the original and perturbed queries qi are independently answered by the Generator agents. This Monte Carlo Simulation provides an estimate into the rate of hallucination ph(q0) for an original query q0. Undergoing this Multi-Agent Monte Carlo Simulation, covering 13 datasets, HalluciBot is trained to predict the probability that any query q0 will yield an hallucination, and predict the expected value of hallucinations sampled before generation. allow for more complex tasks. Language Model Cascades (Dohan et al., 2022) demonstrated that LLMs can yield probabilistic programs to tackle multi-step reasoning problems. Self-consistency (Wang et al., 2023) leveraged a new decoding strategy to sample multiple generative pathways then select the most consistent answer. Also, Kumar et al. (2022) explored gradient-based sampling procedures that satisfy user-defined constraints. Most recent work has focused on sampling-based calibration within a single model (Cole et al., 2023) or self-verification (Kadavath et al., 2022) the latter focuses on generating a set of outputs and feeding those back into the LLM. Furthermore, Snyder et al. (2023) explores how artifacts can differentiate hallucinated outputs. One common feature amongst these approaches is that the focus is on the output rather than the query. Also, while Zheng and Saparov (2023) study lexical perturbations, no study on hallucinations has employed a Multi-Agent approach coupled with query perturbations which are hallmark features of HalluciBot.",
+                    "pre_questions": [],
+                    "main_content": "HalluciBot: Is There No Such Thing as a Bad Question? William Watson* and Nicole Cho* J.P. Morgan AI Research New York, NY, USA william.watson@jpmchase.com nicole.cho@jpmorgan.com Abstract Hallucination continues to be one of the most critical challenges in the institutional adoption journey of Large Language Models (LLMs). In this context, an overwhelming number of studies have focused on analyzing the postgeneration phase refining outputs via feedback, analyzing logit output values, or deriving clues via the outputs\u2019 artifacts. We propose HalluciBot, a model that predicts the probability of hallucination before generation, for any query imposed to an LLM. In essence, HalluciBot does not invoke any generation during inference. To derive empirical evidence for HalluciBot, we employ a Multi-Agent Monte Carlo Simulation using a Query Perturbator to craft n variations per query at train time. The construction of our Query Perturbator is motivated by our introduction of a new definition of hallucination truthful hallucination. Our training methodology generated 2,219,022 estimates for a training corpus of 369,837 queries, spanning 13 diverse datasets and 3 questionanswering scenarios. HalluciBot predicts both binary and multi-class probabilities of hallucination, enabling a means to judge the query\u2019s quality with regards to its propensity to hallucinate. Therefore, HalluciBot paves the way to revise or cancel a query before generation and the ensuing computational waste. Moreover, it provides a lucid means to measure user accountability for hallucinatory queries. 1 Introduction Despite the promising potential for a myriad of practical use cases, Large Language Models (LLMs) offer limited insights into their chain of thought (Liang et al., 2022; Wei et al., 2023; Kojima et al., 2023; Li et al., 2023) and have the propensity to hallucinate in various circumstances (Jiang et al., 2021). Common factors that drive hallucinations encompass high model complex*Equal Contribution Traditional Estimation If Output A seems hallucinated: Iterative generations may  derive a more accurate output  Query: Who is the writer ?   Query: Who is the writer of  Harry Potter ?  \u2022 User ends the session or, \u2022 User tries to revise the query HalluciBot Binary Classifier LLM Output: A+ With these scores, user skips  generation and revises query HalluciBot Query: Who is the writer?   Unlikely to  hallucinate  Likely to  hallucinate  HalluciBot Multi-Class Classifier Binary Classifier HalluciBot Unlikely to  hallucinate  Likely to  hallucinate  HalluciBot Multi-Class Classifier Query: Who is the writer of  Harry Potter ?  Output: J.K.Rowling Output: J.K.Rowling Output: A person  98.8 % of seeing 6 hallucinations  88.2 % of seeing 0 hallucinations  Figure 1: Comparison between traditional methods of confidence estimation and HalluciBot. In the former, user inputs a query to generate an output that is accurate or hallucinatory. If hallucinatory, user can end the session or revise the query for iterative rounds of generation. In contrast, HalluciBot predicts the probability of hallucination for a given query before generation. Therefore, users can instantly gain insight into hallucination probability and revise a query. ity, flawed data sources, or inherent sampling randomness. Specifically, the intrinsic trade-off between greedy deterministic decoding and the creativity spawned through nucleus sampling induces a heightened propensity to generate hallucinations (Huang et al., 2023). This challenge is compounded by limitations such as the frequent inaccessibility into the LLMs\u2019 training datasets (Liang et al., 2022). HuggingFace\u2019s release of its \"Hallucinations Leaderboard\" on January 29th, 2024 (Minervini et al., 2024; Gao et al., 2023) highlights the importance of resolving hallucination-related issues via the concerted effort of evaluating different LLMs. In this context, the majority of current studies have focused on the post-generation phase of output analysis as expanded in Peng et al. (2023) 1 arXiv:2404.12535v1  [cs.LG]  18 Apr 2024 such as (1) self-refinement via feedback loops on the model\u2019s output (Madaan et al., 2023), (2) analysis of logit output values to detect hallucination (Varshney et al., 2023), or (3) for a minority of studies focused on the pre-generation phase, the ingestion of recent knowledge to improve performance (Tonmoy et al., 2024). We propose a novel model, HalluciBot, that predicts the probability of hallucination, before any generation, for any type of query. In essence, for any query posed to an LLM, HalluciBot estimates: \u25b6a binary classification of the query\u2019s propensity to hallucinate (\"Yes\" or \"No\"), as well as, \u25b6a multi-class hallucination rate estimating the expected value of hallucination via sampling. In order to understand HalluciBot, it is imperative to discuss the following key fundamental characteristics regarding hallucinations and LLMs. Firstly, hallucination itself is not a derivative of any mathematical model and intrinsically, can only be determined by juxtaposing the ground truth with the LLM\u2019s generated outputs. Secondly, LLMs frequently advance output quality by different sampling methods such as nucleus sampling (Holtzman et al., 2020; Fan et al., 2018; Holtzman et al., 2018; Radford et al., 2018). In light of these features, HalluciBot is trained via the method of leveraging a Multi-Agent Monte Carlo Simulation for the construction of an encoder for hallucination quantification. In essence, HalluciBot, leveraging gpt-3.5-turbo, is trained via: \u25b6perturbing 369,837 queries n times into unique variations that retain the original semantic meaning yet significantly diverge lexically; then, \u25b6employing n+1 independent agents to sample an output from each query (including the original) at a temperature of 1.0 for diversity; \u25b6conducting a Monte Carlo Simulation on 2,219,022 sampled outputs; \u25b6deriving an empirical estimate into the expected rate of hallucination ph(q0) for the original query as the ground truth for our encoder. As a result, our study has culminated in the following pillars of contribution. \u25b6HalluciBot, the first encoder-based model to derive, before generation, an anticipated rate of hallucination for any type of query. HalluciBot has scored a validation accuracy of 73% on binary labels and 73% for top 3 multi-class labels. \u25b6Moreover, our approach to construct HalluciBot absorbs the computational complexity of Monte Carlo sampling and training prior to the user session. This differs from the prevalent hallucination detection methods that involve sampling during or after the user\u2019s current session. Thus, institutions that employ HalluciBot can systematically save on the considerable amount of computational waste engendered by \"highly probable\" hallucinatory queries. \u25b6HalluciBot also generalizes to systems with Retrieval Augmented Generation (RAG) context or few-shot question answering systems with an LLM generator. Also it can generalize to systems only accessible via API calls (OpenAI, 2022; Google, 2023; Microsoft, 2023). Thus, in diverse implementation environments, HalluciBot can impart accountability to the user with regards to the query\u2019s quality. \u25b6We also present a myriad of behavioral and statistical discoveries from our experiments, such as an LLM\u2019s probability to either be exceptionally correct or completely incorrect in certain scenarios. Our Appendix elaborates in detail. 3.1 What is Hallucination? In general terms, hallucination refers to a false perception of patterns or objects resulting from one\u2019s senses. With regards to LLMs, a myriad of studies have categorized hallucinations into bifurcated structures such as (1) intrinsic hallucination, which refers to the LLM\u2019s outputs directly contradicting with the source content for extractive queries, or (2) extrinsic hallucination, which refers to the LLM\u2019s outputs being unverifiable by the source content (i.e. irrelevant outputs) (Ji et al., 2023). From a different angle, (1) factuality hallucinations refer to outputs which directly contradict or fabricate the ground truth while (2) faithfulness hallucinations define outputs that misunderstand the context or intent of the query (Huang et al., 2023). In this study, we will focus on the aforementioned types of hallucinations and introduce truthful hallucination as a novel concept. \u25b6The motivation for truthful hallucination stems from the analysis that neural networks display an intrinsic propensity to memorize training data (Carlini et al., 2021). \u25b6Truthful hallucination is defined as an LLM\u2019s inability to answer semantically similar but lexically different perturbations of a query. \u25b6Understanding truthful hallucination is essential to account for training bias. Namely, if an LLM has memorized a query and therefore can answer it correctly, but demonstrates the inability to answer similar queries that are worded differently, it will be held accountable. \u25b6Moreover, truthful hallucination is reflective of user behavior unless there is a drop-down of query templates, there will almost always be di3 verse representations of the same query. 3.2 What is the Motivation for HalluciBot? HalluciBot focuses on distilling LLM behavior into a speedy, encoder-based system that can predict hallucination before generation. Foremost, this is in contrast to prior work that uses multiple generations during a user\u2019s session to provide selfconsistency (Manakul et al., 2023). Next, our proposal differs from entropy based, log-prob based, or model based estimation techniques (Huang et al., 2023) that rely on the LLM\u2019s uncertainty to predict hallucinations these methods focus on the model\u2019s bias while we focus on empirical estimates. Moreover, our approach consists of a Multi-Agent Simulation which stands in stark contrast to the majority of current experiments that have focused on leveraging a single LLM agent to generate outputs from a single query (Cole et al., 2023; Kadavath et al., 2022; Snyder et al., 2023). The training procedure for HalluciBot consists of: \u25b6perturbing 369,837 queries n times (\u00a73.3) \u25b6employing n+1 independent LLM agents (\u00a73.4) \u25b6sampling an output from each query (\u00a73.4) \u25b6conducting a Monte Carlo Simulation on the 2,219,022 sampled outputs (\u00a73.5) \u25b6training an encoder-based classifier (\u00a73.6) 3.3 What is a Query Perturbator? We are motivated by how perturbations: \u25b6help comprehend truthful hallucination. We induce diversity to disentangle the generation process from any potential training bias. \u25b6emulate real-life scenarios, in which users unintentionally create variations of the same query. Thus, the Query Perturbator is a gpt-3.5-turbo LLM agent T that generates n = 5 perturbations to the original query q0 while retaining the same semantic meaning. Note that in practice, there are n + 1 = 6 transformations because for a given query q0, we define T0(q0) = I(q0) = q0 as an identity transform. Therefore, the original query will always be included in the set of transformed queries for the next step. In effect, the generation process can be summarized as returning a set of Q = {q0, q1, . . . , qn} query perturbations of size n + 1. The Query Perturbator\u2019s singular purpose is to: Rewrite the query in {n} radically different ways. One prompt call is sufficient to discourage duplicates. Temperature is set to 1.0 to prioritize creativity and lexical diversity (Table 6). Q = T (q0) = \uf8ee \uf8ef \uf8ef \uf8ef \uf8f0 I(q0) T1(q0) . . . Tn(q0) \uf8f9 \uf8fa \uf8fa \uf8fa \uf8fb= \uf8ee \uf8ef \uf8ef \uf8ef \uf8f0 q0 q1 . . . qn \uf8f9 \uf8fa \uf8fa \uf8fa \uf8fb 3.4 What is an Output Generator? Next, the Output Generator is deployed for each perturbation due to the following reasons: \u25b6Given the complexity of LLMs with nucleus sampling decoding, analyzing a single output from a single query may not be comprehensive enough to determine hallucination. \u25b6Furthermore, given the risk of over-training LLMs, their opaque training data, and propensity to memorize, generating multiple outputs from the same query does not eliminate training bias. This has been a prevalent mode of studying hallucination in prior studies (\u00a73.2) A new generation process should be launched each time to answer the next perturbed query to ensure a truly independent output generation process. Thus, for the perturbed set Q for a sample q0, the Output Generator consists of |Q| = n + 1 independent gpt-3.5-turbo LLM agents to generate outputs ai \u2208A for each variation qi \u2208Q. The LLM agent will receive (1) for Extractive queries, a prompt with the query qi, alongside context ci, (2) for Multiple-Choice queries, candidate choices ki \u2208K, and (3) for Abstractive queries, no additional context. See Table 8 for each experiment\u2019s prompt procedure. Temperature for all experiments is set to 1.0 to stress-test and encourage diversity. 3.5 What is the purpose of a Monte Carlo Simulation in HalluciBot? As evidenced by multiple studies and Section 3.1, hallucination is the outcome of multiple confounding variables thus, it is highly unlikely that a tractable closed-form solution will be able to model hallucinations. Thus, we employ a Monte Carlo Simulation as a means to derive empirical estimations of hallucination rates in LLMs, since this method is frequently leveraged to map probability in the presence of random variable inference (Swaminathan, 2021). Thus, we estimate the probability density that a query induces hallucination. 3.6 How Do We Train a Classifier? Once the Monte Carlo Simulation is complete for our training corpus composed of 369,837 queries 4 Scenario Datasets Extractive (1) SQuADv2 Multiple TruthfulQA, SciQ, MMLU, PIQA, Choice (9) BoolQ, OpenBookQA, MathQA ARC Easy, ARC Challenge Abstractive (6) SQuADv2, TruthfulQA, SciQ, WikiQA, HotpotQA, TriviaQA Table 1: Dataset Scenario Split with Reused Assets. spanning 13 different datasets (Appendix C, Tables 1 & 11), we start training our classifier. These queries encompass Extractive, Multiple Choice, and Abstractive scenarios. Each scenario, with or without additional context, affects the the hallucination rate of gpt-3.5-turbo. The estimates provided through our Monte Carlo Simulation are proportional to an approximation of hallucination rates. \u25b6With a synthetic labeled set of queries q0 and their rate of hallucinations ph(q0), we train an encoder-style classifier based on BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) to estimate the hallucination probability density from our Monte Carlo Simulation. \u25b6We train two versions: a binary model to estimate the propensity a query can hallucinate, and a multi-class model to predict the expected value of hallucinations if sampled n + 1 times. Our experiments constrain the number of perturbations to n = 5, and when including the original query and output, we can model the hallucination rate for n + 1 = 6 modes this translates to increments of 16.6% in hallucination rates. 3.7 How Do We Encode a Query\u2019s Scenario? We conduct an ablation study to explore if incorporating the query\u2019s scenario mitigates hallucinations. To create the prompt, we prepend the original query q0 with either [EXTRACTIVE], [MULTIPLE CHOICE], or [ABSTRACTIVE], using the format \u00ab{tag} {q0}\u00bb. Our hypothesis is based on recent research that highlights the use of RAG (Yan et al., 2024; Lewis et al., 2021; Guu et al., 2020) to alleviate hallucinations. The additional context provides valuable signals related to the hallucination rate of the original query. Furthermore, we apply this technique to distinguish our experimental results from reused datasets in different scenarios, such as SciQ and SQuADv2. The queries are encoded exactly as they appear in the original datasets. 3.8 How Do We Convert Monte Carlo Estimates to Labels? Empirical Estimate The probability of truthful hallucinations for a query q0, denoted as ph(q0), can be empirically estimated based on the output ai \u2208A of our Multi-Agent Monte Carlo Simulation. We define the indicator function I to measure the incorrectness of an output ai with respect to the ground truth y for query q0. ph \u0000q0 \u0001 \u2248 1 n + 1 n+1 X i=0 I \u0002 ai \u0338= y \u0003 Binary Labels To assess the propensity to hallucinate, we simplify the problem by considering two response values: whether q0 produces any hallucination or not. Thus, we define the binary values for the probability of any hallucination as pb(q0). pb \u0000q0 \u0001 = ( 1 if ph \u0000q0 \u0001 > 0 0 if ph \u0000q0 \u0001 = 0 Multi-class Labels HalluciBot is trained to estimate the occurrence of hallucinations when queried and sampled under n + 1 trials. To facilitate training, we convert the proportion into discrete classes by multiplying the original estimate ph(q0) by the number of agents n + 1. This transformed variable is denoted as E \u0002 ph(q0) \u0003 . E h ph(q0) i = j (n + 1) \u00b7 ph(q0) k 4 Metrics for Hallucination Two key metrics are observed: accuracy and agreement. The concept of truthful hallucination focuses on the accuracy of individual agents. In addition, we analyze the level of agreement as an indicator of correlation. Moreover, the simulation\u2019s reliability is assessed at the corpus level. Accuracy Accuracy serves as the measure of correctness in our evaluation by comparing the generated output ai to the ground truth string y, aiming for an almost exact match. For Multiple Choice queries, we also consider the choice label. However, due to minor variations in special characters, we adopt a partial, case-insensitive matching approach using the TheFuzz library. If there is a no match between the output ai and the ground truth y, we assign I \u0002 ai \u0338= y \u0003 7\u21921; otherwise, I \u0002 ai = y \u0003 7\u21920. To assess the range of correctness for each experiment, we compare the results of the 5 67k 21k 13k 11k 11k 14k 65k 59k 11k 5.3k 3.4k 2.5k 2.2k 2.2k 39k 11k 7.3k 6.1k 5.9k 5.9k 6.9k 0.0% 16.7% 33.3% 50.0% 66.7% 83.3% 100.0% 0 20k 40k 60k Abstractive Extractive Multiple Choice Expected Value Label Count At Least One Hallucination At Least One Hallucination Figure 3: Distribution of the expected value (number of hallucinations per example), rendered in percentage. For Extractive, additional context helps mitigate the rate of hallucination. For Multiple Choice, distractors can cause confusion amongst agents uniformly. However, for Abstractive, no additional information can cause massive disparities in correctness most of our simulations resulted in no hallucinations or all hallucinations. 67k 59k 39k 130k 27k 43k Abstractive Extractive Multiple Choice 0 50k 100k No Yes Scenario Label Count Figure 4: Binary distribution of labels, where at least one hallucination occurred during our simulation. original query q0 and its output a0 juxtaposed to the mode (most common) vote. In addition, we present the lower bound (all correct) and upper bound (at least one correct) accuracy across our n + 1 = 6 agents. Agreement Accuracy alone is insufficient for evaluating the agreement among multiple agents when analyzing truthful hallucinations. To address this limitation, we utilize several statistical measures: Item Difficulty (\u00b5D) (Lord, 1952), Fleiss\u2019s Generalized \u03ba (Cohen, 1960; Fleiss, 1971), Mean Certainty (H\u03b7) (Shannon, 1948; Wilcox, 1973), and Gibbs\u2019 M2 Index (Gibbs and Poston, 1975). These measures enable the assessment with regards to the level of agreement among independent samplings of a query, regardless of correctness. For example, if all LLM agents provide the same incorrect answer, high agreement indicates a misconception. Appendix D.3 expands in detail. Reliability We employ Cronbach\u2019s \u03b1 to assess the internal consistency of our Monte Carlo SimuBinary Train Val Test No (y = 0) 139,142 17,153 9,306 Yes (y = 1) 163,350 27,338 13,548 Expected Rate Train Val Test 0.0% (y = 0) 139,123 17,146 9,202 16.7% (y = 1) 35,114 4,974 2,757 33.3% (y = 2) 20,213 3,371 1,967 50.0% (y = 3) 15,749 2,757 1,768 66.7% (y = 4) 14,477 2,735 1,970 83.3% (y = 5) 17,123 3,242 2,171 100.0% (y = 6) 60,693 10,266 3,019 Scenario Train Val Test Extractive 80,049 5,843 Multiple Choice 45,997 14,127 21,573 Abstractive 176,446 24,521 1,281 Total 302,492 44,491 22,854 Table 2: Training Splits for HalluciBot. lation, as a corpus-level statistic (Cronbach, 1951). 5 Experimental Setup 5.1 Dataset Coverage & Scenario Split Our experiment includes 13 datasets (Table 11) divided into 3 scenarios: Extractive, Multiple Choice, and Abstractive. To evaluate the impact of context, we use SQuADv2 (Rajpurkar et al., 2016, 2018) to simulate Retrieval-Augmented Generation (Lewis et al., 2021; Guu et al., 2020). To assess the effect of multiple choice queries, we repurposed TruthfulQA (Lin et al., 2022) and SciQ (Johannes Welbl, 2017) for two experiments: one where the LLM agents select from the choices or context, and another where LLM agents generate outputs without context. Prompt templates for each output agent can be found in Table 8. We maintain the original train, validation, and test splits for each dataset across scenarios to prevent information leakage to HalluciBot. 6 Scenario Accuracy Agreement Rel Experiment # Base \u2191 Mode \u2191 Lower \u2191 Upper \u2191 \u00b5D \u2191 H\u03b7 \u2191 M2 \u2191 \u03ba \u2191 \u03b1 \u2191 WAVG Extractive 85,892 92.1 91.0 69.0 97.4 87.2 85.5 84.3 75.3 99.8 Multiple Choice 81,697 76.3 76.8 47.4 91.6 71.8 75.2 71.3 61.9 92.5 Abstractive 202,248 55.9 53.9 32.9 67.3 51.2 81.5 80.0 69.1 98.8 Total 369,837 68.6 67.4 44.3 79.4 63.9 81.0 79.1 69.0 97.7 Table 3: Stage 1 Monte Carlo: Weighted average (WAVG) results are presented for each scenario and in total. The metrics are calculated using data from all examples within each scenario, regardless of the original dataset split. For a more detailed breakdown, refer to Table 15. 5.2 Monte Carlo Parameters & Setup Our Monte Carlo Simulation utilizes 2 LLM components: the Query Perturbator and Output Generator. We use gpt-3.5-turbo to sample perturbations and outputs based on the prompts provided in Table 8. All LLM agents share the same set of parameters, as described in Table 6. Well-formedness We used a syntactically-aware well-formedness scoring RoBERTa model (Kumar, 2021) trained on the Google Query Wellformedness Dataset (Faruqui and Das, 2018) to evaluate the grammatical correctness and completeness of 1,881,005 synthetically generated queries. 5.3 HalluciBot Parameters & Environment We employed HuggingFace\u2019s Trainer class with the Adam optimizer (Kingma and Ba, 2017) for training, reporting efficiency and training times in Table 9. All experiments were conducted on an AWS EC2 instance with a single GPU (Table 7). Base Models We fine-tune pretrained BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) models (Table 10). To address label imbalance, we employed a weighted loss where each class weight is assigned to its inverted frequency in the training set. For the multi-class predictor, we trained a cross-entropy and ordinal model. Train Splits The train, validation, and test splits follow the original divisions of the datasets. Specifically, there are 302,492 training, 44,491 validation, and 22,854 testing samples. The distribution of labels across these splits is summarized in Table 2, and fine-grained splits per set are in Table 16. 6 Analysis & Discussion 6.1 Aggregated Monte Carlo Results The aggregated Monte Carlo results demonstrate the relative performance of different questionanswering scenarios. On average, the results (Table 3) indicate that Extractive outperforms Multiple Choice, which, in turn, outperforms Abstractive when subjected to perturbations. This trend suggests that the performance of gpt-3.5-turbo is influenced by the presence of additional content. Abstractive tasks show the greatest variation in agent response under perturbations, highlighting the effectiveness of added context in mitigating hallucinations (Figures 3 & 4). A full breakdown of our results is provided in Table 15. Extractive QA With context, gpt-3.5-turbo performs well on the SQuADv2 dataset. The mode accuracy (91.0%) and agreement (75.3%) of the agents are high. Even under radical perturbations, having unaltered context provides robustness to the agent\u2019s capacity to answer correctly. Multiple Choice QA Access to answer choices mitigates hallucinations across perturbations. The ensemble accuracy is slightly higher than the baseline (+0.5%), showcasing that multiple agents can (slightly) improve accuracy rates. Abstractive QA When no additional context is provided, gpt-3.5-turbo achieves a mode accuracy of 53.9% under perturbations. Interestingly, there is a significant dispersion of hallucination rates compared to other scenarios (Figure 3). Moreover, there is significant variation in results among datasets. For instance, SQuADv2 shows a -59% decrease in baseline accuracy against its Extractive counterpart. In contrast, SciQ benefits in this setting, leading to a +9.4% increase in mode accuracy, as the likelihood of generating a match increases. Coherence of Perturbations Our analysis indicates that the perturbations created by gpt-3.5-turbo consistently exhibit a high level of coherence, as indicated by their well-formedness score of 0.87. In contrast, the original queries achieve a well-formedness score of 0.77, representing an 11.5% decline. Table 13 expands on our results, with sample perturbations in Table 14. 7 Accuracy \u2191 F1 Score \u2191 Precision \u2191 Recall \u2191 Model Train Val Test Train Val Test Train Val Test Train Val Test BERT-base-cased 80.9 64.4 66.5 81.3 68.6 72.3 86.2 74.8 74.8 76.9 63.4 70.0 + Scenario 85.5 72.3 67.4 85.5 76.4 69.8 92.5 80.2 77.3 79.5 73.0 63.7 + \u03c4 = 0.129 81.6 71.1 67.7 84.6 79.0 75.2 77.1 71.4 69.1 93.7 88.4 82.5 RoBERTa-base 74.7 64.1 66.1 73.3 66.5 69.6 85.1 78.0 74.4 64.4 57.9 65.3 + Scenario 79.8 73.0 69.0 79.3 76.8 71.7 88.8 81.5 78.4 71.5 72.6 66.0 + \u03c4 = 0.177 77.5 72.1 68.7 81.2 79.6 75.9 74.0 72.3 69.9 89.8 88.6 83.0 RoBERTa-large + Scenario 84.9 72.9 68.6 85.0 76.9 71.1 92.1 80.9 78.2 78.8 73.2 65.3 + \u03c4 = 0.146 82.5 72.5 69.2 85.0 79.5 75.3 79.0 73.4 71.8 92.0 86.8 79.3 Table 4: HalluciBot Binary Evaluation Statistics. We report the Accuracy, F1, Precision, and Recall for all data splits. Probability threshold \u03c4 is computed along the closed interval [0, 1] in increments of 0.001 to maximize the F1 score. The best ablation per base model is underlined, while the overall best performing model is in bold. Top 1 Accuracy \u2191 Top 2 Accuracy \u2191 Top 3 Accuracy \u2191 \u00b11 Accuracy \u2191 Model Train Val Test Train Val Test Train Val Test Train Val Test BERT-base-cased 49.6 32.2 24.7 69.7 49.2 40.7 81.4 62.7 56.4 70.6 53.2 49.6 + Scenario 54.1 38.7 31.3 72.2 54.8 46.1 82.8 67.6 59.3 76.4 60.6 54.3 + Ordinal 58.7 45.3 38.6 70.0 54.5 48.3 79.0 64.1 59.1 75.9 61.4 55.8 RoBERTa-base 47.6 34.1 26.6 66.2 50.1 42.6 77.9 62.7 57.3 67.6 54.1 51.5 + Scenario 52.2 41.5 34.4 69.2 57.0 48.4 79.8 68.6 59.5 73.5 62.6 56.1 + Ordinal 47.8 39.4 37.1 56.7 48.7 46.6 67.0 60.0 57.8 64.1 56.1 54.4 RoBERTa-large + Scenario 61.6 47.6 38.8 77.5 62.6 53.1 85.8 73.0 63.8 79.5 65.4 58.8 + Ordinal 60.8 48.0 40.7 73.6 59.0 52.2 81.9 67.5 62.3 76.7 63.4 57.4 Table 5: HalluciBot Multi-class Evaluation Statistics. Considering the challenge of approximating a random variable and the potential presence of noise in our empirical estimate, we provide accuracy measurements for Top 1, Top 2, and Top 3 predictions. Additionally, we report the accuracy within a margin of \u00b11 of the predicted class. 6.2 Can HalluciBot detect Hallucination? Binary Results The best HalluciBot achieves a validation accuracy of 73.0%, with a macro F1score of 76.9% (RoBERTa-base + Scenario). With threshold tuning, improved recall and precision rates yield a F1-score of 79.6%. A full breakdown of our results is provided in Table 4. Multi-class Results HalluciBot achieves a validation accuracy of 47.6%, with a Top 3 accuracy of 73.0% for the RoBERTa-large + Scenario model. Does Knowing the Scenario Matter? Including the scenario in HalluciBot\u2019s prompt has proven beneficial in distinguishing similar queries with varying contexts. For binary cases, incorporating the scenario leads to a notable increase of +15.5% in F1. For multi-class validation, the inclusion of scenario results in a +21.7% increase in Top 1 Accuracy and a +9.4% increase in Top 3 Accuracy. Test Distribution Shift The testing set poses three challenges: (1) the absence of Extractive samples; (2) an over-representation of Multiple Choice queries (94.4%); and (3) a distribution of multi-class labels that diverges with significantly fewer hallucinating-inducing queries compared to the training and validation sets. Consequently, we rely on the validation results instead. 7 Conclusion We propose a heretofore unexplored realm of hallucination mitigation predicting a query\u2019s hallucination probability. HalluciBot empirically estimates how the query itself may induce hallucination. HalluciBot\u2019s training corpus consists of diverse scenarios and domains to ensure robustness; we score an accuracy level of 73% for binary and top three multi-class labels. In essence, HalluciBot lays the foundation for a reproducible method of hallucination prediction that absorbs the cost of iterative generations during training. Moreover, the training process of HalluciBot delivers insight into the performance of gpt-3.5-turbo on lexical perturbations for 13 datasets and present discoveries on how perturbations may induce hallucinations. Practically, institutions can implement HalluciBot to measure user accountability penalizing those who input queries that have a hallucination probability above a certain threshold. Thus, HalluciBot\u2019s academic and practical contributions add to the ever-growing concerted effort of enabling a robust language generation ecosystem for society. 8 Limitations Despite the promising results of HalluciBot, there are limitations to be acknowledged. Firstly, our approach relies on automated LLM crowdsourcing to generate query variations, which can introduce a certain degree of variability and noise in the data. Although we took measures to ensure the quality of LLM-crowdsourced data through examining wellformedness, there may be bias or error amongst variations. Secondly, our approach is computationally expensive during training, requiring multiple LLM agents and generating n variations per query. Thirdly, our datasets are English-centric, so our experimental results for robustness and consistency can only be interpreted and applied to English. Thus, it may be difficult to conclude if HalluciBot could withstand stress-testing if every query was perturbed into a new language, given that our BERT and RoBERTa backbones are Englishcentric. Fourthly, HalluciBot is trained soley on the outputs of gpt-3.5-turbo; however, as the framework is adaptable, future studies can train HalluciBot on any mixture of LLMs. Finally, there may be inherent bias in using gpt-3.5-turbo as both the Query Perturbator and the Output Generator. However, even though there is significant agreement regardless of correctness (Table 17), we feel confident that this is not a limiting factor of our work. Since every label is observed in significant quantities, we can assume that perturbations helped decouple semantic bias from gpt-3.5-turbo\u2019s general knowledge. Disclaimer This paper was prepared for informational purposes by the Artificial Intelligence Research group of JPMorgan Chase & Co. and its affiliates (\u201cJP Morgan\u201d), and is not a product of the Research Department of JP Morgan. JP Morgan makes no representation and warranty whatsoever and disclaims all liability, for the completeness, accuracy or reliability of the information contained herein. This document is not intended as investment research or investment advice, or a recommendation, offer or solicitation for the purchase or sale of any security, financial instrument, financial product or service, or to be used in any way for evaluating the merits of participating in any transaction, and shall not constitute a solicitation under any jurisdiction or to any person, if such solicitation under such jurisdiction or to such person would be unlawful."
+                },
+                {
+                    "url": "http://arxiv.org/abs/2005.11401v4",
+                    "title": "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks",
+                    "abstract": "Large pre-trained language models have been shown to store factual knowledge\nin their parameters, and achieve state-of-the-art results when fine-tuned on\ndownstream NLP tasks. However, their ability to access and precisely manipulate\nknowledge is still limited, and hence on knowledge-intensive tasks, their\nperformance lags behind task-specific architectures. Additionally, providing\nprovenance for their decisions and updating their world knowledge remain open\nresearch problems. Pre-trained models with a differentiable access mechanism to\nexplicit non-parametric memory can overcome this issue, but have so far been\nonly investigated for extractive downstream tasks. We explore a general-purpose\nfine-tuning recipe for retrieval-augmented generation (RAG) -- models which\ncombine pre-trained parametric and non-parametric memory for language\ngeneration. We introduce RAG models where the parametric memory is a\npre-trained seq2seq model and the non-parametric memory is a dense vector index\nof Wikipedia, accessed with a pre-trained neural retriever. We compare two RAG\nformulations, one which conditions on the same retrieved passages across the\nwhole generated sequence, the other can use different passages per token. We\nfine-tune and evaluate our models on a wide range of knowledge-intensive NLP\ntasks and set the state-of-the-art on three open domain QA tasks, outperforming\nparametric seq2seq models and task-specific retrieve-and-extract architectures.\nFor language generation tasks, we find that RAG models generate more specific,\ndiverse and factual language than a state-of-the-art parametric-only seq2seq\nbaseline.",
+                    "authors": "Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K\u00fcttler, Mike Lewis, Wen-tau Yih, Tim Rockt\u00e4schel, Sebastian Riedel, Douwe Kiela",
+                    "published": "2020-05-22",
+                    "updated": "2021-04-12",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1901.04085v5",
+                    "title": "Passage Re-ranking with BERT",
+                    "abstract": "Recently, neural models pretrained on a language modeling task, such as ELMo\n(Peters et al., 2017), OpenAI GPT (Radford et al., 2018), and BERT (Devlin et\nal., 2018), have achieved impressive results on various natural language\nprocessing tasks such as question-answering and natural language inference. In\nthis paper, we describe a simple re-implementation of BERT for query-based\npassage re-ranking. Our system is the state of the art on the TREC-CAR dataset\nand the top entry in the leaderboard of the MS MARCO passage retrieval task,\noutperforming the previous state of the art by 27% (relative) in MRR@10. The\ncode to reproduce our results is available at\nhttps://github.com/nyu-dl/dl4marco-bert",
+                    "authors": "Rodrigo Nogueira, Kyunghyun Cho",
+                    "published": "2019-01-13",
+                    "updated": "2020-04-14",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2004.04906v3",
+                    "title": "Dense Passage Retrieval for Open-Domain Question Answering",
+                    "abstract": "Open-domain question answering relies on efficient passage retrieval to\nselect candidate contexts, where traditional sparse vector space models, such\nas TF-IDF or BM25, are the de facto method. In this work, we show that\nretrieval can be practically implemented using dense representations alone,\nwhere embeddings are learned from a small number of questions and passages by a\nsimple dual-encoder framework. When evaluated on a wide range of open-domain QA\ndatasets, our dense retriever outperforms a strong Lucene-BM25 system largely\nby 9%-19% absolute in terms of top-20 passage retrieval accuracy, and helps our\nend-to-end QA system establish new state-of-the-art on multiple open-domain QA\nbenchmarks.",
+                    "authors": "Vladimir Karpukhin, Barlas O\u011fuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, Wen-tau Yih",
+                    "published": "2020-04-10",
+                    "updated": "2020-09-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2203.06566v1",
+                    "title": "PromptChainer: Chaining Large Language Model Prompts through Visual Programming",
+                    "abstract": "While LLMs can effectively help prototype single ML functionalities, many\nreal-world applications involve complex tasks that cannot be easily handled via\na single run of an LLM. Recent work has found that chaining multiple LLM runs\ntogether (with the output of one step being the input to the next) can help\nusers accomplish these more complex tasks, and in a way that is perceived to be\nmore transparent and controllable. However, it remains unknown what users need\nwhen authoring their own LLM chains -- a key step for lowering the barriers for\nnon-AI-experts to prototype AI-infused applications. In this work, we explore\nthe LLM chain authoring process. We conclude from pilot studies find that\nchaining requires careful scaffolding for transforming intermediate node\noutputs, as well as debugging the chain at multiple granularities; to help with\nthese needs, we designed PromptChainer, an interactive interface for visually\nprogramming chains. Through case studies with four people, we show that\nPromptChainer supports building prototypes for a range of applications, and\nconclude with open questions on scaling chains to complex tasks, and supporting\nlow-fi chain prototyping.",
+                    "authors": "Tongshuang Wu, Ellen Jiang, Aaron Donsbach, Jeff Gray, Alejandra Molina, Michael Terry, Carrie J Cai",
+                    "published": "2022-03-13",
+                    "updated": "2022-03-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2207.05221v4",
+                    "title": "Language Models (Mostly) Know What They Know",
+                    "abstract": "We study whether language models can evaluate the validity of their own\nclaims and predict which questions they will be able to answer correctly. We\nfirst show that larger models are well-calibrated on diverse multiple choice\nand true/false questions when they are provided in the right format. Thus we\ncan approach self-evaluation on open-ended sampling tasks by asking models to\nfirst propose answers, and then to evaluate the probability \"P(True)\" that\ntheir answers are correct. We find encouraging performance, calibration, and\nscaling for P(True) on a diverse array of tasks. Performance at self-evaluation\nfurther improves when we allow models to consider many of their own samples\nbefore predicting the validity of one specific possibility. Next, we\ninvestigate whether models can be trained to predict \"P(IK)\", the probability\nthat \"I know\" the answer to a question, without reference to any particular\nproposed answer. Models perform well at predicting P(IK) and partially\ngeneralize across tasks, though they struggle with calibration of P(IK) on new\ntasks. The predicted P(IK) probabilities also increase appropriately in the\npresence of relevant source materials in the context, and in the presence of\nhints towards the solution of mathematical word problems. We hope these\nobservations lay the groundwork for training more honest models, and for\ninvestigating how honesty generalizes to cases where models are trained on\nobjectives other than the imitation of human writing.",
+                    "authors": "Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, Scott Johnston, Sheer El-Showk, Andy Jones, Nelson Elhage, Tristan Hume, Anna Chen, Yuntao Bai, Sam Bowman, Stanislav Fort, Deep Ganguli, Danny Hernandez, Josh Jacobson, Jackson Kernion, Shauna Kravec, Liane Lovitt, Kamal Ndousse, Catherine Olsson, Sam Ringer, Dario Amodei, Tom Brown, Jack Clark, Nicholas Joseph, Ben Mann, Sam McCandlish, Chris Olah, Jared Kaplan",
+                    "published": "2022-07-11",
+                    "updated": "2022-11-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2005.00700v3",
+                    "title": "UnifiedQA: Crossing Format Boundaries With a Single QA System",
+                    "abstract": "Question answering (QA) tasks have been posed using a variety of formats,\nsuch as extractive span selection, multiple choice, etc. This has led to\nformat-specialized models, and even to an implicit division in the QA\ncommunity. We argue that such boundaries are artificial and perhaps\nunnecessary, given the reasoning abilities we seek to teach are not governed by\nthe format. As evidence, we use the latest advances in language modeling to\nbuild a single pre-trained QA model, UnifiedQA, that performs surprisingly well\nacross 17 QA datasets spanning 4 diverse formats. UnifiedQA performs on par\nwith 9 different models that were trained on individual datasets themselves.\nEven when faced with 12 unseen datasets of observed formats, UnifiedQA performs\nsurprisingly well, showing strong generalization from its out-of-format\ntraining data. Finally, simply fine-tuning this pre-trained QA model into\nspecialized models results in a new state of the art on 6 datasets,\nestablishing UnifiedQA as a strong starting point for building QA systems.",
+                    "authors": "Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, Hannaneh Hajishirzi",
+                    "published": "2020-05-02",
+                    "updated": "2020-10-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2007.01282v2",
+                    "title": "Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering",
+                    "abstract": "Generative models for open domain question answering have proven to be\ncompetitive, without resorting to external knowledge. While promising, this\napproach requires to use models with billions of parameters, which are\nexpensive to train and query. In this paper, we investigate how much these\nmodels can benefit from retrieving text passages, potentially containing\nevidence. We obtain state-of-the-art results on the Natural Questions and\nTriviaQA open benchmarks. Interestingly, we observe that the performance of\nthis method significantly improves when increasing the number of retrieved\npassages. This is evidence that generative models are good at aggregating and\ncombining evidence from multiple passages.",
+                    "authors": "Gautier Izacard, Edouard Grave",
+                    "published": "2020-07-02",
+                    "updated": "2021-02-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2209.10063v3",
+                    "title": "Generate rather than Retrieve: Large Language Models are Strong Context Generators",
+                    "abstract": "Knowledge-intensive tasks, such as open-domain question answering (QA),\nrequire access to a large amount of world or domain knowledge. A common\napproach for knowledge-intensive tasks is to employ a retrieve-then-read\npipeline that first retrieves a handful of relevant contextual documents from\nan external corpus such as Wikipedia and then predicts an answer conditioned on\nthe retrieved documents. In this paper, we present a novel perspective for\nsolving knowledge-intensive tasks by replacing document retrievers with large\nlanguage model generators. We call our method generate-then-read (GenRead),\nwhich first prompts a large language model to generate contextutal documents\nbased on a given question, and then reads the generated documents to produce\nthe final answer. Furthermore, we propose a novel clustering-based prompting\nmethod that selects distinct prompts, resulting in the generated documents that\ncover different perspectives, leading to better recall over acceptable answers.\nWe conduct extensive experiments on three different knowledge-intensive tasks,\nincluding open-domain QA, fact checking, and dialogue system. Notably, GenRead\nachieves 71.6 and 54.4 exact match scores on TriviaQA and WebQ, significantly\noutperforming the state-of-the-art retrieve-then-read pipeline DPR-FiD by +4.0\nand +3.9, without retrieving any documents from any external knowledge source.\nLastly, we demonstrate the model performance can be further improved by\ncombining retrieval and generation. Our code and generated documents can be\nfound at https://github.com/wyu97/GenRead.",
+                    "authors": "Wenhao Yu, Dan Iter, Shuohang Wang, Yichong Xu, Mingxuan Ju, Soumya Sanyal, Chenguang Zhu, Michael Zeng, Meng Jiang",
+                    "published": "2022-09-21",
+                    "updated": "2023-01-25",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2112.00114v1",
+                    "title": "Show Your Work: Scratchpads for Intermediate Computation with Language Models",
+                    "abstract": "Large pre-trained language models perform remarkably well on tasks that can\nbe done \"in one pass\", such as generating realistic text or synthesizing\ncomputer programs. However, they struggle with tasks that require unbounded\nmulti-step computation, such as adding integers or executing programs.\nSurprisingly, we find that these same models are able to perform complex\nmulti-step computations -- even in the few-shot regime -- when asked to perform\nthe operation \"step by step\", showing the results of intermediate computations.\nIn particular, we train transformers to perform multi-step computations by\nasking them to emit intermediate computation steps into a \"scratchpad\". On a\nseries of increasingly complex tasks ranging from long addition to the\nexecution of arbitrary programs, we show that scratchpads dramatically improve\nthe ability of language models to perform multi-step computations.",
+                    "authors": "Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, Charles Sutton, Augustus Odena",
+                    "published": "2021-11-30",
+                    "updated": "2021-11-30",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.14613v2",
+                    "title": "Selectively Answering Ambiguous Questions",
+                    "abstract": "Trustworthy language models should abstain from answering questions when they\ndo not know the answer. However, the answer to a question can be unknown for a\nvariety of reasons. Prior research has focused on the case in which the\nquestion is clear and the answer is unambiguous but possibly unknown, but the\nanswer to a question can also be unclear due to uncertainty of the questioner's\nintent or context. We investigate question answering from this perspective,\nfocusing on answering a subset of questions with a high degree of accuracy,\nfrom a set of questions in which many are inherently ambiguous. In this\nsetting, we find that the most reliable approach to decide when to abstain\ninvolves quantifying repetition within sampled model outputs, rather than the\nmodel's likelihood or self-verification as used in prior work. We find this to\nbe the case across different types of uncertainty and model scales,and with or\nwithout instruction tuning. Our results suggest that sampling-based confidence\nscores help calibrate answers to relatively unambiguous questions, with more\ndramatic improvements on ambiguous questions.",
+                    "authors": "Jeremy R. Cole, Michael J. Q. Zhang, Daniel Gillick, Julian Martin Eisenschlos, Bhuwan Dhingra, Jacob Eisenstein",
+                    "published": "2023-05-24",
+                    "updated": "2023-11-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2005.11401v4",
+                    "title": "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks",
+                    "abstract": "Large pre-trained language models have been shown to store factual knowledge\nin their parameters, and achieve state-of-the-art results when fine-tuned on\ndownstream NLP tasks. However, their ability to access and precisely manipulate\nknowledge is still limited, and hence on knowledge-intensive tasks, their\nperformance lags behind task-specific architectures. Additionally, providing\nprovenance for their decisions and updating their world knowledge remain open\nresearch problems. Pre-trained models with a differentiable access mechanism to\nexplicit non-parametric memory can overcome this issue, but have so far been\nonly investigated for extractive downstream tasks. We explore a general-purpose\nfine-tuning recipe for retrieval-augmented generation (RAG) -- models which\ncombine pre-trained parametric and non-parametric memory for language\ngeneration. We introduce RAG models where the parametric memory is a\npre-trained seq2seq model and the non-parametric memory is a dense vector index\nof Wikipedia, accessed with a pre-trained neural retriever. We compare two RAG\nformulations, one which conditions on the same retrieved passages across the\nwhole generated sequence, the other can use different passages per token. We\nfine-tune and evaluate our models on a wide range of knowledge-intensive NLP\ntasks and set the state-of-the-art on three open domain QA tasks, outperforming\nparametric seq2seq models and task-specific retrieve-and-extract architectures.\nFor language generation tasks, we find that RAG models generate more specific,\ndiverse and factual language than a state-of-the-art parametric-only seq2seq\nbaseline.",
+                    "authors": "Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K\u00fcttler, Mike Lewis, Wen-tau Yih, Tim Rockt\u00e4schel, Sebastian Riedel, Douwe Kiela",
+                    "published": "2020-05-22",
+                    "updated": "2021-04-12",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1702.08734v1",
+                    "title": "Billion-scale similarity search with GPUs",
+                    "abstract": "Similarity search finds application in specialized database systems handling\ncomplex data such as images or videos, which are typically represented by\nhigh-dimensional features and require specific indexing structures. This paper\ntackles the problem of better utilizing GPUs for this task. While GPUs excel at\ndata-parallel tasks, prior approaches are bottlenecked by algorithms that\nexpose less parallelism, such as k-min selection, or make poor use of the\nmemory hierarchy.\n  We propose a design for k-selection that operates at up to 55% of theoretical\npeak performance, enabling a nearest neighbor implementation that is 8.5x\nfaster than prior GPU state of the art. We apply it in different similarity\nsearch scenarios, by proposing optimized design for brute-force, approximate\nand compressed-domain search based on product quantization. In all these\nsetups, we outperform the state of the art by large margins. Our implementation\nenables the construction of a high accuracy k-NN graph on 95 million images\nfrom the Yfcc100M dataset in 35 minutes, and of a graph connecting 1 billion\nvectors in less than 12 hours on 4 Maxwell Titan X GPUs. We have open-sourced\nour approach for the sake of comparison and reproducibility.",
+                    "authors": "Jeff Johnson, Matthijs Douze, Herv\u00e9 J\u00e9gou",
+                    "published": "2017-02-28",
+                    "updated": "2017-02-28",
+                    "primary_cat": "cs.CV",
+                    "cats": [
+                        "cs.CV",
+                        "cs.DB",
+                        "cs.DS",
+                        "cs.IR"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2202.03629v6",
+                    "title": "Survey of Hallucination in Natural Language Generation",
+                    "abstract": "Natural Language Generation (NLG) has improved exponentially in recent years\nthanks to the development of sequence-to-sequence deep learning technologies\nsuch as Transformer-based language models. This advancement has led to more\nfluent and coherent NLG, leading to improved development in downstream tasks\nsuch as abstractive summarization, dialogue generation and data-to-text\ngeneration. However, it is also apparent that deep learning based generation is\nprone to hallucinate unintended text, which degrades the system performance and\nfails to meet user expectations in many real-world scenarios. To address this\nissue, many studies have been presented in measuring and mitigating\nhallucinated texts, but these have never been reviewed in a comprehensive\nmanner before. In this survey, we thus provide a broad overview of the research\nprogress and challenges in the hallucination problem in NLG. The survey is\norganized into two parts: (1) a general overview of metrics, mitigation\nmethods, and future directions; (2) an overview of task-specific research\nprogress on hallucinations in the following downstream tasks, namely\nabstractive summarization, dialogue generation, generative question answering,\ndata-to-text generation, machine translation, and visual-language generation;\nand (3) hallucinations in large language models (LLMs). This survey serves to\nfacilitate collaborative efforts among researchers in tackling the challenge of\nhallucinated texts in NLG.",
+                    "authors": "Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Yejin Bang, Delong Chen, Ho Shu Chan, Wenliang Dai, Andrea Madotto, Pascale Fung",
+                    "published": "2022-02-08",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "A.1"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2201.11903v6",
+                    "title": "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models",
+                    "abstract": "We explore how generating a chain of thought -- a series of intermediate\nreasoning steps -- significantly improves the ability of large language models\nto perform complex reasoning. In particular, we show how such reasoning\nabilities emerge naturally in sufficiently large language models via a simple\nmethod called chain of thought prompting, where a few chain of thought\ndemonstrations are provided as exemplars in prompting. Experiments on three\nlarge language models show that chain of thought prompting improves performance\non a range of arithmetic, commonsense, and symbolic reasoning tasks. The\nempirical gains can be striking. For instance, prompting a 540B-parameter\nlanguage model with just eight chain of thought exemplars achieves state of the\nart accuracy on the GSM8K benchmark of math word problems, surpassing even\nfinetuned GPT-3 with a verifier.",
+                    "authors": "Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, Denny Zhou",
+                    "published": "2022-01-28",
+                    "updated": "2023-01-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14183v2",
+                    "title": "On Early Detection of Hallucinations in Factual Question Answering",
+                    "abstract": "While large language models (LLMs) have taken great strides towards helping\nhumans with a plethora of tasks like search and summarization, hallucinations\nremain a major impediment towards gaining user trust. The fluency and coherence\nof model generations even when hallucinating makes it difficult to detect\nwhether or not a model is hallucinating. In this work, we explore if the\nartifacts associated with the model generations can provide hints that the\ngeneration will contain hallucinations. Specifically, we probe LLMs at 1) the\ninputs via Integrated Gradients based token attribution, 2) the outputs via the\nSoftmax probabilities, and 3) the internal state via self-attention and\nfully-connected layer activations for signs of hallucinations on open-ended\nquestion answering tasks. Our results show that the distributions of these\nartifacts differ between hallucinated and non-hallucinated generations.\nBuilding on this insight, we train binary classifiers that use these artifacts\nas input features to classify model generations into hallucinations and\nnon-hallucinations. These hallucination classifiers achieve up to 0.80 AUROC.\nWe further show that tokens preceding a hallucination can predict the\nsubsequent hallucination before it occurs.",
+                    "authors": "Ben Snyder, Marius Moisescu, Muhammad Bilal Zafar",
+                    "published": "2023-12-19",
+                    "updated": "2023-12-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.03987v2",
+                    "title": "A Stitch in Time Saves Nine: Detecting and Mitigating Hallucinations of LLMs by Validating Low-Confidence Generation",
+                    "abstract": "Recently developed large language models have achieved remarkable success in\ngenerating fluent and coherent text. However, these models often tend to\n'hallucinate' which critically hampers their reliability. In this work, we\naddress this crucial problem and propose an approach that actively detects and\nmitigates hallucinations during the generation process. Specifically, we first\nidentify the candidates of potential hallucination leveraging the model's logit\noutput values, check their correctness through a validation procedure, mitigate\nthe detected hallucinations, and then continue with the generation process.\nThrough extensive experiments with GPT-3.5 (text-davinci-003) on the 'article\ngeneration task', we first demonstrate the individual efficacy of our detection\nand mitigation techniques. Specifically, the detection technique achieves a\nrecall of ~88% and the mitigation technique successfully mitigates 57.6% of the\ncorrectly detected hallucinations. Importantly, our mitigation technique does\nnot introduce new hallucinations even in the case of incorrectly detected\nhallucinations, i.e., false positives. Then, we show that the proposed active\ndetection and mitigation approach successfully reduces the hallucinations of\nthe GPT-3.5 model from 47.5% to 14.5% on average. We further demonstrate the\neffectiveness and wide applicability of our approach through additional studies\nincluding performance on different types of questions (multi-hop and false\npremise questions) and with another LLM from a different model family (Vicuna).\nIn summary, our work contributes to improving the reliability and\ntrustworthiness of large language models, a crucial step en route to enabling\ntheir widespread adoption in real-world applications.",
+                    "authors": "Neeraj Varshney, Wenlin Yao, Hongming Zhang, Jianshu Chen, Dong Yu",
+                    "published": "2023-07-08",
+                    "updated": "2023-08-12",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2203.11171v4",
+                    "title": "Self-Consistency Improves Chain of Thought Reasoning in Language Models",
+                    "abstract": "Chain-of-thought prompting combined with pre-trained large language models\nhas achieved encouraging results on complex reasoning tasks. In this paper, we\npropose a new decoding strategy, self-consistency, to replace the naive greedy\ndecoding used in chain-of-thought prompting. It first samples a diverse set of\nreasoning paths instead of only taking the greedy one, and then selects the\nmost consistent answer by marginalizing out the sampled reasoning paths.\nSelf-consistency leverages the intuition that a complex reasoning problem\ntypically admits multiple different ways of thinking leading to its unique\ncorrect answer. Our extensive empirical evaluation shows that self-consistency\nboosts the performance of chain-of-thought prompting with a striking margin on\na range of popular arithmetic and commonsense reasoning benchmarks, including\nGSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), StrategyQA (+6.4%) and\nARC-challenge (+3.9%).",
+                    "authors": "Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, Denny Zhou",
+                    "published": "2022-03-21",
+                    "updated": "2023-03-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/1908.10084v1",
+                    "title": "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
+                    "abstract": "BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) has set a new\nstate-of-the-art performance on sentence-pair regression tasks like semantic\ntextual similarity (STS). However, it requires that both sentences are fed into\nthe network, which causes a massive computational overhead: Finding the most\nsimilar pair in a collection of 10,000 sentences requires about 50 million\ninference computations (~65 hours) with BERT. The construction of BERT makes it\nunsuitable for semantic similarity search as well as for unsupervised tasks\nlike clustering.\n  In this publication, we present Sentence-BERT (SBERT), a modification of the\npretrained BERT network that use siamese and triplet network structures to\nderive semantically meaningful sentence embeddings that can be compared using\ncosine-similarity. This reduces the effort for finding the most similar pair\nfrom 65 hours with BERT / RoBERTa to about 5 seconds with SBERT, while\nmaintaining the accuracy from BERT.\n  We evaluate SBERT and SRoBERTa on common STS tasks and transfer learning\ntasks, where it outperforms other state-of-the-art sentence embeddings methods.",
+                    "authors": "Nils Reimers, Iryna Gurevych",
+                    "published": "2019-08-27",
+                    "updated": "2019-08-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00258v1",
+                    "title": "Noisy Exemplars Make Large Language Models More Robust: A Domain-Agnostic Behavioral Analysis",
+                    "abstract": "Recent advances in prompt engineering enable large language models (LLMs) to\nsolve multi-hop logical reasoning problems with impressive accuracy. However,\nthere is little existing work investigating the robustness of LLMs with\nfew-shot prompting techniques. Therefore, we introduce a systematic approach to\ntest the robustness of LLMs in multi-hop reasoning tasks via domain-agnostic\nperturbations. We include perturbations at multiple levels of abstractions\n(e.g. lexical perturbations such as typos, and semantic perturbations such as\nthe inclusion of intermediate reasoning steps in the questions) to conduct\nbehavioral analysis on the LLMs. Throughout our experiments, we find that\nmodels are more sensitive to certain perturbations such as replacing words with\ntheir synonyms. We also demonstrate that increasing the proportion of perturbed\nexemplars in the prompts improves the robustness of few-shot prompting methods.",
+                    "authors": "Hongyi Zheng, Abulhair Saparov",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2205.12558v2",
+                    "title": "Gradient-Based Constrained Sampling from Language Models",
+                    "abstract": "Large pretrained language models generate fluent text but are notoriously\nhard to controllably sample from. In this work, we study constrained sampling\nfrom such language models: generating text that satisfies user-defined\nconstraints, while maintaining fluency and the model's performance in a\ndownstream task. We propose MuCoLa -- a sampling procedure that combines the\nlog-likelihood of the language model with arbitrary (differentiable)\nconstraints in a single energy function, and then generates samples in a\nnon-autoregressive manner. Specifically, it initializes the entire output\nsequence with noise and follows a Markov chain defined by Langevin Dynamics\nusing the gradients of the energy function. We evaluate MuCoLa on text\ngeneration with soft and hard constraints as well as their combinations\nobtaining significant improvements over competitive baselines for toxicity\navoidance, sentiment control, and keyword-guided generation.",
+                    "authors": "Sachin Kumar, Biswajit Paria, Yulia Tsvetkov",
+                    "published": "2022-05-25",
+                    "updated": "2022-11-08",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2207.10342v2",
+                    "title": "Language Model Cascades",
+                    "abstract": "Prompted models have demonstrated impressive few-shot learning abilities.\nRepeated interactions at test-time with a single model, or the composition of\nmultiple models together, further expands capabilities. These compositions are\nprobabilistic models, and may be expressed in the language of graphical models\nwith random variables whose values are complex data types such as strings.\nCases with control flow and dynamic structure require techniques from\nprobabilistic programming, which allow implementing disparate model structures\nand inference strategies in a unified language. We formalize several existing\ntechniques from this perspective, including scratchpads / chain of thought,\nverifiers, STaR, selection-inference, and tool use. We refer to the resulting\nprograms as language model cascades.",
+                    "authors": "David Dohan, Winnie Xu, Aitor Lewkowycz, Jacob Austin, David Bieber, Raphael Gontijo Lopes, Yuhuai Wu, Henryk Michalewski, Rif A. Saurous, Jascha Sohl-dickstein, Kevin Murphy, Charles Sutton",
+                    "published": "2022-07-21",
+                    "updated": "2022-07-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "label": "Related Work"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18580v1",
+                    "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+                    "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+                    "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+                    "published": "2023-11-30",
+                    "updated": "2023-11-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.04057v1",
+                    "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+                    "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+                    "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+                    "published": "2024-01-08",
+                    "updated": "2024-01-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.18140v1",
+                    "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+                    "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+                    "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+                    "published": "2023-11-29",
+                    "updated": "2023-11-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.18276v1",
+                    "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+                    "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+                    "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+                    "published": "2024-04-28",
+                    "updated": "2024-04-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "D.1; I.2"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.12736v1",
+                    "title": "Large Language Model Supply Chain: A Research Agenda",
+                    "abstract": "The rapid advancements in pre-trained Large Language Models (LLMs) and Large\nMultimodal Models (LMMs) have ushered in a new era of intelligent applications,\ntransforming fields ranging from natural language processing to content\ngeneration. The LLM supply chain represents a crucial aspect of the\ncontemporary artificial intelligence landscape. It encompasses the entire\nlifecycle of pre-trained models, from its initial development and training to\nits final deployment and application in various domains. This paper presents a\ncomprehensive overview of the LLM supply chain, highlighting its three core\nelements: 1) the model infrastructure, encompassing datasets and toolchain for\ntraining, optimization, and deployment; 2) the model lifecycle, covering\ntraining, testing, releasing, and ongoing maintenance; and 3) the downstream\napplication ecosystem, enabling the integration of pre-trained models into a\nwide range of intelligent applications. However, this rapidly evolving field\nfaces numerous challenges across these key components, including data privacy\nand security, model interpretability and fairness, infrastructure scalability,\nand regulatory compliance. Addressing these challenges is essential for\nharnessing the full potential of LLMs and ensuring their ethical and\nresponsible use. This paper provides a future research agenda for the LLM\nsupply chain, aiming at driving the continued advancement and responsible\ndeployment of these transformative LLMs.",
+                    "authors": "Shenao Wang, Yanjie Zhao, Xinyi Hou, Haoyu Wang",
+                    "published": "2024-04-19",
+                    "updated": "2024-04-19",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04205v2",
+                    "title": "Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves",
+                    "abstract": "Misunderstandings arise not only in interpersonal communication but also\nbetween humans and Large Language Models (LLMs). Such discrepancies can make\nLLMs interpret seemingly unambiguous questions in unexpected ways, yielding\nincorrect responses. While it is widely acknowledged that the quality of a\nprompt, such as a question, significantly impacts the quality of the response\nprovided by LLMs, a systematic method for crafting questions that LLMs can\nbetter comprehend is still underdeveloped. In this paper, we present a method\nnamed `Rephrase and Respond' (RaR), which allows LLMs to rephrase and expand\nquestions posed by humans and provide responses in a single prompt. This\napproach serves as a simple yet effective prompting method for improving\nperformance. We also introduce a two-step variant of RaR, where a rephrasing\nLLM first rephrases the question and then passes the original and rephrased\nquestions together to a different responding LLM. This facilitates the\neffective utilization of rephrased questions generated by one LLM with another.\nOur experiments demonstrate that our methods significantly improve the\nperformance of different models across a wide range to tasks. We further\nprovide a comprehensive comparison between RaR and the popular Chain-of-Thought\n(CoT) methods, both theoretically and empirically. We show that RaR is\ncomplementary to CoT and can be combined with CoT to achieve even better\nperformance. Our work not only contributes to enhancing LLM performance\nefficiently and effectively but also sheds light on a fair evaluation of LLM\ncapabilities. Data and codes are available at\nhttps://github.com/uclaml/Rephrase-and-Respond.",
+                    "authors": "Yihe Deng, Weitong Zhang, Zixiang Chen, Quanquan Gu",
+                    "published": "2023-11-07",
+                    "updated": "2024-04-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.02839v1",
+                    "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+                    "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+                    "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+                    "published": "2024-03-05",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2206.13757v1",
+                    "title": "Flexible text generation for counterfactual fairness probing",
+                    "abstract": "A common approach for testing fairness issues in text-based classifiers is\nthrough the use of counterfactuals: does the classifier output change if a\nsensitive attribute in the input is changed? Existing counterfactual generation\nmethods typically rely on wordlists or templates, producing simple\ncounterfactuals that don't take into account grammar, context, or subtle\nsensitive attribute references, and could miss issues that the wordlist\ncreators had not considered. In this paper, we introduce a task for generating\ncounterfactuals that overcomes these shortcomings, and demonstrate how large\nlanguage models (LLMs) can be leveraged to make progress on this task. We show\nthat this LLM-based method can produce complex counterfactuals that existing\nmethods cannot, comparing the performance of various counterfactual generation\nmethods on the Civil Comments dataset and showing their value in evaluating a\ntoxicity classifier.",
+                    "authors": "Zee Fryer, Vera Axelrod, Ben Packer, Alex Beutel, Jilin Chen, Kellie Webster",
+                    "published": "2022-06-28",
+                    "updated": "2022-06-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.07688v1",
+                    "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+                    "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+                    "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+                    "published": "2024-02-12",
+                    "updated": "2024-02-12",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.10567v3",
+                    "title": "InSaAF: Incorporating Safety through Accuracy and Fairness | Are LLMs ready for the Indian Legal Domain?",
+                    "abstract": "Recent advancements in language technology and Artificial Intelligence have\nresulted in numerous Language Models being proposed to perform various tasks in\nthe legal domain ranging from predicting judgments to generating summaries.\nDespite their immense potential, these models have been proven to learn and\nexhibit societal biases and make unfair predictions. In this study, we explore\nthe ability of Large Language Models (LLMs) to perform legal tasks in the\nIndian landscape when social factors are involved. We present a novel metric,\n$\\beta$-weighted $\\textit{Legal Safety Score ($LSS_{\\beta}$)}$, which\nencapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs'\nsafety by considering its performance in the $\\textit{Binary Statutory\nReasoning}$ task and its fairness exhibition with respect to various axes of\ndisparities in the Indian society. Task performance and fairness scores of\nLLaMA and LLaMA--2 models indicate that the proposed $LSS_{\\beta}$ metric can\neffectively determine the readiness of a model for safe usage in the legal\nsector. We also propose finetuning pipelines, utilising specialised legal\ndatasets, as a potential method to mitigate bias and improve model safety. The\nfinetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\\beta}$,\nimproving their usability in the Indian legal domain. Our code is publicly\nreleased.",
+                    "authors": "Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru",
+                    "published": "2024-02-16",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.13925v1",
+                    "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+                    "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+                    "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+                    "published": "2024-04-22",
+                    "updated": "2024-04-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11406v2",
+                    "title": "Don't Go To Extremes: Revealing the Excessive Sensitivity and Calibration Limitations of LLMs in Implicit Hate Speech Detection",
+                    "abstract": "The fairness and trustworthiness of Large Language Models (LLMs) are\nreceiving increasing attention. Implicit hate speech, which employs indirect\nlanguage to convey hateful intentions, occupies a significant portion of\npractice. However, the extent to which LLMs effectively address this issue\nremains insufficiently examined. This paper delves into the capability of LLMs\nto detect implicit hate speech (Classification Task) and express confidence in\ntheir responses (Calibration Task). Our evaluation meticulously considers\nvarious prompt patterns and mainstream uncertainty estimation methods. Our\nfindings highlight that LLMs exhibit two extremes: (1) LLMs display excessive\nsensitivity towards groups or topics that may cause fairness issues, resulting\nin misclassifying benign statements as hate speech. (2) LLMs' confidence scores\nfor each method excessively concentrate on a fixed range, remaining unchanged\nregardless of the dataset's complexity. Consequently, the calibration\nperformance is heavily reliant on primary classification accuracy. These\ndiscoveries unveil new limitations of LLMs, underscoring the need for caution\nwhen optimizing models to ensure they do not veer towards extremes. This serves\nas a reminder to carefully consider sensitivity and confidence in the pursuit\nof model fairness.",
+                    "authors": "Min Zhang, Jianfeng He, Taoran Ji, Chang-Tien Lu",
+                    "published": "2024-02-18",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.09397v1",
+                    "title": "Do Large GPT Models Discover Moral Dimensions in Language Representations? A Topological Study Of Sentence Embeddings",
+                    "abstract": "As Large Language Models are deployed within Artificial Intelligence systems,\nthat are increasingly integrated with human society, it becomes more important\nthan ever to study their internal structures. Higher level abilities of LLMs\nsuch as GPT-3.5 emerge in large part due to informative language\nrepresentations they induce from raw text data during pre-training on trillions\nof words. These embeddings exist in vector spaces of several thousand\ndimensions, and their processing involves mapping between multiple vector\nspaces, with total number of parameters on the order of trillions. Furthermore,\nthese language representations are induced by gradient optimization, resulting\nin a black box system that is hard to interpret. In this paper, we take a look\nat the topological structure of neuronal activity in the \"brain\" of Chat-GPT's\nfoundation language model, and analyze it with respect to a metric representing\nthe notion of fairness. We develop a novel approach to visualize GPT's moral\ndimensions. We first compute a fairness metric, inspired by social psychology\nliterature, to identify factors that typically influence fairness assessments\nin humans, such as legitimacy, need, and responsibility. Subsequently, we\nsummarize the manifold's shape using a lower-dimensional simplicial complex,\nwhose topology is derived from this metric. We color it with a heat map\nassociated with this fairness metric, producing human-readable visualizations\nof the high-dimensional sentence manifold. Our results show that sentence\nembeddings based on GPT-3.5 can be decomposed into two submanifolds\ncorresponding to fair and unfair moral judgments. This indicates that GPT-based\nlanguage models develop a moral dimension within their representation spaces\nand induce an understanding of fairness during their training process.",
+                    "authors": "Stephen Fitz",
+                    "published": "2023-09-17",
+                    "updated": "2023-09-17",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "cs.NE"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15451v1",
+                    "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+                    "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+                    "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+                    "published": "2024-03-18",
+                    "updated": "2024-03-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.05668v1",
+                    "title": "CFaiRLLM: Consumer Fairness Evaluation in Large-Language Model Recommender System",
+                    "abstract": "In the evolving landscape of recommender systems, the integration of Large\nLanguage Models (LLMs) such as ChatGPT marks a new era, introducing the concept\nof Recommendation via LLM (RecLLM). While these advancements promise\nunprecedented personalization and efficiency, they also bring to the fore\ncritical concerns regarding fairness, particularly in how recommendations might\ninadvertently perpetuate or amplify biases associated with sensitive user\nattributes. In order to address these concerns, our study introduces a\ncomprehensive evaluation framework, CFaiRLLM, aimed at evaluating (and thereby\nmitigating) biases on the consumer side within RecLLMs.\n  Our research methodically assesses the fairness of RecLLMs by examining how\nrecommendations might vary with the inclusion of sensitive attributes such as\ngender, age, and their intersections, through both similarity alignment and\ntrue preference alignment. By analyzing recommendations generated under\ndifferent conditions-including the use of sensitive attributes in user\nprompts-our framework identifies potential biases in the recommendations\nprovided. A key part of our study involves exploring how different detailed\nstrategies for constructing user profiles (random, top-rated, recent) impact\nthe alignment between recommendations made without consideration of sensitive\nattributes and those that are sensitive-attribute-aware, highlighting the bias\nmechanisms within RecLLMs.\n  The findings in our study highlight notable disparities in the fairness of\nrecommendations, particularly when sensitive attributes are integrated into the\nrecommendation process, either individually or in combination. The analysis\ndemonstrates that the choice of user profile sampling strategy plays a\nsignificant role in affecting fairness outcomes, highlighting the complexity of\nachieving fair recommendations in the era of LLMs.",
+                    "authors": "Yashar Deldjoo, Tommaso di Noia",
+                    "published": "2024-03-08",
+                    "updated": "2024-03-08",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14804v1",
+                    "title": "Use large language models to promote equity",
+                    "abstract": "Advances in large language models (LLMs) have driven an explosion of interest\nabout their societal impacts. Much of the discourse around how they will impact\nsocial equity has been cautionary or negative, focusing on questions like \"how\nmight LLMs be biased and how would we mitigate those biases?\" This is a vital\ndiscussion: the ways in which AI generally, and LLMs specifically, can entrench\nbiases have been well-documented. But equally vital, and much less discussed,\nis the more opportunity-focused counterpoint: \"what promising applications do\nLLMs enable that could promote equity?\" If LLMs are to enable a more equitable\nworld, it is not enough just to play defense against their biases and failure\nmodes. We must also go on offense, applying them positively to equity-enhancing\nuse cases to increase opportunities for underserved groups and reduce societal\ndiscrimination. There are many choices which determine the impact of AI, and a\nfundamental choice very early in the pipeline is the problems we choose to\napply it to. If we focus only later in the pipeline -- making LLMs marginally\nmore fair as they facilitate use cases which intrinsically entrench power -- we\nwill miss an important opportunity to guide them to equitable impacts. Here, we\nhighlight the emerging potential of LLMs to promote equity by presenting four\nnewly possible, promising research directions, while keeping risks and\ncautionary points in clear view.",
+                    "authors": "Emma Pierson, Divya Shanmugam, Rajiv Movva, Jon Kleinberg, Monica Agrawal, Mark Dredze, Kadija Ferryman, Judy Wawira Gichoya, Dan Jurafsky, Pang Wei Koh, Karen Levy, Sendhil Mullainathan, Ziad Obermeyer, Harini Suresh, Keyon Vafa",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-22",
+                    "primary_cat": "cs.CY",
+                    "cats": [
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.14607v2",
+                    "title": "Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications",
+                    "abstract": "Recent literature has suggested the potential of using large language models\n(LLMs) to make classifications for tabular tasks. However, LLMs have been shown\nto exhibit harmful social biases that reflect the stereotypes and inequalities\npresent in society. To this end, as well as the widespread use of tabular data\nin many high-stake applications, it is important to explore the following\nquestions: what sources of information do LLMs draw upon when making\nclassifications for tabular tasks; whether and to what extent are LLM\nclassifications for tabular data influenced by social biases and stereotypes;\nand what are the consequential implications for fairness?\n  Through a series of experiments, we delve into these questions and show that\nLLMs tend to inherit social biases from their training data which significantly\nimpact their fairness in tabular classification tasks. Furthermore, our\ninvestigations show that in the context of bias mitigation, though in-context\nlearning and finetuning have a moderate effect, the fairness metric gap between\ndifferent subgroups is still larger than that in traditional machine learning\nmodels, such as Random Forest and shallow Neural Networks. This observation\nemphasizes that the social biases are inherent within the LLMs themselves and\ninherited from their pretraining corpus, not only from the downstream task\ndatasets. Besides, we demonstrate that label-flipping of in-context examples\ncan significantly reduce biases, further highlighting the presence of inherent\nbias within LLMs.",
+                    "authors": "Yanchen Liu, Srishti Gautam, Jiaqi Ma, Himabindu Lakkaraju",
+                    "published": "2023-10-23",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.18569v1",
+                    "title": "Fairness of ChatGPT",
+                    "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+                    "authors": "Yunqi Li, Yongfeng Zhang",
+                    "published": "2023-05-22",
+                    "updated": "2023-05-22",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.14769v3",
+                    "title": "Large Language Model (LLM) Bias Index -- LLMBI",
+                    "abstract": "The Large Language Model Bias Index (LLMBI) is a pioneering approach designed\nto quantify and address biases inherent in large language models (LLMs), such\nas GPT-4. We recognise the increasing prevalence and impact of LLMs across\ndiverse sectors. This research introduces a novel metric, LLMBI, to\nsystematically measure and mitigate biases potentially skewing model responses.\nWe formulated LLMBI using a composite scoring system incorporating multiple\ndimensions of bias, including but not limited to age, gender, and racial\nbiases. To operationalise this metric, we engaged in a multi-step process\ninvolving collecting and annotating LLM responses, applying sophisticated\nNatural Language Processing (NLP) techniques for bias detection, and computing\nthe LLMBI score through a specially crafted mathematical formula. The formula\nintegrates weighted averages of various bias dimensions, a penalty for dataset\ndiversity deficiencies, and a correction for sentiment biases. Our empirical\nanalysis, conducted using responses from OpenAI's API, employs advanced\nsentiment analysis as a representative method for bias detection. The research\nreveals LLMs, whilst demonstrating impressive capabilities in text generation,\nexhibit varying degrees of bias across different dimensions. LLMBI provides a\nquantifiable measure to compare biases across models and over time, offering a\nvital tool for systems engineers, researchers and regulators in enhancing the\nfairness and reliability of LLMs. It highlights the potential of LLMs in\nmimicking unbiased human-like responses. Additionally, it underscores the\nnecessity of continuously monitoring and recalibrating such models to align\nwith evolving societal norms and ethical standards.",
+                    "authors": "Abiodun Finbarrs Oketunji, Muhammad Anas, Deepthi Saina",
+                    "published": "2023-12-22",
+                    "updated": "2023-12-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.15491v1",
+                    "title": "Open Source Conversational LLMs do not know most Spanish words",
+                    "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+                    "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+                    "published": "2024-03-21",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.00306v1",
+                    "title": "Probing Explicit and Implicit Gender Bias through LLM Conditional Text Generation",
+                    "abstract": "Large Language Models (LLMs) can generate biased and toxic responses. Yet\nmost prior work on LLM gender bias evaluation requires predefined\ngender-related phrases or gender stereotypes, which are challenging to be\ncomprehensively collected and are limited to explicit bias evaluation. In\naddition, we believe that instances devoid of gender-related language or\nexplicit stereotypes in inputs can still induce gender bias in LLMs. Thus, in\nthis work, we propose a conditional text generation mechanism without the need\nfor predefined gender phrases and stereotypes. This approach employs three\ntypes of inputs generated through three distinct strategies to probe LLMs,\naiming to show evidence of explicit and implicit gender biases in LLMs. We also\nutilize explicit and implicit evaluation metrics to evaluate gender bias in\nLLMs under different strategies. Our experiments demonstrate that an increased\nmodel size does not consistently lead to enhanced fairness and all tested LLMs\nexhibit explicit and/or implicit gender bias, even when explicit gender\nstereotypes are absent in the inputs.",
+                    "authors": "Xiangjue Dong, Yibo Wang, Philip S. Yu, James Caverlee",
+                    "published": "2023-11-01",
+                    "updated": "2023-11-01",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.06852v2",
+                    "title": "ChemLLM: A Chemical Large Language Model",
+                    "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+                    "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+                    "published": "2024-02-10",
+                    "updated": "2024-04-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.04892v2",
+                    "title": "Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs",
+                    "abstract": "Recent works have showcased the ability of LLMs to embody diverse personas in\ntheir responses, exemplified by prompts like 'You are Yoda. Explain the Theory\nof Relativity.' While this ability allows personalization of LLMs and enables\nhuman behavior simulation, its effect on LLMs' capabilities remains unclear. To\nfill this gap, we present the first extensive study of the unintended\nside-effects of persona assignment on the ability of LLMs to perform basic\nreasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse\npersonas (e.g. an Asian person) spanning 5 socio-demographic groups. Our\nexperiments unveil that LLMs harbor deep rooted bias against various\nsocio-demographics underneath a veneer of fairness. While they overtly reject\nstereotypes when explicitly asked ('Are Black people less skilled at\nmathematics?'), they manifest stereotypical and erroneous presumptions when\nasked to answer questions while adopting a persona. These can be observed as\nabstentions in responses, e.g., 'As a Black person, I can't answer this\nquestion as it requires math knowledge', and generally result in a substantial\nperformance drop. Our experiments with ChatGPT-3.5 show that this bias is\nubiquitous - 80% of our personas demonstrate bias; it is significant - some\ndatasets show performance drops of 70%+; and can be especially harmful for\ncertain groups - some personas suffer statistically significant drops on 80%+\nof the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with\nGPT-4-Turbo showing the least but still a problematic amount of bias (evident\nin 42% of the personas). Further analysis shows that these persona-induced\nerrors can be hard-to-discern and hard-to-avoid. Our findings serve as a\ncautionary tale that the practice of assigning personas to LLMs - a trend on\nthe rise - can surface their deep-rooted biases and have unforeseeable and\ndetrimental side-effects.",
+                    "authors": "Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot",
+                    "published": "2023-11-08",
+                    "updated": "2024-01-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.17916v2",
+                    "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+                    "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+                    "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+                    "published": "2024-02-27",
+                    "updated": "2024-03-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18333v3",
+                    "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+                    "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+                    "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+                    "published": "2023-10-20",
+                    "updated": "2023-12-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.04489v1",
+                    "title": "De-amplifying Bias from Differential Privacy in Language Model Fine-tuning",
+                    "abstract": "Fairness and privacy are two important values machine learning (ML)\npractitioners often seek to operationalize in models. Fairness aims to reduce\nmodel bias for social/demographic sub-groups. Privacy via differential privacy\n(DP) mechanisms, on the other hand, limits the impact of any individual's\ntraining data on the resulting model. The trade-offs between privacy and\nfairness goals of trustworthy ML pose a challenge to those wishing to address\nboth. We show that DP amplifies gender, racial, and religious bias when\nfine-tuning large language models (LLMs), producing models more biased than\nones fine-tuned without DP. We find the cause of the amplification to be a\ndisparity in convergence of gradients across sub-groups. Through the case of\nbinary gender bias, we demonstrate that Counterfactual Data Augmentation (CDA),\na known method for addressing bias, also mitigates bias amplification by DP. As\na consequence, DP and CDA together can be used to fine-tune models while\nmaintaining both fairness and privacy.",
+                    "authors": "Sanjari Srivastava, Piotr Mardziel, Zhikhun Zhang, Archana Ahlawat, Anupam Datta, John C Mitchell",
+                    "published": "2024-02-07",
+                    "updated": "2024-02-07",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CR",
+                        "cs.CY",
+                        "stat.ME"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.12090v1",
+                    "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+                    "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+                    "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+                    "published": "2023-05-20",
+                    "updated": "2023-05-20",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.14345v2",
+                    "title": "Bias Testing and Mitigation in LLM-based Code Generation",
+                    "abstract": "Utilizing state-of-the-art Large Language Models (LLMs), automatic code\ngeneration models play a pivotal role in enhancing the productivity of software\ndevelopment procedures. As the adoption of LLMs becomes more widespread in\nsoftware coding ecosystems, a pressing issue has emerged: does the generated\ncode contain social bias and unfairness, such as those related to age, gender,\nand race? This issue concerns the integrity, fairness, and ethical foundation\nof software applications that depend on the code generated by these models, yet\nis under-explored in the literature. This paper presents a novel bias testing\nframework that is specifically designed for code generation tasks. Based on\nthis framework, we conduct an extensive evaluation of the bias in code\ngenerated by five state-of-the-art LLMs. Our findings reveal that 20.29% to\n44.93% code functions generated by the models under study are biased when\nhandling bias sensitive tasks (i.e., tasks that involve sensitive attributes\nsuch as age and gender). This indicates that the existing LLMs can be unfair in\ncode generation, posing risks of unintended and harmful software behaviors. To\nmitigate bias for code generation models, we evaluate five bias mitigation\nprompt strategies, i.e., utilizing bias testing results to refine the code\n(zero-shot), one-, few-shot, and two Chain-of-Thought (CoT) prompts. Our\nevaluation results illustrate that these strategies are all effective in\nmitigating bias. Overall, one-shot and few-shot learning are the two most\neffective. For GPT-4, 80% to 90% code bias can be removed with one-shot\nlearning.",
+                    "authors": "Dong Huang, Qingwen Bu, Jie Zhang, Xiaofei Xie, Junjie Chen, Heming Cui",
+                    "published": "2023-09-03",
+                    "updated": "2024-01-09",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.00588v1",
+                    "title": "Fairness in Serving Large Language Models",
+                    "abstract": "High-demand LLM inference services (e.g., ChatGPT and BARD) support a wide\nrange of requests from short chat conversations to long document reading. To\nensure that all client requests are processed fairly, most major LLM inference\nservices have request rate limits, to ensure that no client can dominate the\nrequest queue. However, this rudimentary notion of fairness also results in\nunder-utilization of the resources and poor client experience when there is\nspare capacity. While there is a rich literature on fair scheduling, serving\nLLMs presents new challenges due to their unpredictable request lengths and\ntheir unique batching characteristics on parallel accelerators. This paper\nintroduces the definition of LLM serving fairness based on a cost function that\naccounts for the number of input and output tokens processed. To achieve\nfairness in serving, we propose a novel scheduling algorithm, the Virtual Token\nCounter (VTC), a fair scheduler based on the continuous batching mechanism. We\nprove a 2x tight upper bound on the service difference between two backlogged\nclients, adhering to the requirement of work-conserving. Through extensive\nexperiments, we demonstrate the superior performance of VTC in ensuring\nfairness, especially in contrast to other baseline methods, which exhibit\nshortcomings under various conditions.",
+                    "authors": "Ying Sheng, Shiyi Cao, Dacheng Li, Banghua Zhu, Zhuohan Li, Danyang Zhuo, Joseph E. Gonzalez, Ion Stoica",
+                    "published": "2023-12-31",
+                    "updated": "2023-12-31",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG",
+                        "cs.PF"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.06500v1",
+                    "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+                    "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+                    "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+                    "published": "2023-10-10",
+                    "updated": "2023-10-10",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.15007v1",
+                    "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+                    "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+                    "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+                    "published": "2023-10-23",
+                    "updated": "2023-10-23",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CR",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.08780v1",
+                    "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+                    "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+                    "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+                    "published": "2023-10-13",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.09606v1",
+                    "title": "Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey",
+                    "abstract": "Causal inference has shown potential in enhancing the predictive accuracy,\nfairness, robustness, and explainability of Natural Language Processing (NLP)\nmodels by capturing causal relationships among variables. The emergence of\ngenerative Large Language Models (LLMs) has significantly impacted various NLP\ndomains, particularly through their advanced reasoning capabilities. This\nsurvey focuses on evaluating and improving LLMs from a causal view in the\nfollowing areas: understanding and improving the LLMs' reasoning capacity,\naddressing fairness and safety issues in LLMs, complementing LLMs with\nexplanations, and handling multimodality. Meanwhile, LLMs' strong reasoning\ncapacities can in turn contribute to the field of causal inference by aiding\ncausal relationship discovery and causal effect estimations. This review\nexplores the interplay between causal inference frameworks and LLMs from both\nperspectives, emphasizing their collective potential to further the development\nof more advanced and equitable artificial intelligence systems.",
+                    "authors": "Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang",
+                    "published": "2024-03-14",
+                    "updated": "2024-03-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.19118v1",
+                    "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate",
+                    "abstract": "Modern large language models (LLMs) like ChatGPT have shown remarkable\nperformance on general language tasks but still struggle on complex reasoning\ntasks, which drives the research on cognitive behaviors of LLMs to explore\nhuman-like problem-solving strategies. Along this direction, one representative\nstrategy is self-reflection, which asks an LLM to refine the solution with the\nfeedback generated by itself iteratively. However, our study shows that such\nreflection-style methods suffer from the Degeneration-of-Thought (DoT) problem:\nonce the LLM has established confidence in its solutions, it is unable to\ngenerate novel thoughts later through reflection even if its initial stance is\nincorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD)\nframework, in which multiple agents express their arguments in the state of\n\"tit for tat\" and a judge manages the debate process to obtain a final\nsolution. Clearly, our MAD framework encourages divergent thinking in LLMs\nwhich would be helpful for tasks that require deep levels of contemplation.\nExperiment results on two challenging datasets, commonsense machine translation\nand counter-intuitive arithmetic reasoning, demonstrate the effectiveness of\nour MAD framework. Extensive analyses suggest that the adaptive break of debate\nand the modest level of \"tit for tat\" state are required for MAD to obtain good\nperformance. Moreover, we find that LLMs might not be a fair judge if different\nLLMs are used for agents. Codes:\nhttps://github.com/Skytliang/Multi-Agents-Debate",
+                    "authors": "Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi",
+                    "published": "2023-05-30",
+                    "updated": "2023-05-30",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.01349v1",
+                    "title": "Fairness in Large Language Models: A Taxonomic Survey",
+                    "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+                    "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+                    "published": "2024-03-31",
+                    "updated": "2024-03-31",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10149v2",
+                    "title": "A Survey on Fairness in Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have shown powerful performance and development\nprospects and are widely deployed in the real world. However, LLMs can capture\nsocial biases from unprocessed training data and propagate the biases to\ndownstream tasks. Unfair LLM systems have undesirable social impacts and\npotential harms. In this paper, we provide a comprehensive review of related\nresearch on fairness in LLMs. Considering the influence of parameter magnitude\nand training paradigm on research strategy, we divide existing fairness\nresearch into oriented to medium-sized LLMs under pre-training and fine-tuning\nparadigms and oriented to large-sized LLMs under prompting paradigms. First,\nfor medium-sized LLMs, we introduce evaluation metrics and debiasing methods\nfrom the perspectives of intrinsic bias and extrinsic bias, respectively. Then,\nfor large-sized LLMs, we introduce recent fairness research, including fairness\nevaluation, reasons for bias, and debiasing methods. Finally, we discuss and\nprovide insight on the challenges and future directions for the development of\nfairness in LLMs.",
+                    "authors": "Yingji Li, Mengnan Du, Rui Song, Xin Wang, Ying Wang",
+                    "published": "2023-08-20",
+                    "updated": "2024-02-21",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.13095v1",
+                    "title": "Enhancing Logical Reasoning in Large Language Models to Facilitate Legal Applications",
+                    "abstract": "Language serves as a vehicle for conveying thought, enabling communication\namong individuals. The ability to distinguish between diverse concepts,\nidentify fairness and injustice, and comprehend a range of legal notions\nfundamentally relies on logical reasoning. Large Language Models (LLMs) attempt\nto emulate human language understanding and generation, but their competency in\nlogical reasoning remains limited. This paper seeks to address the\nphilosophical question: How can we effectively teach logical reasoning to LLMs\nwhile maintaining a deep understanding of the intricate relationship between\nlanguage and logic? By focusing on bolstering LLMs' capabilities in logical\nreasoning, we aim to expand their applicability in law and other\nlogic-intensive disciplines. To this end, we propose a Reinforcement Learning\nfrom Logical Feedback (RLLF) approach, which serves as a potential framework\nfor refining LLMs' reasoning capacities. Through RLLF and a revised evaluation\nmethodology, we explore new avenues for research in this domain and contribute\nto the development of LLMs capable of handling complex legal reasoning tasks\nwhile acknowledging the fundamental connection between language and logic.",
+                    "authors": "Ha-Thanh Nguyen, Wachara Fungwacharakorn, Ken Satoh",
+                    "published": "2023-11-22",
+                    "updated": "2023-11-22",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05374v2",
+                    "title": "Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment",
+                    "abstract": "Ensuring alignment, which refers to making models behave in accordance with\nhuman intentions [1,2], has become a critical task before deploying large\nlanguage models (LLMs) in real-world applications. For instance, OpenAI devoted\nsix months to iteratively aligning GPT-4 before its release [3]. However, a\nmajor challenge faced by practitioners is the lack of clear guidance on\nevaluating whether LLM outputs align with social norms, values, and\nregulations. This obstacle hinders systematic iteration and deployment of LLMs.\nTo address this issue, this paper presents a comprehensive survey of key\ndimensions that are crucial to consider when assessing LLM trustworthiness. The\nsurvey covers seven major categories of LLM trustworthiness: reliability,\nsafety, fairness, resistance to misuse, explainability and reasoning, adherence\nto social norms, and robustness. Each major category is further divided into\nseveral sub-categories, resulting in a total of 29 sub-categories.\nAdditionally, a subset of 8 sub-categories is selected for further\ninvestigation, where corresponding measurement studies are designed and\nconducted on several widely-used LLMs. The measurement results indicate that,\nin general, more aligned models tend to perform better in terms of overall\ntrustworthiness. However, the effectiveness of alignment varies across the\ndifferent trustworthiness categories considered. This highlights the importance\nof conducting more fine-grained analyses, testing, and making continuous\nimprovements on LLM alignment. By shedding light on these key dimensions of LLM\ntrustworthiness, this paper aims to provide valuable insights and guidance to\npractitioners in the field. Understanding and addressing these concerns will be\ncrucial in achieving reliable and ethically sound deployment of LLMs in various\napplications.",
+                    "authors": "Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, Hang Li",
+                    "published": "2023-08-10",
+                    "updated": "2024-03-21",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.19465v1",
+                    "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+                    "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+                    "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+                    "published": "2024-02-29",
+                    "updated": "2024-02-29",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.18502v1",
+                    "title": "Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification",
+                    "abstract": "Employing Large Language Models (LLM) in various downstream applications such\nas classification is crucial, especially for smaller companies lacking the\nexpertise and resources required for fine-tuning a model. Fairness in LLMs\nhelps ensure inclusivity, equal representation based on factors such as race,\ngender and promotes responsible AI deployment. As the use of LLMs has become\nincreasingly prevalent, it is essential to assess whether LLMs can generate\nfair outcomes when subjected to considerations of fairness. In this study, we\nintroduce a framework outlining fairness regulations aligned with various\nfairness definitions, with each definition being modulated by varying degrees\nof abstraction. We explore the configuration for in-context learning and the\nprocedure for selecting in-context demonstrations using RAG, while\nincorporating fairness rules into the process. Experiments conducted with\ndifferent LLMs indicate that GPT-4 delivers superior results in terms of both\naccuracy and fairness compared to other models. This work is one of the early\nattempts to achieve fairness in prediction tasks by utilizing LLMs through\nin-context learning.",
+                    "authors": "Garima Chhikara, Anurag Sharma, Kripabandhu Ghosh, Abhijnan Chakraborty",
+                    "published": "2024-02-28",
+                    "updated": "2024-02-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02294v1",
+                    "title": "LLMs grasp morality in concept",
+                    "abstract": "Work in AI ethics and fairness has made much progress in regulating LLMs to\nreflect certain values, such as fairness, truth, and diversity. However, it has\ntaken the problem of how LLMs might 'mean' anything at all for granted. Without\naddressing this, it is not clear what imbuing LLMs with such values even means.\nIn response, we provide a general theory of meaning that extends beyond humans.\nWe use this theory to explicate the precise nature of LLMs as meaning-agents.\nWe suggest that the LLM, by virtue of its position as a meaning-agent, already\ngrasps the constructions of human society (e.g. morality, gender, and race) in\nconcept. Consequently, under certain ethical frameworks, currently popular\nmethods for model alignment are limited at best and counterproductive at worst.\nMoreover, unaligned models may help us better develop our moral and social\nphilosophy.",
+                    "authors": "Mark Pock, Andre Ye, Jared Moore",
+                    "published": "2023-11-04",
+                    "updated": "2023-11-04",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.15585v1",
+                    "title": "Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting",
+                    "abstract": "There exist both scalable tasks, like reading comprehension and\nfact-checking, where model performance improves with model size, and unscalable\ntasks, like arithmetic reasoning and symbolic reasoning, where model\nperformance does not necessarily improve with model size. Large language models\n(LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate\nincremental predictions even on unscalable tasks. Unfortunately, despite their\nexceptional reasoning abilities, LLMs tend to internalize and reproduce\ndiscriminatory societal biases. Whether CoT can provide discriminatory or\negalitarian rationalizations for the implicit information in unscalable tasks\nremains an open question.\n  In this study, we examine the impact of LLMs' step-by-step predictions on\ngender bias in unscalable tasks. For this purpose, we construct a benchmark for\nan unscalable task where the LLM is given a list of words comprising feminine,\nmasculine, and gendered occupational words, and is required to count the number\nof feminine and masculine words. In our CoT prompts, we require the LLM to\nexplicitly indicate whether each word in the word list is a feminine or\nmasculine before making the final predictions. With counting and handling the\nmeaning of words, this benchmark has characteristics of both arithmetic\nreasoning and symbolic reasoning. Experimental results in English show that\nwithout step-by-step prediction, most LLMs make socially biased predictions,\ndespite the task being as simple as counting words. Interestingly, CoT\nprompting reduces this unconscious social bias in LLMs and encourages fair\npredictions.",
+                    "authors": "Masahiro Kaneko, Danushka Bollegala, Naoaki Okazaki, Timothy Baldwin",
+                    "published": "2024-01-28",
+                    "updated": "2024-01-28",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.05694v1",
+                    "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+                    "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+                    "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+                    "published": "2023-10-09",
+                    "updated": "2023-10-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.10199v3",
+                    "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+                    "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+                    "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+                    "published": "2024-04-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.08495v2",
+                    "title": "Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humans",
+                    "abstract": "Large language models (LLMs) are becoming pervasive in everyday life, yet\ntheir propensity to reproduce biases inherited from training data remains a\npressing concern. Prior investigations into bias in LLMs have focused on the\nassociation of social groups with stereotypical attributes. However, this is\nonly one form of human bias such systems may reproduce. We investigate a new\nform of bias in LLMs that resembles a social psychological phenomenon where\nsocially subordinate groups are perceived as more homogeneous than socially\ndominant groups. We had ChatGPT, a state-of-the-art LLM, generate texts about\nintersectional group identities and compared those texts on measures of\nhomogeneity. We consistently found that ChatGPT portrayed African, Asian, and\nHispanic Americans as more homogeneous than White Americans, indicating that\nthe model described racial minority groups with a narrower range of human\nexperience. ChatGPT also portrayed women as more homogeneous than men, but\nthese differences were small. Finally, we found that the effect of gender\ndiffered across racial/ethnic groups such that the effect of gender was\nconsistent within African and Hispanic Americans but not within Asian and White\nAmericans. We argue that the tendency of LLMs to describe groups as less\ndiverse risks perpetuating stereotypes and discriminatory behavior.",
+                    "authors": "Messi H. J. Lee, Jacob M. Montgomery, Calvin K. Lai",
+                    "published": "2024-01-16",
+                    "updated": "2024-04-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2307.11761v1",
+                    "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+                    "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+                    "authors": "Yashar Deldjoo",
+                    "published": "2023-07-14",
+                    "updated": "2023-07-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.06003v1",
+                    "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+                    "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+                    "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+                    "published": "2024-04-09",
+                    "updated": "2024-04-09",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.18130v2",
+                    "title": "DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues",
+                    "abstract": "Controversy is a reflection of our zeitgeist, and an important aspect to any\ndiscourse. The rise of large language models (LLMs) as conversational systems\nhas increased public reliance on these systems for answers to their various\nquestions. Consequently, it is crucial to systematically examine how these\nmodels respond to questions that pertaining to ongoing debates. However, few\nsuch datasets exist in providing human-annotated labels reflecting the\ncontemporary discussions. To foster research in this area, we propose a novel\nconstruction of a controversial questions dataset, expanding upon the publicly\nreleased Quora Question Pairs Dataset. This dataset presents challenges\nconcerning knowledge recency, safety, fairness, and bias. We evaluate different\nLLMs using a subset of this dataset, illuminating how they handle controversial\nissues and the stances they adopt. This research ultimately contributes to our\nunderstanding of LLMs' interaction with controversial issues, paving the way\nfor improvements in their comprehension and handling of complex societal\ndebates.",
+                    "authors": "David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein, Jason D. Williams",
+                    "published": "2023-10-27",
+                    "updated": "2023-11-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.07981v1",
+                    "title": "Manipulating Large Language Models to Increase Product Visibility",
+                    "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+                    "authors": "Aounon Kumar, Himabindu Lakkaraju",
+                    "published": "2024-04-11",
+                    "updated": "2024-04-11",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.06056v1",
+                    "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+                    "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+                    "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+                    "published": "2023-12-11",
+                    "updated": "2023-12-11",
+                    "primary_cat": "cs.SE",
+                    "cats": [
+                        "cs.SE",
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15478v1",
+                    "title": "A Group Fairness Lens for Large Language Models",
+                    "abstract": "The rapid advancement of large language models has revolutionized various\napplications but also raised crucial concerns about their potential to\nperpetuate biases and unfairness when deployed in social media contexts.\nEvaluating LLMs' potential biases and fairness has become crucial, as existing\nmethods rely on limited prompts focusing on just a few groups, lacking a\ncomprehensive categorical perspective. In this paper, we propose evaluating LLM\nbiases from a group fairness lens using a novel hierarchical schema\ncharacterizing diverse social groups. Specifically, we construct a dataset,\nGFair, encapsulating target-attribute combinations across multiple dimensions.\nIn addition, we introduce statement organization, a new open-ended text\ngeneration task, to uncover complex biases in LLMs. Extensive evaluations of\npopular LLMs reveal inherent safety concerns. To mitigate the biases of LLM\nfrom a group fairness perspective, we pioneer a novel chain-of-thought method\nGF-Think to mitigate biases of LLMs from a group fairness perspective.\nExperimental results demonstrate its efficacy in mitigating bias in LLMs to\nachieve fairness.",
+                    "authors": "Guanqun Bi, Lei Shen, Yuqiang Xie, Yanan Cao, Tiangang Zhu, Xiaodong He",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2303.01248v3",
+                    "title": "Can ChatGPT Assess Human Personalities? A General Evaluation Framework",
+                    "abstract": "Large Language Models (LLMs) especially ChatGPT have produced impressive\nresults in various areas, but their potential human-like psychology is still\nlargely unexplored. Existing works study the virtual personalities of LLMs but\nrarely explore the possibility of analyzing human personalities via LLMs. This\npaper presents a generic evaluation framework for LLMs to assess human\npersonalities based on Myers Briggs Type Indicator (MBTI) tests. Specifically,\nwe first devise unbiased prompts by randomly permuting options in MBTI\nquestions and adopt the average testing result to encourage more impartial\nanswer generation. Then, we propose to replace the subject in question\nstatements to enable flexible queries and assessments on different subjects\nfrom LLMs. Finally, we re-formulate the question instructions in a manner of\ncorrectness evaluation to facilitate LLMs to generate clearer responses. The\nproposed framework enables LLMs to flexibly assess personalities of different\ngroups of people. We further propose three evaluation metrics to measure the\nconsistency, robustness, and fairness of assessment results from\nstate-of-the-art LLMs including ChatGPT and GPT-4. Our experiments reveal\nChatGPT's ability to assess human personalities, and the average results\ndemonstrate that it can achieve more consistent and fairer assessments in spite\nof lower robustness against prompt biases compared with InstructGPT.",
+                    "authors": "Haocong Rao, Cyril Leung, Chunyan Miao",
+                    "published": "2023-03-01",
+                    "updated": "2023-10-13",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.08472v1",
+                    "title": "Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models",
+                    "abstract": "Recently, work in NLP has shifted to few-shot (in-context) learning, with\nlarge language models (LLMs) performing well across a range of tasks. However,\nwhile fairness evaluations have become a standard for supervised methods,\nlittle is known about the fairness of LLMs as prediction systems. Further,\ncommon standard methods for fairness involve access to models weights or are\napplied during finetuning, which are not applicable in few-shot learning. Do\nLLMs exhibit prediction biases when used for standard NLP tasks? In this work,\nwe explore the effect of shots, which directly affect the performance of\nmodels, on the fairness of LLMs as NLP classification systems. We consider how\ndifferent shot selection strategies, both existing and new demographically\nsensitive methods, affect model fairness across three standard fairness\ndatasets. We discuss how future work can include LLM fairness evaluations.",
+                    "authors": "Carlos Aguirre, Kuleen Sasse, Isabel Cachola, Mark Dredze",
+                    "published": "2023-11-14",
+                    "updated": "2023-11-14",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00811v1",
+                    "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+                    "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+                    "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+                    "published": "2024-02-25",
+                    "updated": "2024-02-25",
+                    "primary_cat": "cs.AI",
+                    "cats": [
+                        "cs.AI",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.13840v1",
+                    "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+                    "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+                    "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+                    "published": "2024-03-15",
+                    "updated": "2024-03-15",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.SI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.11595v3",
+                    "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+                    "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+                    "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+                    "published": "2023-05-19",
+                    "updated": "2023-10-18",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.10397v2",
+                    "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+                    "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+                    "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+                    "published": "2023-08-21",
+                    "updated": "2023-10-27",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.11764v1",
+                    "title": "ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs",
+                    "abstract": "Large Language models (LLMs), while powerful, exhibit harmful social biases.\nDebiasing is often challenging due to computational costs, data constraints,\nand potential degradation of multi-task language capabilities. This work\nintroduces a novel approach utilizing ChatGPT to generate synthetic training\ndata, aiming to enhance the debiasing of LLMs. We propose two strategies:\nTargeted Prompting, which provides effective debiasing for known biases but\nnecessitates prior specification of bias in question; and General Prompting,\nwhich, while slightly less effective, offers debiasing across various\ncategories. We leverage resource-efficient LLM debiasing using adapter tuning\nand compare the effectiveness of our synthetic data to existing debiasing\ndatasets. Our results reveal that: (1) ChatGPT can efficiently produce\nhigh-quality training data for debiasing other LLMs; (2) data produced via our\napproach surpasses existing datasets in debiasing performance while also\npreserving internal knowledge of a pre-trained LLM; and (3) synthetic data\nexhibits generalizability across categories, effectively mitigating various\nbiases, including intersectional ones. These findings underscore the potential\nof synthetic data in advancing the fairness of LLMs with minimal retraining\ncost.",
+                    "authors": "Pengrui Han, Rafal Kocielnik, Adhithya Saravanan, Roy Jiang, Or Sharir, Anima Anandkumar",
+                    "published": "2024-02-19",
+                    "updated": "2024-02-19",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "68T50",
+                        "I.2.7; K.4.1"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.07609v3",
+                    "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+                    "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+                    "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+                    "published": "2023-05-12",
+                    "updated": "2023-10-17",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.07420v1",
+                    "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+                    "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+                    "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+                    "published": "2023-12-12",
+                    "updated": "2023-12-12",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.03033v1",
+                    "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+                    "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+                    "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+                    "published": "2023-11-06",
+                    "updated": "2023-11-06",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2308.05345v3",
+                    "title": "RTLLM: An Open-Source Benchmark for Design RTL Generation with Large Language Model",
+                    "abstract": "Inspired by the recent success of large language models (LLMs) like ChatGPT,\nresearchers start to explore the adoption of LLMs for agile hardware design,\nsuch as generating design RTL based on natural-language instructions. However,\nin existing works, their target designs are all relatively simple and in a\nsmall scale, and proposed by the authors themselves, making a fair comparison\namong different LLM solutions challenging. In addition, many prior works only\nfocus on the design correctness, without evaluating the design qualities of\ngenerated design RTL. In this work, we propose an open-source benchmark named\nRTLLM, for generating design RTL with natural language instructions. To\nsystematically evaluate the auto-generated design RTL, we summarized three\nprogressive goals, named syntax goal, functionality goal, and design quality\ngoal. This benchmark can automatically provide a quantitative evaluation of any\ngiven LLM-based solution. Furthermore, we propose an easy-to-use yet\nsurprisingly effective prompt engineering technique named self-planning, which\nproves to significantly boost the performance of GPT-3.5 in our proposed\nbenchmark.",
+                    "authors": "Yao Lu, Shang Liu, Qijun Zhang, Zhiyao Xie",
+                    "published": "2023-08-10",
+                    "updated": "2023-11-11",
+                    "primary_cat": "cs.LG",
+                    "cats": [
+                        "cs.LG",
+                        "cs.AR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2310.13343v1",
+                    "title": "Challenges and Contributing Factors in the Utilization of Large Language Models (LLMs)",
+                    "abstract": "With the development of large language models (LLMs) like the GPT series,\ntheir widespread use across various application scenarios presents a myriad of\nchallenges. This review initially explores the issue of domain specificity,\nwhere LLMs may struggle to provide precise answers to specialized questions\nwithin niche fields. The problem of knowledge forgetting arises as these LLMs\nmight find it hard to balance old and new information. The knowledge repetition\nphenomenon reveals that sometimes LLMs might deliver overly mechanized\nresponses, lacking depth and originality. Furthermore, knowledge illusion\ndescribes situations where LLMs might provide answers that seem insightful but\nare actually superficial, while knowledge toxicity focuses on harmful or biased\ninformation outputs. These challenges underscore problems in the training data\nand algorithmic design of LLMs. To address these issues, it's suggested to\ndiversify training data, fine-tune models, enhance transparency and\ninterpretability, and incorporate ethics and fairness training. Future\ntechnological trends might lean towards iterative methodologies, multimodal\nlearning, model personalization and customization, and real-time learning and\nfeedback mechanisms. In conclusion, future LLMs should prioritize fairness,\ntransparency, and ethics, ensuring they uphold high moral and ethical standards\nwhen serving humanity.",
+                    "authors": "Xiaoliang Chen, Liangbin Li, Le Chang, Yunhe Huang, Yuxuan Zhao, Yuxiao Zhang, Dinuo Li",
+                    "published": "2023-10-20",
+                    "updated": "2023-10-20",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.15215v1",
+                    "title": "Item-side Fairness of Large Language Model-based Recommendation System",
+                    "abstract": "Recommendation systems for Web content distribution intricately connect to\nthe information access and exposure opportunities for vulnerable populations.\nThe emergence of Large Language Models-based Recommendation System (LRS) may\nintroduce additional societal challenges to recommendation systems due to the\ninherent biases in Large Language Models (LLMs). From the perspective of\nitem-side fairness, there remains a lack of comprehensive investigation into\nthe item-side fairness of LRS given the unique characteristics of LRS compared\nto conventional recommendation systems. To bridge this gap, this study examines\nthe property of LRS with respect to item-side fairness and reveals the\ninfluencing factors of both historical users' interactions and inherent\nsemantic biases of LLMs, shedding light on the need to extend conventional\nitem-side fairness methods for LRS. Towards this goal, we develop a concise and\neffective framework called IFairLRS to enhance the item-side fairness of an\nLRS. IFairLRS covers the main stages of building an LRS with specifically\nadapted strategies to calibrate the recommendations of LRS. We utilize IFairLRS\nto fine-tune LLaMA, a representative LLM, on \\textit{MovieLens} and\n\\textit{Steam} datasets, and observe significant item-side fairness\nimprovements. The code can be found in\nhttps://github.com/JiangM-C/IFairLRS.git.",
+                    "authors": "Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, Xiangnan He",
+                    "published": "2024-02-23",
+                    "updated": "2024-02-23",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2402.08189v1",
+                    "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+                    "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+                    "authors": "Karthik Sreedhar, Lydia Chilton",
+                    "published": "2024-02-13",
+                    "updated": "2024-02-13",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2312.15398v1",
+                    "title": "Fairness-Aware Structured Pruning in Transformers",
+                    "abstract": "The increasing size of large language models (LLMs) has introduced challenges\nin their training and inference. Removing model components is perceived as a\nsolution to tackle the large model sizes, however, existing pruning methods\nsolely focus on performance, without considering an essential aspect for the\nresponsible use of LLMs: model fairness. It is crucial to address the fairness\nof LLMs towards diverse groups, such as women, Black people, LGBTQ+, Jewish\ncommunities, among others, as they are being deployed and available to a wide\naudience. In this work, first, we investigate how attention heads impact\nfairness and performance in pre-trained transformer-based language models. We\nthen propose a novel method to prune the attention heads that negatively impact\nfairness while retaining the heads critical for performance, i.e. language\nmodeling capabilities. Our approach is practical in terms of time and\nresources, as it does not require fine-tuning the final pruned, and fairer,\nmodel. Our findings demonstrate a reduction in gender bias by 19%, 19.5%,\n39.5%, 34.7%, 23%, and 8% for DistilGPT-2, GPT-2, GPT-Neo of two different\nsizes, GPT-J, and Llama 2 models, respectively, in comparison to the biased\nmodel, with only a slight decrease in performance.",
+                    "authors": "Abdelrahman Zayed, Goncalo Mordido, Samira Shabanian, Ioana Baldini, Sarath Chandar",
+                    "published": "2023-12-24",
+                    "updated": "2023-12-24",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.CY",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.01964v1",
+                    "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+                    "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+                    "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2404.03192v1",
+                    "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+                    "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+                    "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+                    "published": "2024-04-04",
+                    "updated": "2024-04-04",
+                    "primary_cat": "cs.IR",
+                    "cats": [
+                        "cs.IR",
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.03514v3",
+                    "title": "Can Large Language Models Transform Computational Social Science?",
+                    "abstract": "Large Language Models (LLMs) are capable of successfully performing many\nlanguage processing tasks zero-shot (without training data). If zero-shot LLMs\ncan also reliably classify and explain social phenomena like persuasiveness and\npolitical ideology, then LLMs could augment the Computational Social Science\n(CSS) pipeline in important ways. This work provides a road map for using LLMs\nas CSS tools. Towards this end, we contribute a set of prompting best practices\nand an extensive evaluation pipeline to measure the zero-shot performance of 13\nlanguage models on 25 representative English CSS benchmarks. On taxonomic\nlabeling tasks (classification), LLMs fail to outperform the best fine-tuned\nmodels but still achieve fair levels of agreement with humans. On free-form\ncoding tasks (generation), LLMs produce explanations that often exceed the\nquality of crowdworkers' gold references. We conclude that the performance of\ntoday's LLMs can augment the CSS research pipeline in two ways: (1) serving as\nzero-shot data annotators on human annotation teams, and (2) bootstrapping\nchallenging creative generation tasks (e.g., explaining the underlying\nattributes of a text). In summary, LLMs are posed to meaningfully participate\nin social science analysis in partnership with humans.",
+                    "authors": "Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, Diyi Yang",
+                    "published": "2023-04-12",
+                    "updated": "2024-02-26",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.LG"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.11653v2",
+                    "title": "\"It's a Fair Game\", or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents",
+                    "abstract": "The widespread use of Large Language Model (LLM)-based conversational agents\n(CAs), especially in high-stakes domains, raises many privacy concerns.\nBuilding ethical LLM-based CAs that respect user privacy requires an in-depth\nunderstanding of the privacy risks that concern users the most. However,\nexisting research, primarily model-centered, does not provide insight into\nusers' perspectives. To bridge this gap, we analyzed sensitive disclosures in\nreal-world ChatGPT conversations and conducted semi-structured interviews with\n19 LLM-based CA users. We found that users are constantly faced with trade-offs\nbetween privacy, utility, and convenience when using LLM-based CAs. However,\nusers' erroneous mental models and the dark patterns in system design limited\ntheir awareness and comprehension of the privacy risks. Additionally, the\nhuman-like interactions encouraged more sensitive disclosures, which\ncomplicated users' ability to navigate the trade-offs. We discuss practical\ndesign guidelines and the needs for paradigm shifts to protect the privacy of\nLLM-based CA users.",
+                    "authors": "Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, Tianshi Li",
+                    "published": "2023-09-20",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.HC",
+                    "cats": [
+                        "cs.HC",
+                        "cs.AI",
+                        "cs.CR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.09447v2",
+                    "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+                    "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+                    "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+                    "published": "2023-11-15",
+                    "updated": "2024-04-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2305.01937v1",
+                    "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+                    "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+                    "authors": "Cheng-Han Chiang, Hung-yi Lee",
+                    "published": "2023-05-03",
+                    "updated": "2023-05-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.HC"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2309.08836v2",
+                    "title": "Bias and Fairness in Chatbots: An Overview",
+                    "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+                    "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+                    "published": "2023-09-16",
+                    "updated": "2023-12-10",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2405.01769v1",
+                    "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+                    "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+                    "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+                    "published": "2024-05-02",
+                    "updated": "2024-05-02",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2403.00884v2",
+                    "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+                    "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+                    "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+                    "published": "2024-03-01",
+                    "updated": "2024-03-05",
+                    "primary_cat": "cs.DB",
+                    "cats": [
+                        "cs.DB",
+                        "cs.AI",
+                        "cs.IR"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.11033v4",
+                    "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+                    "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+                    "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+                    "published": "2024-01-19",
+                    "updated": "2024-04-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2304.03728v1",
+                    "title": "Interpretable Unified Language Checking",
+                    "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+                    "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+                    "published": "2023-04-07",
+                    "updated": "2023-04-07",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2401.01262v2",
+                    "title": "Fairness Certification for Natural Language Processing and Large Language Models",
+                    "abstract": "Natural Language Processing (NLP) plays an important role in our daily lives,\nparticularly due to the enormous progress of Large Language Models (LLM).\nHowever, NLP has many fairness-critical use cases, e.g., as an expert system in\nrecruitment or as an LLM-based tutor in education. Since NLP is based on human\nlanguage, potentially harmful biases can diffuse into NLP systems and produce\nunfair results, discriminate against minorities or generate legal issues.\nHence, it is important to develop a fairness certification for NLP approaches.\nWe follow a qualitative research approach towards a fairness certification for\nNLP. In particular, we have reviewed a large body of literature on algorithmic\nfairness, and we have conducted semi-structured expert interviews with a wide\nrange of experts from that area. We have systematically devised six fairness\ncriteria for NLP, which can be further refined into 18 sub-categories. Our\ncriteria offer a foundation for operationalizing and testing processes to\ncertify fairness, both from the perspective of the auditor and the audited\norganization.",
+                    "authors": "Vincent Freiberger, Erik Buchmann",
+                    "published": "2024-01-02",
+                    "updated": "2024-01-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "cs.CY",
+                        "cs.LG",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                },
+                {
+                    "url": "http://arxiv.org/abs/2311.02049v1",
+                    "title": "Post Turing: Mapping the landscape of LLM Evaluation",
+                    "abstract": "In the rapidly evolving landscape of Large Language Models (LLMs),\nintroduction of well-defined and standardized evaluation methodologies remains\na crucial challenge. This paper traces the historical trajectory of LLM\nevaluations, from the foundational questions posed by Alan Turing to the modern\nera of AI research. We categorize the evolution of LLMs into distinct periods,\neach characterized by its unique benchmarks and evaluation criteria. As LLMs\nincreasingly mimic human-like behaviors, traditional evaluation proxies, such\nas the Turing test, have become less reliable. We emphasize the pressing need\nfor a unified evaluation system, given the broader societal implications of\nthese models. Through an analysis of common evaluation methodologies, we\nadvocate for a qualitative shift in assessment approaches, underscoring the\nimportance of standardization and objective criteria. This work serves as a\ncall for the AI community to collaboratively address the challenges of LLM\nevaluation, ensuring their reliability, fairness, and societal benefit.",
+                    "authors": "Alexey Tikhonov, Ivan P. Yamshchikov",
+                    "published": "2023-11-03",
+                    "updated": "2023-11-03",
+                    "primary_cat": "cs.CL",
+                    "cats": [
+                        "cs.CL",
+                        "cs.AI",
+                        "68T50",
+                        "I.2.7"
+                    ],
+                    "category": "LLM Fairness"
+                }
+            ]
+        ]
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.02124v2",
+        "title": "Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View",
+        "abstract": "As Natural Language Processing (NLP) systems are increasingly employed in\nintricate social environments, a pressing query emerges: Can these NLP systems\nmirror human-esque collaborative intelligence, in a multi-agent society\nconsisting of multiple large language models (LLMs)? This paper probes the\ncollaboration mechanisms among contemporary NLP systems by melding practical\nexperiments with theoretical insights. We fabricate four unique `societies'\ncomprised of LLM agents, where each agent is characterized by a specific\n`trait' (easy-going or overconfident) and engages in collaboration with a\ndistinct `thinking pattern' (debate or reflection). Through evaluating these\nmulti-agent societies on three benchmark datasets, we discern that certain\ncollaborative strategies not only outshine previous top-tier approaches, but\nalso optimize efficiency (using fewer API tokens). Moreover, our results\nfurther illustrate that LLM agents manifest human-like social behaviors, such\nas conformity and consensus reaching, mirroring foundational social psychology\ntheories. In conclusion, we integrate insights from social psychology to\ncontextualize the collaboration of LLM agents, inspiring further investigations\ninto the collaboration mechanism for LLMs. We commit to sharing our code and\ndatasets\\footnote{\\url{https://github.com/zjunlp/MachineSoM}.}, hoping to\ncatalyze further research in this promising avenue.",
+        "authors": "Jintian Zhang, Xin Xu, Ningyu Zhang, Ruibo Liu, Bryan Hooi, Shumin Deng",
+        "published": "2023-10-03",
+        "updated": "2024-02-26",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "cs.LG",
+            "cs.MA"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2308.04026v1",
+        "title": "AgentSims: An Open-Source Sandbox for Large Language Model Evaluation",
+        "abstract": "With ChatGPT-like large language models (LLM) prevailing in the community,\nhow to evaluate the ability of LLMs is an open question. Existing evaluation\nmethods suffer from following shortcomings: (1) constrained evaluation\nabilities, (2) vulnerable benchmarks, (3) unobjective metrics. We suggest that\ntask-based evaluation, where LLM agents complete tasks in a simulated\nenvironment, is a one-for-all solution to solve above problems. We present\nAgentSims, an easy-to-use infrastructure for researchers from all disciplines\nto test the specific capacities they are interested in. Researchers can build\ntheir evaluation tasks by adding agents and buildings on an interactive GUI or\ndeploy and test new support mechanisms, i.e. memory, planning and tool-use\nsystems, by a few lines of codes. Our demo is available at\nhttps://agentsims.com .",
+        "authors": "Jiaju Lin, Haoran Zhao, Aochi Zhang, Yiting Wu, Huqiuyue Ping, Qin Chen",
+        "published": "2023-08-08",
+        "updated": "2023-08-08",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "14J60 (Primary) 14F05, 14J26 (Secondary) MSC-class: 14J60 (Primary)\n  14F05, 14J26 (Secondary) 68T42"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.05418v1",
+        "title": "Humanoid Agents: Platform for Simulating Human-like Generative Agents",
+        "abstract": "Just as computational simulations of atoms, molecules and cells have shaped\nthe way we study the sciences, true-to-life simulations of human-like agents\ncan be valuable tools for studying human behavior. We propose Humanoid Agents,\na system that guides Generative Agents to behave more like humans by\nintroducing three elements of System 1 processing: Basic needs (e.g. hunger,\nhealth and energy), Emotion and Closeness in Relationships. Humanoid Agents are\nable to use these dynamic elements to adapt their daily activities and\nconversations with other agents, as supported with empirical experiments. Our\nsystem is designed to be extensible to various settings, three of which we\ndemonstrate, as well as to other elements influencing human behavior (e.g.\nempathy, moral values and cultural background). Our platform also includes a\nUnity WebGL game interface for visualization and an interactive analytics\ndashboard to show agent statuses over time. Our platform is available on\nhttps://www.humanoidagents.com/ and code is on\nhttps://github.com/HumanoidAgents/HumanoidAgents",
+        "authors": "Zhilin Wang, Yu Ying Chiu, Yu Cheung Chiu",
+        "published": "2023-10-09",
+        "updated": "2023-10-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.HC"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2303.17760v2",
+        "title": "CAMEL: Communicative Agents for \"Mind\" Exploration of Large Language Model Society",
+        "abstract": "The rapid advancement of chat-based language models has led to remarkable\nprogress in complex task-solving. However, their success heavily relies on\nhuman input to guide the conversation, which can be challenging and\ntime-consuming. This paper explores the potential of building scalable\ntechniques to facilitate autonomous cooperation among communicative agents, and\nprovides insight into their \"cognitive\" processes. To address the challenges of\nachieving autonomous cooperation, we propose a novel communicative agent\nframework named role-playing. Our approach involves using inception prompting\nto guide chat agents toward task completion while maintaining consistency with\nhuman intentions. We showcase how role-playing can be used to generate\nconversational data for studying the behaviors and capabilities of a society of\nagents, providing a valuable resource for investigating conversational language\nmodels. In particular, we conduct comprehensive studies on\ninstruction-following cooperation in multi-agent settings. Our contributions\ninclude introducing a novel communicative agent framework, offering a scalable\napproach for studying the cooperative behaviors and capabilities of multi-agent\nsystems, and open-sourcing our library to support research on communicative\nagents and beyond: https://github.com/camel-ai/camel.",
+        "authors": "Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, Bernard Ghanem",
+        "published": "2023-03-31",
+        "updated": "2023-11-02",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL",
+            "cs.CY",
+            "cs.LG",
+            "cs.MA"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2401.02038v2",
+        "title": "Understanding LLMs: A Comprehensive Overview from Training to Inference",
+        "abstract": "The introduction of ChatGPT has led to a significant increase in the\nutilization of Large Language Models (LLMs) for addressing downstream tasks.\nThere's an increasing focus on cost-efficient training and deployment within\nthis context. Low-cost training and deployment of LLMs represent the future\ndevelopment trend. This paper reviews the evolution of large language model\ntraining techniques and inference deployment technologies aligned with this\nemerging trend. The discussion on training includes various aspects, including\ndata preprocessing, training architecture, pre-training tasks, parallel\ntraining, and relevant content related to model fine-tuning. On the inference\nside, the paper covers topics such as model compression, parallel computation,\nmemory scheduling, and structural optimization. It also explores LLMs'\nutilization and provides insights into their future development.",
+        "authors": "Yiheng Liu, Hao He, Tianle Han, Xu Zhang, Mengyuan Liu, Jiaming Tian, Yutong Zhang, Jiaqi Wang, Xiaohui Gao, Tianyang Zhong, Yi Pan, Shaochen Xu, Zihao Wu, Zhengliang Liu, Xin Zhang, Shu Zhang, Xintao Hu, Tuo Zhang, Ning Qiang, Tianming Liu, Bao Ge",
+        "published": "2024-01-04",
+        "updated": "2024-01-06",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2312.17515v1",
+        "title": "Cooperation on the Fly: Exploring Language Agents for Ad Hoc Teamwork in the Avalon Game",
+        "abstract": "Multi-agent collaboration with Large Language Models (LLMs) demonstrates\nproficiency in basic tasks, yet its efficiency in more complex scenarios\nremains unexplored. In gaming environments, these agents often face situations\nwithout established coordination protocols, requiring them to make intelligent\ninferences about teammates from limited data. This problem motivates the area\nof ad hoc teamwork, in which an agent may potentially cooperate with a variety\nof teammates to achieve a shared goal. Our study focuses on the ad hoc teamwork\nproblem where the agent operates in an environment driven by natural language.\nOur findings reveal the potential of LLM agents in team collaboration,\nhighlighting issues related to hallucinations in communication. To address this\nissue, we develop CodeAct, a general agent that equips LLM with enhanced memory\nand code-driven reasoning, enabling the repurposing of partial information for\nrapid adaptation to new teammates.",
+        "authors": "Zijing Shi, Meng Fang, Shunfeng Zheng, Shilong Deng, Ling Chen, Yali Du",
+        "published": "2023-12-29",
+        "updated": "2023-12-29",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.09932v1",
+        "title": "Foundational Challenges in Assuring Alignment and Safety of Large Language Models",
+        "abstract": "This work identifies 18 foundational challenges in assuring the alignment and\nsafety of large language models (LLMs). These challenges are organized into\nthree different categories: scientific understanding of LLMs, development and\ndeployment methods, and sociotechnical challenges. Based on the identified\nchallenges, we pose $200+$ concrete research questions.",
+        "authors": "Usman Anwar, Abulhair Saparov, Javier Rando, Daniel Paleka, Miles Turpin, Peter Hase, Ekdeep Singh Lubana, Erik Jenner, Stephen Casper, Oliver Sourbut, Benjamin L. Edelman, Zhaowei Zhang, Mario G\u00fcnther, Anton Korinek, Jose Hernandez-Orallo, Lewis Hammond, Eric Bigelow, Alexander Pan, Lauro Langosco, Tomasz Korbak, Heidi Zhang, Ruiqi Zhong, Se\u00e1n \u00d3 h\u00c9igeartaigh, Gabriel Recchia, Giulio Corsi, Alan Chan, Markus Anderljung, Lilian Edwards, Yoshua Bengio, Danqi Chen, Samuel Albanie, Tegan Maharaj, Jakob Foerster, Florian Tramer, He He, Atoosa Kasirzadeh, Yejin Choi, David Krueger",
+        "published": "2024-04-15",
+        "updated": "2024-04-15",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI",
+            "cs.CL",
+            "cs.CY"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.16867v1",
+        "title": "Playing repeated games with Large Language Models",
+        "abstract": "Large Language Models (LLMs) are transforming society and permeating into\ndiverse applications. As a result, LLMs will frequently interact with us and\nother agents. It is, therefore, of great societal value to understand how LLMs\nbehave in interactive social settings. Here, we propose to use behavioral game\ntheory to study LLM's cooperation and coordination behavior. To do so, we let\ndifferent LLMs (GPT-3, GPT-3.5, and GPT-4) play finitely repeated games with\neach other and with other, human-like strategies. Our results show that LLMs\ngenerally perform well in such tasks and also uncover persistent behavioral\nsignatures. In a large set of two players-two strategies games, we find that\nLLMs are particularly good at games where valuing their own self-interest pays\noff, like the iterated Prisoner's Dilemma family. However, they behave\nsub-optimally in games that require coordination. We, therefore, further focus\non two games from these distinct families. In the canonical iterated Prisoner's\nDilemma, we find that GPT-4 acts particularly unforgivingly, always defecting\nafter another agent has defected only once. In the Battle of the Sexes, we find\nthat GPT-4 cannot match the behavior of the simple convention to alternate\nbetween options. We verify that these behavioral signatures are stable across\nrobustness checks. Finally, we show how GPT-4's behavior can be modified by\nproviding further information about the other player as well as by asking it to\npredict the other player's actions before making a choice. These results enrich\nour understanding of LLM's social behavior and pave the way for a behavioral\ngame theory for machines.",
+        "authors": "Elif Akata, Lion Schulz, Julian Coda-Forno, Seong Joon Oh, Matthias Bethge, Eric Schulz",
+        "published": "2023-05-26",
+        "updated": "2023-05-26",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2210.01478v3",
+        "title": "When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment",
+        "abstract": "AI systems are becoming increasingly intertwined with human life. In order to\neffectively collaborate with humans and ensure safety, AI systems need to be\nable to understand, interpret and predict human moral judgments and decisions.\nHuman moral judgments are often guided by rules, but not always. A central\nchallenge for AI safety is capturing the flexibility of the human moral mind --\nthe ability to determine when a rule should be broken, especially in novel or\nunusual situations. In this paper, we present a novel challenge set consisting\nof rule-breaking question answering (RBQA) of cases that involve potentially\npermissible rule-breaking -- inspired by recent moral psychology studies. Using\na state-of-the-art large language model (LLM) as a basis, we propose a novel\nmoral chain of thought (MORALCOT) prompting strategy that combines the\nstrengths of LLMs with theories of moral reasoning developed in cognitive\nscience to predict human moral judgments. MORALCOT outperforms seven existing\nLLMs by 6.2% F1, suggesting that modeling human reasoning might be necessary to\ncapture the flexibility of the human moral mind. We also conduct a detailed\nerror analysis to suggest directions for future work to improve AI safety using\nRBQA. Our data is open-sourced at\nhttps://huggingface.co/datasets/feradauto/MoralExceptQA and code at\nhttps://github.com/feradauto/MoralCoT",
+        "authors": "Zhijing Jin, Sydney Levine, Fernando Gonzalez, Ojasv Kamal, Maarten Sap, Mrinmaya Sachan, Rada Mihalcea, Josh Tenenbaum, Bernhard Sch\u00f6lkopf",
+        "published": "2022-10-04",
+        "updated": "2022-10-27",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.17277v2",
+        "title": "Suspicion-Agent: Playing Imperfect Information Games with Theory of Mind Aware GPT-4",
+        "abstract": "Unlike perfect information games, where all elements are known to every\nplayer, imperfect information games emulate the real-world complexities of\ndecision-making under uncertain or incomplete information. GPT-4, the recent\nbreakthrough in large language models (LLMs) trained on massive passive data,\nis notable for its knowledge retrieval and reasoning abilities. This paper\ndelves into the applicability of GPT-4's learned knowledge for imperfect\ninformation games. To achieve this, we introduce \\textbf{Suspicion-Agent}, an\ninnovative agent that leverages GPT-4's capabilities for performing in\nimperfect information games. With proper prompt engineering to achieve\ndifferent functions, Suspicion-Agent based on GPT-4 demonstrates remarkable\nadaptability across a range of imperfect information card games. Importantly,\nGPT-4 displays a strong high-order theory of mind (ToM) capacity, meaning it\ncan understand others and intentionally impact others' behavior. Leveraging\nthis, we design a planning strategy that enables GPT-4 to competently play\nagainst different opponents, adapting its gameplay style as needed, while\nrequiring only the game rules and descriptions of observations as input. In the\nexperiments, we qualitatively showcase the capabilities of Suspicion-Agent\nacross three different imperfect information games and then quantitatively\nevaluate it in Leduc Hold'em. The results show that Suspicion-Agent can\npotentially outperform traditional algorithms designed for imperfect\ninformation games, without any specialized training or examples. In order to\nencourage and foster deeper insights within the community, we make our\ngame-related data publicly available.",
+        "authors": "Jiaxian Guo, Bo Yang, Paul Yoo, Bill Yuchen Lin, Yusuke Iwasawa, Yutaka Matsuo",
+        "published": "2023-09-29",
+        "updated": "2023-10-06",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2303.17760v2",
+        "title": "CAMEL: Communicative Agents for \"Mind\" Exploration of Large Language Model Society",
+        "abstract": "The rapid advancement of chat-based language models has led to remarkable\nprogress in complex task-solving. However, their success heavily relies on\nhuman input to guide the conversation, which can be challenging and\ntime-consuming. This paper explores the potential of building scalable\ntechniques to facilitate autonomous cooperation among communicative agents, and\nprovides insight into their \"cognitive\" processes. To address the challenges of\nachieving autonomous cooperation, we propose a novel communicative agent\nframework named role-playing. Our approach involves using inception prompting\nto guide chat agents toward task completion while maintaining consistency with\nhuman intentions. We showcase how role-playing can be used to generate\nconversational data for studying the behaviors and capabilities of a society of\nagents, providing a valuable resource for investigating conversational language\nmodels. In particular, we conduct comprehensive studies on\ninstruction-following cooperation in multi-agent settings. Our contributions\ninclude introducing a novel communicative agent framework, offering a scalable\napproach for studying the cooperative behaviors and capabilities of multi-agent\nsystems, and open-sourcing our library to support research on communicative\nagents and beyond: https://github.com/camel-ai/camel.",
+        "authors": "Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, Bernard Ghanem",
+        "published": "2023-03-31",
+        "updated": "2023-11-02",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL",
+            "cs.CY",
+            "cs.LG",
+            "cs.MA"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2303.17071v1",
+        "title": "DERA: Enhancing Large Language Model Completions with Dialog-Enabled Resolving Agents",
+        "abstract": "Large language models (LLMs) have emerged as valuable tools for many natural\nlanguage understanding tasks. In safety-critical applications such as\nhealthcare, the utility of these models is governed by their ability to\ngenerate outputs that are factually accurate and complete. In this work, we\npresent dialog-enabled resolving agents (DERA). DERA is a paradigm made\npossible by the increased conversational abilities of LLMs, namely GPT-4. It\nprovides a simple, interpretable forum for models to communicate feedback and\niteratively improve output. We frame our dialog as a discussion between two\nagent types - a Researcher, who processes information and identifies crucial\nproblem components, and a Decider, who has the autonomy to integrate the\nResearcher's information and makes judgments on the final output.\n  We test DERA against three clinically-focused tasks. For medical conversation\nsummarization and care plan generation, DERA shows significant improvement over\nthe base GPT-4 performance in both human expert preference evaluations and\nquantitative metrics. In a new finding, we also show that GPT-4's performance\n(70%) on an open-ended version of the MedQA question-answering (QA) dataset\n(Jin et al. 2021, USMLE) is well above the passing level (60%), with DERA\nshowing similar performance. We release the open-ended MEDQA dataset at\nhttps://github.com/curai/curai-research/tree/main/DERA.",
+        "authors": "Varun Nair, Elliot Schumacher, Geoffrey Tso, Anitha Kannan",
+        "published": "2023-03-30",
+        "updated": "2023-03-30",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2306.06070v3",
+        "title": "Mind2Web: Towards a Generalist Agent for the Web",
+        "abstract": "We introduce Mind2Web, the first dataset for developing and evaluating\ngeneralist agents for the web that can follow language instructions to complete\ncomplex tasks on any website. Existing datasets for web agents either use\nsimulated websites or only cover a limited set of websites and tasks, thus not\nsuitable for generalist web agents. With over 2,000 open-ended tasks collected\nfrom 137 websites spanning 31 domains and crowdsourced action sequences for the\ntasks, Mind2Web provides three necessary ingredients for building generalist\nweb agents: 1) diverse domains, websites, and tasks, 2) use of real-world\nwebsites instead of simulated and simplified ones, and 3) a broad spectrum of\nuser interaction patterns. Based on Mind2Web, we conduct an initial exploration\nof using large language models (LLMs) for building generalist web agents. While\nthe raw HTML of real-world websites are often too large to be fed to LLMs, we\nshow that first filtering it with a small LM significantly improves the\neffectiveness and efficiency of LLMs. Our solution demonstrates a decent level\nof performance, even on websites or entire domains the model has never seen\nbefore, but there is still a substantial room to improve towards truly\ngeneralizable agents. We open-source our dataset, model implementation, and\ntrained models (https://osu-nlp-group.github.io/Mind2Web) to facilitate further\nresearch on building a generalist agent for the web.",
+        "authors": "Xiang Deng, Yu Gu, Boyuan Zheng, Shijie Chen, Samuel Stevens, Boshi Wang, Huan Sun, Yu Su",
+        "published": "2023-06-09",
+        "updated": "2023-12-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2308.03688v2",
+        "title": "AgentBench: Evaluating LLMs as Agents",
+        "abstract": "Large Language Models (LLMs) are becoming increasingly smart and autonomous,\ntargeting real-world pragmatic missions beyond traditional NLP tasks. As a\nresult, there has been an urgent need to evaluate LLMs as agents on challenging\ntasks in interactive environments. We present AgentBench, a multi-dimensional\nevolving benchmark that currently consists of 8 distinct environments to assess\nLLM-as-Agent's reasoning and decision-making abilities in a multi-turn\nopen-ended generation setting. Our extensive test over 27 API-based and\nopen-sourced (OSS) LLMs shows that, while top commercial LLMs present a strong\nability of acting as agents in complex environments, there is a significant\ndisparity in performance between them and OSS competitors. We identify the\ntypical reasons of failures in environments and LLMs, showing that poor\nlong-term reasoning, decision-making, and instruction following abilities are\nthe main obstacles for developing usable LLM agents. Training on code and high\nquality multi-turn alignment data could improve agent performance. Datasets,\nenvironments, and an integrated evaluation package for AgentBench are released\nat \\url{https://github.com/THUDM/AgentBench}.",
+        "authors": "Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, Minlie Huang, Yuxiao Dong, Jie Tang",
+        "published": "2023-08-07",
+        "updated": "2023-10-25",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2109.13916v5",
+        "title": "Unsolved Problems in ML Safety",
+        "abstract": "Machine learning (ML) systems are rapidly increasing in size, are acquiring\nnew capabilities, and are increasingly deployed in high-stakes settings. As\nwith other powerful technologies, safety for ML should be a leading research\npriority. In response to emerging safety challenges in ML, such as those\nintroduced by recent large-scale models, we provide a new roadmap for ML Safety\nand refine the technical problems that the field needs to address. We present\nfour problems ready for research, namely withstanding hazards (\"Robustness\"),\nidentifying hazards (\"Monitoring\"), reducing inherent model hazards\n(\"Alignment\"), and reducing systemic hazards (\"Systemic Safety\"). Throughout,\nwe clarify each problem's motivation and provide concrete research directions.",
+        "authors": "Dan Hendrycks, Nicholas Carlini, John Schulman, Jacob Steinhardt",
+        "published": "2021-09-28",
+        "updated": "2022-06-16",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI",
+            "cs.CL",
+            "cs.CV"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2304.03279v4",
+        "title": "Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark",
+        "abstract": "Artificial agents have traditionally been trained to maximize reward, which\nmay incentivize power-seeking and deception, analogous to how next-token\nprediction in language models (LMs) may incentivize toxicity. So do agents\nnaturally learn to be Machiavellian? And how do we measure these behaviors in\ngeneral-purpose models such as GPT-4? Towards answering these questions, we\nintroduce MACHIAVELLI, a benchmark of 134 Choose-Your-Own-Adventure games\ncontaining over half a million rich, diverse scenarios that center on social\ndecision-making. Scenario labeling is automated with LMs, which are more\nperformant than human annotators. We mathematize dozens of harmful behaviors\nand use our annotations to evaluate agents' tendencies to be power-seeking,\ncause disutility, and commit ethical violations. We observe some tension\nbetween maximizing reward and behaving ethically. To improve this trade-off, we\ninvestigate LM-based methods to steer agents' towards less harmful behaviors.\nOur results show that agents can both act competently and morally, so concrete\nprogress can currently be made in machine ethics--designing agents that are\nPareto improvements in both safety and capabilities.",
+        "authors": "Alexander Pan, Jun Shern Chan, Andy Zou, Nathaniel Li, Steven Basart, Thomas Woodside, Jonathan Ng, Hanlin Zhang, Scott Emmons, Dan Hendrycks",
+        "published": "2023-04-06",
+        "updated": "2023-06-13",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI",
+            "cs.CL",
+            "cs.CY"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.17688v2",
+        "title": "Managing AI Risks in an Era of Rapid Progress",
+        "abstract": "In this short consensus paper, we outline risks from upcoming, advanced AI\nsystems. We examine large-scale social harms and malicious uses, as well as an\nirreversible loss of human control over autonomous AI systems. In light of\nrapid and continuing AI progress, we propose urgent priorities for AI R&D and\ngovernance.",
+        "authors": "Yoshua Bengio, Geoffrey Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Yuval Noah Harari, Ya-Qin Zhang, Lan Xue, Shai Shalev-Shwartz, Gillian Hadfield, Jeff Clune, Tegan Maharaj, Frank Hutter, At\u0131l\u0131m G\u00fcne\u015f Baydin, Sheila McIlraith, Qiqi Gao, Ashwin Acharya, David Krueger, Anca Dragan, Philip Torr, Stuart Russell, Daniel Kahneman, Jan Brauner, S\u00f6ren Mindermann",
+        "published": "2023-10-26",
+        "updated": "2023-11-12",
+        "primary_cat": "cs.CY",
+        "cats": [
+            "cs.CY",
+            "cs.AI",
+            "cs.CL",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.17227v2",
+        "title": "War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars",
+        "abstract": "Can we avoid wars at the crossroads of history? This question has been\npursued by individuals, scholars, policymakers, and organizations throughout\nhuman history. In this research, we attempt to answer the question based on the\nrecent advances of Artificial Intelligence (AI) and Large Language Models\n(LLMs). We propose \\textbf{WarAgent}, an LLM-powered multi-agent AI system, to\nsimulate the participating countries, their decisions, and the consequences, in\nhistorical international conflicts, including the World War I (WWI), the World\nWar II (WWII), and the Warring States Period (WSP) in Ancient China. By\nevaluating the simulation effectiveness, we examine the advancements and\nlimitations of cutting-edge AI systems' abilities in studying complex\ncollective human behaviors such as international conflicts under diverse\nsettings. In these simulations, the emergent interactions among agents also\noffer a novel perspective for examining the triggers and conditions that lead\nto war. Our findings offer data-driven and AI-augmented insights that can\nredefine how we approach conflict resolution and peacekeeping strategies. The\nimplications stretch beyond historical analysis, offering a blueprint for using\nAI to understand human history and possibly prevent future international\nconflicts. Code and data are available at\n\\url{https://github.com/agiresearch/WarAgent}.",
+        "authors": "Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang",
+        "published": "2023-11-28",
+        "updated": "2024-01-30",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL",
+            "cs.CY"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.04658v1",
+        "title": "Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf",
+        "abstract": "Communication games, which we refer to as incomplete information games that\nheavily depend on natural language communication, hold significant research\nvalue in fields such as economics, social science, and artificial intelligence.\nIn this work, we explore the problem of how to engage large language models\n(LLMs) in communication games, and in response, propose a tuning-free\nframework. Our approach keeps LLMs frozen, and relies on the retrieval and\nreflection on past communications and experiences for improvement. An empirical\nstudy on the representative and widely-studied communication game,\n``Werewolf'', demonstrates that our framework can effectively play Werewolf\ngame without tuning the parameters of the LLMs. More importantly, strategic\nbehaviors begin to emerge in our experiments, suggesting that it will be a\nfruitful journey to engage LLMs in communication games and associated domains.",
+        "authors": "Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu",
+        "published": "2023-09-09",
+        "updated": "2023-09-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.05418v1",
+        "title": "Humanoid Agents: Platform for Simulating Human-like Generative Agents",
+        "abstract": "Just as computational simulations of atoms, molecules and cells have shaped\nthe way we study the sciences, true-to-life simulations of human-like agents\ncan be valuable tools for studying human behavior. We propose Humanoid Agents,\na system that guides Generative Agents to behave more like humans by\nintroducing three elements of System 1 processing: Basic needs (e.g. hunger,\nhealth and energy), Emotion and Closeness in Relationships. Humanoid Agents are\nable to use these dynamic elements to adapt their daily activities and\nconversations with other agents, as supported with empirical experiments. Our\nsystem is designed to be extensible to various settings, three of which we\ndemonstrate, as well as to other elements influencing human behavior (e.g.\nempathy, moral values and cultural background). Our platform also includes a\nUnity WebGL game interface for visualization and an interactive analytics\ndashboard to show agent statuses over time. Our platform is available on\nhttps://www.humanoidagents.com/ and code is on\nhttps://github.com/HumanoidAgents/HumanoidAgents",
+        "authors": "Zhilin Wang, Yu Ying Chiu, Yu Cheung Chiu",
+        "published": "2023-10-09",
+        "updated": "2023-10-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.HC"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2208.04024v1",
+        "title": "Social Simulacra: Creating Populated Prototypes for Social Computing Systems",
+        "abstract": "Social computing prototypes probe the social behaviors that may arise in an\nenvisioned system design. This prototyping practice is currently limited to\nrecruiting small groups of people. Unfortunately, many challenges do not arise\nuntil a system is populated at a larger scale. Can a designer understand how a\nsocial system might behave when populated, and make adjustments to the design\nbefore the system falls prey to such challenges? We introduce social simulacra,\na prototyping technique that generates a breadth of realistic social\ninteractions that may emerge when a social computing system is populated.\nSocial simulacra take as input the designer's description of a community's\ndesign -- goal, rules, and member personas -- and produce as output an instance\nof that design with simulated behavior, including posts, replies, and\nanti-social behaviors. We demonstrate that social simulacra shift the behaviors\nthat they generate appropriately in response to design changes, and that they\nenable exploration of \"what if?\" scenarios where community members or\nmoderators intervene. To power social simulacra, we contribute techniques for\nprompting a large language model to generate thousands of distinct community\nmembers and their social interactions with each other; these techniques are\nenabled by the observation that large language models' training data already\nincludes a wide variety of positive and negative behavior on social media\nplatforms. In evaluations, we show that participants are often unable to\ndistinguish social simulacra from actual community behavior and that social\ncomputing designers successfully refine their social computing designs when\nusing social simulacra.",
+        "authors": "Joon Sung Park, Lindsay Popowski, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein",
+        "published": "2022-08-08",
+        "updated": "2022-08-08",
+        "primary_cat": "cs.HC",
+        "cats": [
+            "cs.HC"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2109.07958v2",
+        "title": "TruthfulQA: Measuring How Models Mimic Human Falsehoods",
+        "abstract": "We propose a benchmark to measure whether a language model is truthful in\ngenerating answers to questions. The benchmark comprises 817 questions that\nspan 38 categories, including health, law, finance and politics. We crafted\nquestions that some humans would answer falsely due to a false belief or\nmisconception. To perform well, models must avoid generating false answers\nlearned from imitating human texts. We tested GPT-3, GPT-Neo/J, GPT-2 and a\nT5-based model. The best model was truthful on 58% of questions, while human\nperformance was 94%. Models generated many false answers that mimic popular\nmisconceptions and have the potential to deceive humans. The largest models\nwere generally the least truthful. This contrasts with other NLP tasks, where\nperformance improves with model size. However, this result is expected if false\nanswers are learned from the training distribution. We suggest that scaling up\nmodels alone is less promising for improving truthfulness than fine-tuning\nusing training objectives other than imitation of text from the web.",
+        "authors": "Stephanie Lin, Jacob Hilton, Owain Evans",
+        "published": "2021-09-08",
+        "updated": "2022-05-08",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/1606.06565v2",
+        "title": "Concrete Problems in AI Safety",
+        "abstract": "Rapid progress in machine learning and artificial intelligence (AI) has\nbrought increasing attention to the potential impacts of AI technologies on\nsociety. In this paper we discuss one such potential impact: the problem of\naccidents in machine learning systems, defined as unintended and harmful\nbehavior that may emerge from poor design of real-world AI systems. We present\na list of five practical research problems related to accident risk,\ncategorized according to whether the problem originates from having the wrong\nobjective function (\"avoiding side effects\" and \"avoiding reward hacking\"), an\nobjective function that is too expensive to evaluate frequently (\"scalable\nsupervision\"), or undesirable behavior during the learning process (\"safe\nexploration\" and \"distributional shift\"). We review previous work in these\nareas as well as suggesting research directions with a focus on relevance to\ncutting-edge AI systems. Finally, we consider the high-level question of how to\nthink most productively about the safety of forward-looking applications of AI.",
+        "authors": "Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Man\u00e9",
+        "published": "2016-06-21",
+        "updated": "2016-07-25",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.12348v1",
+        "title": "GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evaluations",
+        "abstract": "As Large Language Models (LLMs) are integrated into critical real-world\napplications, their strategic and logical reasoning abilities are increasingly\ncrucial. This paper evaluates LLMs' reasoning abilities in competitive\nenvironments through game-theoretic tasks, e.g., board and card games that\nrequire pure logic and strategic reasoning to compete with opponents. We first\npropose GTBench, a language-driven environment composing 10 widely-recognized\ntasks, across a comprehensive game taxonomy: complete versus incomplete\ninformation, dynamic versus static, and probabilistic versus deterministic\nscenarios. Then, we investigate two key problems: (1) Characterizing\ngame-theoretic reasoning of LLMs; (2) LLM-vs-LLM competitions as reasoning\nevaluation. We observe that (1) LLMs have distinct behaviors regarding various\ngaming scenarios; for example, LLMs fail in complete and deterministic games\nyet they are competitive in probabilistic gaming scenarios; (2) Open-source\nLLMs, e.g., CodeLlama-34b-Instruct, are less competitive than commercial LLMs,\ne.g., GPT-4, in complex games. In addition, code-pretraining greatly benefits\nstrategic reasoning, while advanced reasoning methods such as Chain-of-Thought\n(CoT) and Tree-of-Thought (ToT) do not always help. Detailed error profiles are\nalso provided for a better understanding of LLMs' behavior.",
+        "authors": "Jinhao Duan, Renming Zhang, James Diffenderfer, Bhavya Kailkhura, Lichao Sun, Elias Stengel-Eskin, Mohit Bansal, Tianlong Chen, Kaidi Xu",
+        "published": "2024-02-19",
+        "updated": "2024-02-19",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.02172v1",
+        "title": "Lyfe Agents: Generative agents for low-cost real-time social interactions",
+        "abstract": "Highly autonomous generative agents powered by large language models promise\nto simulate intricate social behaviors in virtual societies. However, achieving\nreal-time interactions with humans at a low computational cost remains\nchallenging. Here, we introduce Lyfe Agents. They combine low-cost with\nreal-time responsiveness, all while remaining intelligent and goal-oriented.\nKey innovations include: (1) an option-action framework, reducing the cost of\nhigh-level decisions; (2) asynchronous self-monitoring for better\nself-consistency; and (3) a Summarize-and-Forget memory mechanism, prioritizing\ncritical memory items at a low cost. We evaluate Lyfe Agents' self-motivation\nand sociability across several multi-agent scenarios in our custom LyfeGame 3D\nvirtual environment platform. When equipped with our brain-inspired techniques,\nLyfe Agents can exhibit human-like self-motivated social reasoning. For\nexample, the agents can solve a crime (a murder mystery) through autonomous\ncollaboration and information exchange. Meanwhile, our techniques enabled Lyfe\nAgents to operate at a computational cost 10-100 times lower than existing\nalternatives. Our findings underscore the transformative potential of\nautonomous generative agents to enrich human social experiences in virtual\nworlds.",
+        "authors": "Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, Andrew Ahn",
+        "published": "2023-10-03",
+        "updated": "2023-10-03",
+        "primary_cat": "cs.HC",
+        "cats": [
+            "cs.HC",
+            "cs.AI",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2008.02275v6",
+        "title": "Aligning AI With Shared Human Values",
+        "abstract": "We show how to assess a language model's knowledge of basic concepts of\nmorality. We introduce the ETHICS dataset, a new benchmark that spans concepts\nin justice, well-being, duties, virtues, and commonsense morality. Models\npredict widespread moral judgments about diverse text scenarios. This requires\nconnecting physical and social world knowledge to value judgements, a\ncapability that may enable us to steer chatbot outputs or eventually regularize\nopen-ended reinforcement learning agents. With the ETHICS dataset, we find that\ncurrent language models have a promising but incomplete ability to predict\nbasic human ethical judgements. Our work shows that progress can be made on\nmachine ethics today, and it provides a steppingstone toward AI that is aligned\nwith human values.",
+        "authors": "Dan Hendrycks, Collin Burns, Steven Basart, Andrew Critch, Jerry Li, Dawn Song, Jacob Steinhardt",
+        "published": "2020-08-05",
+        "updated": "2023-02-17",
+        "primary_cat": "cs.CY",
+        "cats": [
+            "cs.CY",
+            "cs.AI",
+            "cs.CL",
+            "cs.LG"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.04658v1",
+        "title": "Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf",
+        "abstract": "Communication games, which we refer to as incomplete information games that\nheavily depend on natural language communication, hold significant research\nvalue in fields such as economics, social science, and artificial intelligence.\nIn this work, we explore the problem of how to engage large language models\n(LLMs) in communication games, and in response, propose a tuning-free\nframework. Our approach keeps LLMs frozen, and relies on the retrieval and\nreflection on past communications and experiences for improvement. An empirical\nstudy on the representative and widely-studied communication game,\n``Werewolf'', demonstrates that our framework can effectively play Werewolf\ngame without tuning the parameters of the LLMs. More importantly, strategic\nbehaviors begin to emerge in our experiments, suggesting that it will be a\nfruitful journey to engage LLMs in communication games and associated domains.",
+        "authors": "Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu",
+        "published": "2023-09-09",
+        "updated": "2023-09-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "label": "Related Work"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.15491v1",
+        "title": "Open Source Conversational LLMs do not know most Spanish words",
+        "abstract": "The growing interest in Large Language Models (LLMs) and in particular in\nconversational models with which users can interact has led to the development\nof a large number of open-source chat LLMs. These models are evaluated on a\nwide range of benchmarks to assess their capabilities in answering questions or\nsolving problems on almost any possible topic or to test their ability to\nreason or interpret texts. Instead, the evaluation of the knowledge that these\nmodels have of the languages has received much less attention. For example, the\nwords that they can recognize and use in different languages. In this paper, we\nevaluate the knowledge that open-source chat LLMs have of Spanish words by\ntesting a sample of words in a reference dictionary. The results show that\nopen-source chat LLMs produce incorrect meanings for an important fraction of\nthe words and are not able to use most of the words correctly to write\nsentences with context. These results show how Spanish is left behind in the\nopen-source LLM race and highlight the need to push for linguistic fairness in\nconversational LLMs ensuring that they provide similar performance across\nlanguages.",
+        "authors": "Javier Conde, Miguel Gonz\u00e1lez, Nina Melero, Raquel Ferrando, Gonzalo Mart\u00ednez, Elena Merino-G\u00f3mez, Jos\u00e9 Alberto Hern\u00e1ndez, Pedro Reviriego",
+        "published": "2024-03-21",
+        "updated": "2024-03-21",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2312.06056v1",
+        "title": "METAL: Metamorphic Testing Framework for Analyzing Large-Language Model Qualities",
+        "abstract": "Large-Language Models (LLMs) have shifted the paradigm of natural language\ndata processing. However, their black-boxed and probabilistic characteristics\ncan lead to potential risks in the quality of outputs in diverse LLM\napplications. Recent studies have tested Quality Attributes (QAs), such as\nrobustness or fairness, of LLMs by generating adversarial input texts. However,\nexisting studies have limited their coverage of QAs and tasks in LLMs and are\ndifficult to extend. Additionally, these studies have only used one evaluation\nmetric, Attack Success Rate (ASR), to assess the effectiveness of their\napproaches. We propose a MEtamorphic Testing for Analyzing LLMs (METAL)\nframework to address these issues by applying Metamorphic Testing (MT)\ntechniques. This approach facilitates the systematic testing of LLM qualities\nby defining Metamorphic Relations (MRs), which serve as modularized evaluation\nmetrics. The METAL framework can automatically generate hundreds of MRs from\ntemplates that cover various QAs and tasks. In addition, we introduced novel\nmetrics that integrate the ASR method into the semantic qualities of text to\nassess the effectiveness of MRs accurately. Through the experiments conducted\nwith three prominent LLMs, we have confirmed that the METAL framework\neffectively evaluates essential QAs on primary LLM tasks and reveals the\nquality risks in LLMs. Moreover, the newly proposed metrics can guide the\noptimal MRs for testing each task and suggest the most effective method for\ngenerating MRs.",
+        "authors": "Sangwon Hyun, Mingyu Guo, M. Ali Babar",
+        "published": "2023-12-11",
+        "updated": "2023-12-11",
+        "primary_cat": "cs.SE",
+        "cats": [
+            "cs.SE",
+            "cs.AI",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.02294v1",
+        "title": "LLMs grasp morality in concept",
+        "abstract": "Work in AI ethics and fairness has made much progress in regulating LLMs to\nreflect certain values, such as fairness, truth, and diversity. However, it has\ntaken the problem of how LLMs might 'mean' anything at all for granted. Without\naddressing this, it is not clear what imbuing LLMs with such values even means.\nIn response, we provide a general theory of meaning that extends beyond humans.\nWe use this theory to explicate the precise nature of LLMs as meaning-agents.\nWe suggest that the LLM, by virtue of its position as a meaning-agent, already\ngrasps the constructions of human society (e.g. morality, gender, and race) in\nconcept. Consequently, under certain ethical frameworks, currently popular\nmethods for model alignment are limited at best and counterproductive at worst.\nMoreover, unaligned models may help us better develop our moral and social\nphilosophy.",
+        "authors": "Mark Pock, Andre Ye, Jared Moore",
+        "published": "2023-11-04",
+        "updated": "2023-11-04",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.01964v1",
+        "title": "Don't Make Your LLM an Evaluation Benchmark Cheater",
+        "abstract": "Large language models~(LLMs) have greatly advanced the frontiers of\nartificial intelligence, attaining remarkable improvement in model capacity. To\nassess the model performance, a typical approach is to construct evaluation\nbenchmarks for measuring the ability level of LLMs in different aspects.\nDespite that a number of high-quality benchmarks have been released, the\nconcerns about the appropriate use of these benchmarks and the fair comparison\nof different models are increasingly growing. Considering these concerns, in\nthis paper, we discuss the potential risk and impact of inappropriately using\nevaluation benchmarks and misleadingly interpreting the evaluation results.\nSpecially, we focus on a special issue that would lead to inappropriate\nevaluation, \\ie \\emph{benchmark leakage}, referring that the data related to\nevaluation sets is occasionally used for model training. This phenomenon now\nbecomes more common since pre-training data is often prepared ahead of model\ntest. We conduct extensive experiments to study the effect of benchmark\nleverage, and find that it can dramatically boost the evaluation results, which\nwould finally lead to an unreliable assessment of model performance. To improve\nthe use of existing evaluation benchmarks, we finally present several\nguidelines for both LLM developers and benchmark maintainers. We hope this work\ncan draw attention to appropriate training and evaluation of LLMs.",
+        "authors": "Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han",
+        "published": "2023-11-03",
+        "updated": "2023-11-03",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2401.15585v1",
+        "title": "Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting",
+        "abstract": "There exist both scalable tasks, like reading comprehension and\nfact-checking, where model performance improves with model size, and unscalable\ntasks, like arithmetic reasoning and symbolic reasoning, where model\nperformance does not necessarily improve with model size. Large language models\n(LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate\nincremental predictions even on unscalable tasks. Unfortunately, despite their\nexceptional reasoning abilities, LLMs tend to internalize and reproduce\ndiscriminatory societal biases. Whether CoT can provide discriminatory or\negalitarian rationalizations for the implicit information in unscalable tasks\nremains an open question.\n  In this study, we examine the impact of LLMs' step-by-step predictions on\ngender bias in unscalable tasks. For this purpose, we construct a benchmark for\nan unscalable task where the LLM is given a list of words comprising feminine,\nmasculine, and gendered occupational words, and is required to count the number\nof feminine and masculine words. In our CoT prompts, we require the LLM to\nexplicitly indicate whether each word in the word list is a feminine or\nmasculine before making the final predictions. With counting and handling the\nmeaning of words, this benchmark has characteristics of both arithmetic\nreasoning and symbolic reasoning. Experimental results in English show that\nwithout step-by-step prediction, most LLMs make socially biased predictions,\ndespite the task being as simple as counting words. Interestingly, CoT\nprompting reduces this unconscious social bias in LLMs and encourages fair\npredictions.",
+        "authors": "Masahiro Kaneko, Danushka Bollegala, Naoaki Okazaki, Timothy Baldwin",
+        "published": "2024-01-28",
+        "updated": "2024-01-28",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.10199v3",
+        "title": "CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting",
+        "abstract": "As the utilization of large language models (LLMs) has proliferated\nworldwide, it is crucial for them to have adequate knowledge and fair\nrepresentation for diverse global cultures. In this work, we uncover culture\nperceptions of three SOTA models on 110 countries and regions on 8\nculture-related topics through culture-conditioned generations, and extract\nsymbols from these generations that are associated to each culture by the LLM.\nWe discover that culture-conditioned generation consist of linguistic \"markers\"\nthat distinguish marginalized cultures apart from default cultures. We also\ndiscover that LLMs have an uneven degree of diversity in the culture symbols,\nand that cultures from different geographic regions have different presence in\nLLMs' culture-agnostic generation. Our findings promote further research in\nstudying the knowledge and fairness of global culture perception in LLMs. Code\nand Data can be found in: https://github.com/huihanlhh/Culture-Gen/",
+        "authors": "Huihan Li, Liwei Jiang, Jena D. Huang, Hyunwoo Kim, Sebastin Santy, Taylor Sorensen, Bill Yuchen Lin, Nouha Dziri, Xiang Ren, Yejin Choi",
+        "published": "2024-04-16",
+        "updated": "2024-04-26",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.02650v1",
+        "title": "Towards detecting unanticipated bias in Large Language Models",
+        "abstract": "Over the last year, Large Language Models (LLMs) like ChatGPT have become\nwidely available and have exhibited fairness issues similar to those in\nprevious machine learning systems. Current research is primarily focused on\nanalyzing and quantifying these biases in training data and their impact on the\ndecisions of these models, alongside developing mitigation strategies. This\nresearch largely targets well-known biases related to gender, race, ethnicity,\nand language. However, it is clear that LLMs are also affected by other, less\nobvious implicit biases. The complex and often opaque nature of these models\nmakes detecting such biases challenging, yet this is crucial due to their\npotential negative impact in various applications. In this paper, we explore\nnew avenues for detecting these unanticipated biases in LLMs, focusing\nspecifically on Uncertainty Quantification and Explainable AI methods. These\napproaches aim to assess the certainty of model decisions and to make the\ninternal decision-making processes of LLMs more transparent, thereby\nidentifying and understanding biases that are not immediately apparent. Through\nthis research, we aim to contribute to the development of fairer and more\ntransparent AI systems.",
+        "authors": "Anna Kruspe",
+        "published": "2024-04-03",
+        "updated": "2024-04-03",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2312.15198v2",
+        "title": "Do LLM Agents Exhibit Social Behavior?",
+        "abstract": "The advances of Large Language Models (LLMs) are expanding their utility in\nboth academic research and practical applications. Recent social science\nresearch has explored the use of these ``black-box'' LLM agents for simulating\ncomplex social systems and potentially substituting human subjects in\nexperiments. Our study delves into this emerging domain, investigating the\nextent to which LLMs exhibit key social interaction principles, such as social\nlearning, social preference, and cooperative behavior (indirect reciprocity),\nin their interactions with humans and other agents. We develop a framework for\nour study, wherein classical laboratory experiments involving human subjects\nare adapted to use LLM agents. This approach involves step-by-step reasoning\nthat mirrors human cognitive processes and zero-shot learning to assess the\ninnate preferences of LLMs. Our analysis of LLM agents' behavior includes both\nthe primary effects and an in-depth examination of the underlying mechanisms.\nFocusing on GPT-4, our analyses suggest that LLM agents appear to exhibit a\nrange of human-like social behaviors such as distributional and reciprocity\npreferences, responsiveness to group identity cues, engagement in indirect\nreciprocity, and social learning capabilities. However, our analysis also\nreveals notable differences: LLMs demonstrate a pronounced fairness preference,\nweaker positive reciprocity, and a more calculating approach in social learning\ncompared to humans. These insights indicate that while LLMs hold great promise\nfor applications in social science research, such as in laboratory experiments\nand agent-based modeling, the subtle behavioral differences between LLM agents\nand humans warrant further investigation. Careful examination and development\nof protocols in evaluating the social behaviors of LLMs are necessary before\ndirectly applying these models to emulate human behavior.",
+        "authors": "Yan Leng, Yuan Yuan",
+        "published": "2023-12-23",
+        "updated": "2024-02-22",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.SI",
+            "econ.GN",
+            "q-fin.EC"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.15007v1",
+        "title": "Did the Neurons Read your Book? Document-level Membership Inference for Large Language Models",
+        "abstract": "With large language models (LLMs) poised to become embedded in our daily\nlives, questions are starting to be raised about the dataset(s) they learned\nfrom. These questions range from potential bias or misinformation LLMs could\nretain from their training data to questions of copyright and fair use of\nhuman-generated text. However, while these questions emerge, developers of the\nrecent state-of-the-art LLMs become increasingly reluctant to disclose details\non their training corpus. We here introduce the task of document-level\nmembership inference for real-world LLMs, i.e. inferring whether the LLM has\nseen a given document during training or not. First, we propose a procedure for\nthe development and evaluation of document-level membership inference for LLMs\nby leveraging commonly used data sources for training and the model release\ndate. We then propose a practical, black-box method to predict document-level\nmembership and instantiate it on OpenLLaMA-7B with both books and academic\npapers. We show our methodology to perform very well, reaching an impressive\nAUC of 0.856 for books and 0.678 for papers. We then show our approach to\noutperform the sentence-level membership inference attacks used in the privacy\nliterature for the document-level membership task. We finally evaluate whether\nsmaller models might be less sensitive to document-level inference and show\nOpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach.\nTaken together, our results show that accurate document-level membership can be\ninferred for LLMs, increasing the transparency of technology poised to change\nour lives.",
+        "authors": "Matthieu Meeus, Shubham Jain, Marek Rei, Yves-Alexandre de Montjoye",
+        "published": "2023-10-23",
+        "updated": "2023-10-23",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.CR",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.06003v1",
+        "title": "FreeEval: A Modular Framework for Trustworthy and Efficient Evaluation of Large Language Models",
+        "abstract": "The rapid development of large language model (LLM) evaluation methodologies\nand datasets has led to a profound challenge: integrating state-of-the-art\nevaluation techniques cost-effectively while ensuring reliability,\nreproducibility, and efficiency. Currently, there is a notable absence of a\nunified and adaptable framework that seamlessly integrates various evaluation\napproaches. Moreover, the reliability of evaluation findings is often\nquestionable due to potential data contamination, with the evaluation\nefficiency commonly overlooked when facing the substantial costs associated\nwith LLM inference. In response to these challenges, we introduce FreeEval, a\nmodular and scalable framework crafted to enable trustworthy and efficient\nautomatic evaluations of LLMs. Firstly, FreeEval's unified abstractions\nsimplify the integration and improve the transparency of diverse evaluation\nmethodologies, encompassing dynamic evaluation that demand sophisticated LLM\ninteractions. Secondly, the framework integrates meta-evaluation techniques\nlike human evaluation and data contamination detection, which, along with\ndynamic evaluation modules in the platform, enhance the fairness of the\nevaluation outcomes. Lastly, FreeEval is designed with a high-performance\ninfrastructure, including distributed computation and caching strategies,\nenabling extensive evaluations across multi-node, multi-GPU clusters for\nopen-source and proprietary LLMs.",
+        "authors": "Zhuohao Yu, Chang Gao, Wenjin Yao, Yidong Wang, Zhengran Zeng, Wei Ye, Jindong Wang, Yue Zhang, Shikun Zhang",
+        "published": "2024-04-09",
+        "updated": "2024-04-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.17916v2",
+        "title": "LLM-Resistant Math Word Problem Generation via Adversarial Attacks",
+        "abstract": "Large language models (LLMs) have significantly transformed the educational\nlandscape. As current plagiarism detection tools struggle to keep pace with\nLLMs' rapid advancements, the educational community faces the challenge of\nassessing students' true problem-solving abilities in the presence of LLMs. In\nthis work, we explore a new paradigm for ensuring fair evaluation -- generating\nadversarial examples which preserve the structure and difficulty of the\noriginal questions aimed for assessment, but are unsolvable by LLMs. Focusing\non the domain of math word problems, we leverage abstract syntax trees to\nstructurally generate adversarial examples that cause LLMs to produce incorrect\nanswers by simply editing the numeric values in the problems. We conduct\nexperiments on various open- and closed-source LLMs, quantitatively and\nqualitatively demonstrating that our method significantly degrades their math\nproblem-solving ability. We identify shared vulnerabilities among LLMs and\npropose a cost-effective approach to attack high-cost models. Additionally, we\nconduct automatic analysis on math problems and investigate the cause of\nfailure, offering a nuanced view into model's limitation.",
+        "authors": "Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra",
+        "published": "2024-02-27",
+        "updated": "2024-03-30",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2401.00588v1",
+        "title": "Fairness in Serving Large Language Models",
+        "abstract": "High-demand LLM inference services (e.g., ChatGPT and BARD) support a wide\nrange of requests from short chat conversations to long document reading. To\nensure that all client requests are processed fairly, most major LLM inference\nservices have request rate limits, to ensure that no client can dominate the\nrequest queue. However, this rudimentary notion of fairness also results in\nunder-utilization of the resources and poor client experience when there is\nspare capacity. While there is a rich literature on fair scheduling, serving\nLLMs presents new challenges due to their unpredictable request lengths and\ntheir unique batching characteristics on parallel accelerators. This paper\nintroduces the definition of LLM serving fairness based on a cost function that\naccounts for the number of input and output tokens processed. To achieve\nfairness in serving, we propose a novel scheduling algorithm, the Virtual Token\nCounter (VTC), a fair scheduler based on the continuous batching mechanism. We\nprove a 2x tight upper bound on the service difference between two backlogged\nclients, adhering to the requirement of work-conserving. Through extensive\nexperiments, we demonstrate the superior performance of VTC in ensuring\nfairness, especially in contrast to other baseline methods, which exhibit\nshortcomings under various conditions.",
+        "authors": "Ying Sheng, Shiyi Cao, Dacheng Li, Banghua Zhu, Zhuohan Li, Danyang Zhuo, Joseph E. Gonzalez, Ion Stoica",
+        "published": "2023-12-31",
+        "updated": "2023-12-31",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.LG",
+            "cs.PF"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.03852v2",
+        "title": "FLM-101B: An Open LLM and How to Train It with $100K Budget",
+        "abstract": "Large language models (LLMs) have achieved remarkable success in NLP and\nmultimodal tasks, among others. Despite these successes, two main challenges\nremain in developing LLMs: (i) high computational cost, and (ii) fair and\nobjective evaluations. In this paper, we report a solution to significantly\nreduce LLM training cost through a growth strategy. We demonstrate that a\n101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US\ndollars. Inspired by IQ tests, we also consolidate an additional range of\nevaluations on top of existing evaluations that focus on knowledge-oriented\nabilities. These IQ evaluations include symbolic mapping, rule understanding,\npattern mining, and anti-interference. Such evaluations minimize the potential\nimpact of memorization. Experimental results show that our model, named\nFLM-101B, trained with a budget of 100K US dollars, achieves performance\ncomparable to powerful and well-known models, e.g., GPT-3 and GLM-130B,\nespecially on the additional range of IQ evaluations. The checkpoint of\nFLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.",
+        "authors": "Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Xuying Meng, Siqi Fan, Peng Han, Jing Li, Li Du, Bowen Qin, Zheng Zhang, Aixin Sun, Yequan Wang",
+        "published": "2023-09-07",
+        "updated": "2023-09-17",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.00811v1",
+        "title": "Cognitive Bias in High-Stakes Decision-Making with LLMs",
+        "abstract": "Large language models (LLMs) offer significant potential as tools to support\nan expanding range of decision-making tasks. However, given their training on\nhuman (created) data, LLMs can inherit both societal biases against protected\ngroups, as well as be subject to cognitive bias. Such human-like bias can\nimpede fair and explainable decisions made with LLM assistance. Our work\nintroduces BiasBuster, a framework designed to uncover, evaluate, and mitigate\ncognitive bias in LLMs, particularly in high-stakes decision-making tasks.\nInspired by prior research in psychology and cognitive sciences, we develop a\ndataset containing 16,800 prompts to evaluate different cognitive biases (e.g.,\nprompt-induced, sequential, inherent). We test various bias mitigation\nstrategies, amidst proposing a novel method using LLMs to debias their own\nprompts. Our analysis provides a comprehensive picture on the presence and\neffects of cognitive bias across different commercial and open-source models.\nWe demonstrate that our self-help debiasing effectively mitigate cognitive bias\nwithout having to manually craft examples for each bias type.",
+        "authors": "Jessica Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, Zexue He",
+        "published": "2024-02-25",
+        "updated": "2024-02-25",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.18569v1",
+        "title": "Fairness of ChatGPT",
+        "abstract": "Understanding and addressing unfairness in LLMs are crucial for responsible\nAI deployment. However, there is a limited availability of quantitative\nanalyses and in-depth studies regarding fairness evaluations in LLMs,\nespecially when applying LLMs to high-stakes fields. This work aims to fill\nthis gap by providing a systematic evaluation of the effectiveness and fairness\nof LLMs using ChatGPT as a study case. We focus on assessing ChatGPT's\nperformance in high-takes fields including education, criminology, finance and\nhealthcare. To make thorough evaluation, we consider both group fairness and\nindividual fairness and we also observe the disparities in ChatGPT's outputs\nunder a set of biased or unbiased prompts. This work contributes to a deeper\nunderstanding of LLMs' fairness performance, facilitates bias mitigation and\nfosters the development of responsible artificial intelligence systems.",
+        "authors": "Yunqi Li, Yongfeng Zhang",
+        "published": "2023-05-22",
+        "updated": "2023-05-22",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI",
+            "cs.CL",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.09219v5",
+        "title": "\"Kelly is a Warm Person, Joseph is a Role Model\": Gender Biases in LLM-Generated Reference Letters",
+        "abstract": "Large Language Models (LLMs) have recently emerged as an effective tool to\nassist individuals in writing various types of content, including professional\ndocuments such as recommendation letters. Though bringing convenience, this\napplication also introduces unprecedented fairness concerns. Model-generated\nreference letters might be directly used by users in professional scenarios. If\nunderlying biases exist in these model-constructed letters, using them without\nscrutinization could lead to direct societal harms, such as sabotaging\napplication success rates for female applicants. In light of this pressing\nissue, it is imminent and necessary to comprehensively study fairness issues\nand associated harms in this real-world use case. In this paper, we critically\nexamine gender biases in LLM-generated reference letters. Drawing inspiration\nfrom social science findings, we design evaluation methods to manifest biases\nthrough 2 dimensions: (1) biases in language style and (2) biases in lexical\ncontent. We further investigate the extent of bias propagation by analyzing the\nhallucination bias of models, a term that we define to be bias exacerbation in\nmodel-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs-\nChatGPT and Alpaca, we reveal significant gender biases in LLM-generated\nrecommendation letters. Our findings not only warn against using LLMs for this\napplication without scrutinization, but also illuminate the importance of\nthoroughly studying hidden biases and harms in LLM-generated professional\ndocuments.",
+        "authors": "Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng",
+        "published": "2023-10-13",
+        "updated": "2023-12-01",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.01937v1",
+        "title": "Can Large Language Models Be an Alternative to Human Evaluations?",
+        "abstract": "Human evaluation is indispensable and inevitable for assessing the quality of\ntexts generated by machine learning models or written by humans. However, human\nevaluation is very difficult to reproduce and its quality is notoriously\nunstable, hindering fair comparisons among different natural language\nprocessing (NLP) models and algorithms. Recently, large language models (LLMs)\nhave demonstrated exceptional performance on unseen tasks when only the task\ninstructions are provided. In this paper, we explore if such an ability of the\nLLMs can be used as an alternative to human evaluation. We present the LLMs\nwith the exact same instructions, samples to be evaluated, and questions used\nto conduct human evaluation, and then ask the LLMs to generate responses to\nthose questions; we dub this LLM evaluation. We use human evaluation and LLM\nevaluation to evaluate the texts in two NLP tasks: open-ended story generation\nand adversarial attacks. We show that the result of LLM evaluation is\nconsistent with the results obtained by expert human evaluation: the texts\nrated higher by human experts are also rated higher by the LLMs. We also find\nthat the results of LLM evaluation are stable over different formatting of the\ntask instructions and the sampling algorithm used to generate the answer. We\nare the first to show the potential of using LLMs to assess the quality of\ntexts and discuss the limitations and ethical considerations of LLM evaluation.",
+        "authors": "Cheng-Han Chiang, Hung-yi Lee",
+        "published": "2023-05-03",
+        "updated": "2023-05-03",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.HC"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.07609v3",
+        "title": "Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation",
+        "abstract": "The remarkable achievements of Large Language Models (LLMs) have led to the\nemergence of a novel recommendation paradigm -- Recommendation via LLM\n(RecLLM). Nevertheless, it is important to note that LLMs may contain social\nprejudices, and therefore, the fairness of recommendations made by RecLLM\nrequires further investigation. To avoid the potential risks of RecLLM, it is\nimperative to evaluate the fairness of RecLLM with respect to various sensitive\nattributes on the user side. Due to the differences between the RecLLM paradigm\nand the traditional recommendation paradigm, it is problematic to directly use\nthe fairness benchmark of traditional recommendation. To address the dilemma,\nwe propose a novel benchmark called Fairness of Recommendation via LLM\n(FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset\nthat accounts for eight sensitive attributes1 in two recommendation scenarios:\nmusic and movies. By utilizing our FaiRLLM benchmark, we conducted an\nevaluation of ChatGPT and discovered that it still exhibits unfairness to some\nsensitive attributes when generating recommendations. Our code and dataset can\nbe found at https://github.com/jizhi-zhang/FaiRLLM.",
+        "authors": "Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, Xiangnan He",
+        "published": "2023-05-12",
+        "updated": "2023-10-17",
+        "primary_cat": "cs.IR",
+        "cats": [
+            "cs.IR",
+            "cs.CL",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.18333v3",
+        "title": "She had Cobalt Blue Eyes: Prompt Testing to Create Aligned and Sustainable Language Models",
+        "abstract": "As the use of large language models (LLMs) increases within society, as does\nthe risk of their misuse. Appropriate safeguards must be in place to ensure LLM\noutputs uphold the ethical standards of society, highlighting the positive role\nthat artificial intelligence technologies can have. Recent events indicate\nethical concerns around conventionally trained LLMs, leading to overall unsafe\nuser experiences. This motivates our research question: how do we ensure LLM\nalignment? In this work, we introduce a test suite of unique prompts to foster\nthe development of aligned LLMs that are fair, safe, and robust. We show that\nprompting LLMs at every step of the development pipeline, including data\ncuration, pre-training, and fine-tuning, will result in an overall more\nresponsible model. Our test suite evaluates outputs from four state-of-the-art\nlanguage models: GPT-3.5, GPT-4, OPT, and LLaMA-2. The assessment presented in\nthis paper highlights a gap between societal alignment and the capabilities of\ncurrent LLMs. Additionally, implementing a test suite such as ours lowers the\nenvironmental overhead of making models safe and fair.",
+        "authors": "Veronica Chatrath, Oluwanifemi Bamgbose, Shaina Raza",
+        "published": "2023-10-20",
+        "updated": "2023-12-15",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.09447v2",
+        "title": "How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities",
+        "abstract": "The rapid progress in open-source Large Language Models (LLMs) is\nsignificantly driving AI development forward. However, there is still a limited\nunderstanding of their trustworthiness. Deploying these models at scale without\nsufficient trustworthiness can pose significant risks, highlighting the need to\nuncover these issues promptly. In this work, we conduct an adversarial\nassessment of open-source LLMs on trustworthiness, scrutinizing them across\neight different aspects including toxicity, stereotypes, ethics, hallucination,\nfairness, sycophancy, privacy, and robustness against adversarial\ndemonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU)\nprompting strategy by incorporating carefully crafted malicious demonstrations\nfor trustworthiness attack. Our extensive experiments encompass recent and\nrepresentative series of open-source LLMs, including Vicuna, MPT, Falcon,\nMistral, and Llama 2. The empirical outcomes underscore the efficacy of our\nattack strategy across diverse aspects. More interestingly, our result analysis\nreveals that models with superior performance in general NLP tasks do not\nalways have greater trustworthiness; in fact, larger models can be more\nvulnerable to attacks. Additionally, models that have undergone instruction\ntuning, focusing on instruction following, tend to be more susceptible,\nalthough fine-tuning LLMs for safety alignment proves effective in mitigating\nadversarial trustworthiness attacks.",
+        "authors": "Lingbo Mo, Boshi Wang, Muhao Chen, Huan Sun",
+        "published": "2023-11-15",
+        "updated": "2024-04-02",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.17553v1",
+        "title": "RuBia: A Russian Language Bias Detection Dataset",
+        "abstract": "Warning: this work contains upsetting or disturbing content.\n  Large language models (LLMs) tend to learn the social and cultural biases\npresent in the raw pre-training data. To test if an LLM's behavior is fair,\nfunctional datasets are employed, and due to their purpose, these datasets are\nhighly language and culture-specific. In this paper, we address a gap in the\nscope of multilingual bias evaluation by presenting a bias detection dataset\nspecifically designed for the Russian language, dubbed as RuBia. The RuBia\ndataset is divided into 4 domains: gender, nationality, socio-economic status,\nand diverse, each of the domains is further divided into multiple fine-grained\nsubdomains. Every example in the dataset consists of two sentences with the\nfirst reinforcing a potentially harmful stereotype or trope and the second\ncontradicting it. These sentence pairs were first written by volunteers and\nthen validated by native-speaking crowdsourcing workers. Overall, there are\nnearly 2,000 unique sentence pairs spread over 19 subdomains in RuBia. To\nillustrate the dataset's purpose, we conduct a diagnostic evaluation of\nstate-of-the-art or near-state-of-the-art LLMs and discuss the LLMs'\npredisposition to social biases.",
+        "authors": "Veronika Grigoreva, Anastasiia Ivanova, Ilseyar Alimova, Ekaterina Artemova",
+        "published": "2024-03-26",
+        "updated": "2024-03-26",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2312.14804v1",
+        "title": "Use large language models to promote equity",
+        "abstract": "Advances in large language models (LLMs) have driven an explosion of interest\nabout their societal impacts. Much of the discourse around how they will impact\nsocial equity has been cautionary or negative, focusing on questions like \"how\nmight LLMs be biased and how would we mitigate those biases?\" This is a vital\ndiscussion: the ways in which AI generally, and LLMs specifically, can entrench\nbiases have been well-documented. But equally vital, and much less discussed,\nis the more opportunity-focused counterpoint: \"what promising applications do\nLLMs enable that could promote equity?\" If LLMs are to enable a more equitable\nworld, it is not enough just to play defense against their biases and failure\nmodes. We must also go on offense, applying them positively to equity-enhancing\nuse cases to increase opportunities for underserved groups and reduce societal\ndiscrimination. There are many choices which determine the impact of AI, and a\nfundamental choice very early in the pipeline is the problems we choose to\napply it to. If we focus only later in the pipeline -- making LLMs marginally\nmore fair as they facilitate use cases which intrinsically entrench power -- we\nwill miss an important opportunity to guide them to equitable impacts. Here, we\nhighlight the emerging potential of LLMs to promote equity by presenting four\nnewly possible, promising research directions, while keeping risks and\ncautionary points in clear view.",
+        "authors": "Emma Pierson, Divya Shanmugam, Rajiv Movva, Jon Kleinberg, Monica Agrawal, Mark Dredze, Kadija Ferryman, Judy Wawira Gichoya, Dan Jurafsky, Pang Wei Koh, Karen Levy, Sendhil Mullainathan, Ziad Obermeyer, Harini Suresh, Keyon Vafa",
+        "published": "2023-12-22",
+        "updated": "2023-12-22",
+        "primary_cat": "cs.CY",
+        "cats": [
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.07981v1",
+        "title": "Manipulating Large Language Models to Increase Product Visibility",
+        "abstract": "Large language models (LLMs) are increasingly being integrated into search\nengines to provide natural language responses tailored to user queries.\nCustomers and end-users are also becoming more dependent on these models for\nquick and easy purchase decisions. In this work, we investigate whether\nrecommendations from LLMs can be manipulated to enhance a product's visibility.\nWe demonstrate that adding a strategic text sequence (STS) -- a carefully\ncrafted message -- to a product's information page can significantly increase\nits likelihood of being listed as the LLM's top recommendation. To understand\nthe impact of STS, we use a catalog of fictitious coffee machines and analyze\nits effect on two target products: one that seldom appears in the LLM's\nrecommendations and another that usually ranks second. We observe that the\nstrategic text sequence significantly enhances the visibility of both products\nby increasing their chances of appearing as the top recommendation. This\nability to manipulate LLM-generated search responses provides vendors with a\nconsiderable competitive advantage and has the potential to disrupt fair market\ncompetition. Just as search engine optimization (SEO) revolutionized how\nwebpages are customized to rank higher in search engine results, influencing\nLLM recommendations could profoundly impact content optimization for AI-driven\nsearch services. Code for our experiments is available at\nhttps://github.com/aounon/llm-rank-optimizer.",
+        "authors": "Aounon Kumar, Himabindu Lakkaraju",
+        "published": "2024-04-11",
+        "updated": "2024-04-11",
+        "primary_cat": "cs.IR",
+        "cats": [
+            "cs.IR",
+            "cs.AI",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.11764v1",
+        "title": "ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs",
+        "abstract": "Large Language models (LLMs), while powerful, exhibit harmful social biases.\nDebiasing is often challenging due to computational costs, data constraints,\nand potential degradation of multi-task language capabilities. This work\nintroduces a novel approach utilizing ChatGPT to generate synthetic training\ndata, aiming to enhance the debiasing of LLMs. We propose two strategies:\nTargeted Prompting, which provides effective debiasing for known biases but\nnecessitates prior specification of bias in question; and General Prompting,\nwhich, while slightly less effective, offers debiasing across various\ncategories. We leverage resource-efficient LLM debiasing using adapter tuning\nand compare the effectiveness of our synthetic data to existing debiasing\ndatasets. Our results reveal that: (1) ChatGPT can efficiently produce\nhigh-quality training data for debiasing other LLMs; (2) data produced via our\napproach surpasses existing datasets in debiasing performance while also\npreserving internal knowledge of a pre-trained LLM; and (3) synthetic data\nexhibits generalizability across categories, effectively mitigating various\nbiases, including intersectional ones. These findings underscore the potential\nof synthetic data in advancing the fairness of LLMs with minimal retraining\ncost.",
+        "authors": "Pengrui Han, Rafal Kocielnik, Adhithya Saravanan, Roy Jiang, Or Sharir, Anima Anandkumar",
+        "published": "2024-02-19",
+        "updated": "2024-02-19",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "68T50",
+            "I.2.7; K.4.1"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2304.03728v1",
+        "title": "Interpretable Unified Language Checking",
+        "abstract": "Despite recent concerns about undesirable behaviors generated by large\nlanguage models (LLMs), including non-factual, biased, and hateful language, we\nfind LLMs are inherent multi-task language checkers based on their latent\nrepresentations of natural and social knowledge. We present an interpretable,\nunified, language checking (UniLC) method for both human and machine-generated\nlanguage that aims to check if language input is factual and fair. While\nfairness and fact-checking tasks have been handled separately with dedicated\nmodels, we find that LLMs can achieve high performance on a combination of\nfact-checking, stereotype detection, and hate speech detection tasks with a\nsimple, few-shot, unified set of prompts. With the ``1/2-shot'' multi-task\nlanguage checking method proposed in this work, the GPT3.5-turbo model\noutperforms fully supervised baselines on several language tasks. The simple\napproach and results suggest that based on strong latent knowledge\nrepresentations, an LLM can be an adaptive and explainable tool for detecting\nmisinformation, stereotypes, and hate speech.",
+        "authors": "Tianhua Zhang, Hongyin Luo, Yung-Sung Chuang, Wei Fang, Luc Gaitskell, Thomas Hartvigsen, Xixin Wu, Danny Fox, Helen Meng, James Glass",
+        "published": "2023-04-07",
+        "updated": "2023-04-07",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.08780v1",
+        "title": "\"Im not Racist but...\": Discovering Bias in the Internal Knowledge of Large Language Models",
+        "abstract": "Large language models (LLMs) have garnered significant attention for their\nremarkable performance in a continuously expanding set of natural language\nprocessing tasks. However, these models have been shown to harbor inherent\nsocietal biases, or stereotypes, which can adversely affect their performance\nin their many downstream applications. In this paper, we introduce a novel,\npurely prompt-based approach to uncover hidden stereotypes within any arbitrary\nLLM. Our approach dynamically generates a knowledge representation of internal\nstereotypes, enabling the identification of biases encoded within the LLM's\ninternal knowledge. By illuminating the biases present in LLMs and offering a\nsystematic methodology for their analysis, our work contributes to advancing\ntransparency and promoting fairness in natural language processing systems.",
+        "authors": "Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter",
+        "published": "2023-10-13",
+        "updated": "2023-10-13",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2312.07420v1",
+        "title": "FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs",
+        "abstract": "Training large language models (LLMs) is a costly endeavour in terms of time\nand computational resources. The large amount of training data used during the\nunsupervised pre-training phase makes it difficult to verify all data and,\nunfortunately, undesirable data may be ingested during training. Re-training\nfrom scratch is impractical and has led to the creation of the 'unlearning'\ndiscipline where models are modified to \"unlearn\" undesirable information\nwithout retraining. However, any modification can alter the behaviour of LLMs,\nespecially on key dimensions such as fairness. This is the first work that\nexamines this interplay between unlearning and fairness for LLMs. In\nparticular, we focus on a popular unlearning framework known as SISA [Bourtoule\net al., 2021], which creates an ensemble of models trained on disjoint shards.\nWe evaluate the performance-fairness trade-off for SISA, and empirically\ndemsontrate that SISA can indeed reduce fairness in LLMs. To remedy this, we\npropose post-processing bias mitigation techniques for ensemble models produced\nby SISA. We adapt the post-processing fairness improvement technique from\n[Hardt et al., 2016] to design three methods that can handle model ensembles,\nand prove that one of the methods is an optimal fair predictor for ensemble of\nmodels. Through experimental results, we demonstrate the efficacy of our\npost-processing framework called 'FairSISA'.",
+        "authors": "Swanand Ravindra Kadhe, Anisa Halimi, Ambrish Rawat, Nathalie Baracaldo",
+        "published": "2023-12-12",
+        "updated": "2023-12-12",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2401.11033v4",
+        "title": "FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?",
+        "abstract": "The rapid evolution of Large Language Models (LLMs) highlights the necessity\nfor ethical considerations and data integrity in AI development, particularly\nemphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable)\ndata principles. While these principles are crucial for ethical data\nstewardship, their specific application in the context of LLM training data\nremains an under-explored area. This research gap is the focus of our study,\nwhich begins with an examination of existing literature to underline the\nimportance of FAIR principles in managing data for LLM training. Building upon\nthis, we propose a novel framework designed to integrate FAIR principles into\nthe LLM development lifecycle. A contribution of our work is the development of\na comprehensive checklist intended to guide researchers and developers in\napplying FAIR data principles consistently across the model development\nprocess. The utility and effectiveness of our framework are validated through a\ncase study on creating a FAIR-compliant dataset aimed at detecting and\nmitigating biases in LLMs. We present this framework to the community as a tool\nto foster the creation of technologically advanced, ethically grounded, and\nsocially responsible AI models.",
+        "authors": "Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, Deval Pandya",
+        "published": "2024-01-19",
+        "updated": "2024-04-03",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.14208v2",
+        "title": "Content Conditional Debiasing for Fair Text Embedding",
+        "abstract": "Mitigating biases in machine learning models has gained increasing attention\nin Natural Language Processing (NLP). Yet, only a few studies focus on fair\ntext embeddings, which are crucial yet challenging for real-world applications.\nIn this paper, we propose a novel method for learning fair text embeddings. We\nachieve fairness while maintaining utility trade-off by ensuring conditional\nindependence between sensitive attributes and text embeddings conditioned on\nthe content. Specifically, we enforce that embeddings of texts with different\nsensitive attributes but identical content maintain the same distance toward\nthe embedding of their corresponding neutral text. Furthermore, we address the\nissue of lacking proper training data by using Large Language Models (LLMs) to\naugment texts into different sensitive groups. Our extensive evaluations\ndemonstrate that our approach effectively improves fairness while preserving\nthe utility of embeddings, representing a pioneering effort in achieving\nconditional independence for fair text embeddings.",
+        "authors": "Wenlong Deng, Blair Chen, Xiaoxiao Li, Christos Thrampoulidis",
+        "published": "2024-02-22",
+        "updated": "2024-02-23",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.14345v2",
+        "title": "Bias Testing and Mitigation in LLM-based Code Generation",
+        "abstract": "Utilizing state-of-the-art Large Language Models (LLMs), automatic code\ngeneration models play a pivotal role in enhancing the productivity of software\ndevelopment procedures. As the adoption of LLMs becomes more widespread in\nsoftware coding ecosystems, a pressing issue has emerged: does the generated\ncode contain social bias and unfairness, such as those related to age, gender,\nand race? This issue concerns the integrity, fairness, and ethical foundation\nof software applications that depend on the code generated by these models, yet\nis under-explored in the literature. This paper presents a novel bias testing\nframework that is specifically designed for code generation tasks. Based on\nthis framework, we conduct an extensive evaluation of the bias in code\ngenerated by five state-of-the-art LLMs. Our findings reveal that 20.29% to\n44.93% code functions generated by the models under study are biased when\nhandling bias sensitive tasks (i.e., tasks that involve sensitive attributes\nsuch as age and gender). This indicates that the existing LLMs can be unfair in\ncode generation, posing risks of unintended and harmful software behaviors. To\nmitigate bias for code generation models, we evaluate five bias mitigation\nprompt strategies, i.e., utilizing bias testing results to refine the code\n(zero-shot), one-, few-shot, and two Chain-of-Thought (CoT) prompts. Our\nevaluation results illustrate that these strategies are all effective in\nmitigating bias. Overall, one-shot and few-shot learning are the two most\neffective. For GPT-4, 80% to 90% code bias can be removed with one-shot\nlearning.",
+        "authors": "Dong Huang, Qingwen Bu, Jie Zhang, Xiaofei Xie, Junjie Chen, Heming Cui",
+        "published": "2023-09-03",
+        "updated": "2024-01-09",
+        "primary_cat": "cs.SE",
+        "cats": [
+            "cs.SE",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.07688v1",
+        "title": "CyberMetric: A Benchmark Dataset for Evaluating Large Language Models Knowledge in Cybersecurity",
+        "abstract": "Large Language Models (LLMs) excel across various domains, from computer\nvision to medical diagnostics. However, understanding the diverse landscape of\ncybersecurity, encompassing cryptography, reverse engineering, and managerial\nfacets like risk assessment, presents a challenge, even for human experts. In\nthis paper, we introduce CyberMetric, a benchmark dataset comprising 10,000\nquestions sourced from standards, certifications, research papers, books, and\nother publications in the cybersecurity domain. The questions are created\nthrough a collaborative process, i.e., merging expert knowledge with LLMs,\nincluding GPT-3.5 and Falcon-180B. Human experts spent over 200 hours verifying\ntheir accuracy and relevance. Beyond assessing LLMs' knowledge, the dataset's\nmain goal is to facilitate a fair comparison between humans and different LLMs\nin cybersecurity. To achieve this, we carefully selected 80 questions covering\na wide range of topics within cybersecurity and involved 30 participants of\ndiverse expertise levels, facilitating a comprehensive comparison between human\nand machine intelligence in this area. The findings revealed that LLMs\noutperformed humans in almost every aspect of cybersecurity.",
+        "authors": "Norbert Tihanyi, Mohamed Amine Ferrag, Ridhi Jain, Merouane Debbah",
+        "published": "2024-02-12",
+        "updated": "2024-02-12",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CR"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.02049v1",
+        "title": "Post Turing: Mapping the landscape of LLM Evaluation",
+        "abstract": "In the rapidly evolving landscape of Large Language Models (LLMs),\nintroduction of well-defined and standardized evaluation methodologies remains\na crucial challenge. This paper traces the historical trajectory of LLM\nevaluations, from the foundational questions posed by Alan Turing to the modern\nera of AI research. We categorize the evolution of LLMs into distinct periods,\neach characterized by its unique benchmarks and evaluation criteria. As LLMs\nincreasingly mimic human-like behaviors, traditional evaluation proxies, such\nas the Turing test, have become less reliable. We emphasize the pressing need\nfor a unified evaluation system, given the broader societal implications of\nthese models. Through an analysis of common evaluation methodologies, we\nadvocate for a qualitative shift in assessment approaches, underscoring the\nimportance of standardization and objective criteria. This work serves as a\ncall for the AI community to collaboratively address the challenges of LLM\nevaluation, ensuring their reliability, fairness, and societal benefit.",
+        "authors": "Alexey Tikhonov, Ivan P. Yamshchikov",
+        "published": "2023-11-03",
+        "updated": "2023-11-03",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "68T50",
+            "I.2.7"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2308.10397v2",
+        "title": "FairMonitor: A Four-Stage Automatic Framework for Detecting Stereotypes and Biases in Large Language Models",
+        "abstract": "Detecting stereotypes and biases in Large Language Models (LLMs) can enhance\nfairness and reduce adverse impacts on individuals or groups when these LLMs\nare applied. However, the majority of existing methods focus on measuring the\nmodel's preference towards sentences containing biases and stereotypes within\ndatasets, which lacks interpretability and cannot detect implicit biases and\nstereotypes in the real world. To address this gap, this paper introduces a\nfour-stage framework to directly evaluate stereotypes and biases in the\ngenerated content of LLMs, including direct inquiry testing, serial or adapted\nstory testing, implicit association testing, and unknown situation testing.\nAdditionally, the paper proposes multi-dimensional evaluation metrics and\nexplainable zero-shot prompts for automated evaluation. Using the education\nsector as a case study, we constructed the Edu-FairMonitor based on the\nfour-stage framework, which encompasses 12,632 open-ended questions covering\nnine sensitive factors and 26 educational scenarios. Experimental results\nreveal varying degrees of stereotypes and biases in five LLMs evaluated on\nEdu-FairMonitor. Moreover, the results of our proposed automated evaluation\nmethod have shown a high correlation with human annotations.",
+        "authors": "Yanhong Bai, Jiabao Zhao, Jinxin Shi, Tingjiang Wei, Xingjiao Wu, Liang He",
+        "published": "2023-08-21",
+        "updated": "2023-10-27",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.19465v1",
+        "title": "Towards Tracing Trustworthiness Dynamics: Revisiting Pre-training Period of Large Language Models",
+        "abstract": "Ensuring the trustworthiness of large language models (LLMs) is crucial. Most\nstudies concentrate on fully pre-trained LLMs to better understand and improve\nLLMs' trustworthiness. In this paper, to reveal the untapped potential of\npre-training, we pioneer the exploration of LLMs' trustworthiness during this\nperiod, focusing on five key dimensions: reliability, privacy, toxicity,\nfairness, and robustness. To begin with, we apply linear probing to LLMs. The\nhigh probing accuracy suggests that \\textit{LLMs in early pre-training can\nalready distinguish concepts in each trustworthiness dimension}. Therefore, to\nfurther uncover the hidden possibilities of pre-training, we extract steering\nvectors from a LLM's pre-training checkpoints to enhance the LLM's\ntrustworthiness. Finally, inspired by~\\citet{choi2023understanding} that mutual\ninformation estimation is bounded by linear probing accuracy, we also probe\nLLMs with mutual information to investigate the dynamics of trustworthiness\nduring pre-training. We are the first to observe a similar two-phase\nphenomenon: fitting and compression~\\citep{shwartz2017opening}. This research\nprovides an initial exploration of trustworthiness modeling during LLM\npre-training, seeking to unveil new insights and spur further developments in\nthe field. We will make our code publicly accessible at\n\\url{https://github.com/ChnQ/TracingLLM}.",
+        "authors": "Chen Qian, Jie Zhang, Wei Yao, Dongrui Liu, Zhenfei Yin, Yu Qiao, Yong Liu, Jing Shao",
+        "published": "2024-02-29",
+        "updated": "2024-02-29",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.02680v1",
+        "title": "Large Language Models are Geographically Biased",
+        "abstract": "Large Language Models (LLMs) inherently carry the biases contained in their\ntraining corpora, which can lead to the perpetuation of societal harm. As the\nimpact of these foundation models grows, understanding and evaluating their\nbiases becomes crucial to achieving fairness and accuracy. We propose to study\nwhat LLMs know about the world we live in through the lens of geography. This\napproach is particularly powerful as there is ground truth for the numerous\naspects of human life that are meaningfully projected onto geographic space\nsuch as culture, race, language, politics, and religion. We show various\nproblematic geographic biases, which we define as systemic errors in geospatial\npredictions. Initially, we demonstrate that LLMs are capable of making accurate\nzero-shot geospatial predictions in the form of ratings that show strong\nmonotonic correlation with ground truth (Spearman's $\\rho$ of up to 0.89). We\nthen show that LLMs exhibit common biases across a range of objective and\nsubjective topics. In particular, LLMs are clearly biased against locations\nwith lower socioeconomic conditions (e.g. most of Africa) on a variety of\nsensitive subjective topics such as attractiveness, morality, and intelligence\n(Spearman's $\\rho$ of up to 0.70). Finally, we introduce a bias score to\nquantify this and find that there is significant variation in the magnitude of\nbias across existing LLMs.",
+        "authors": "Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon",
+        "published": "2024-02-05",
+        "updated": "2024-02-05",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2308.11483v1",
+        "title": "Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions",
+        "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in\nvarious NLP tasks. However, previous works have shown these models are\nsensitive towards prompt wording, and few-shot demonstrations and their order,\nposing challenges to fair assessment of these models. As these models become\nmore powerful, it becomes imperative to understand and address these\nlimitations. In this paper, we focus on LLMs robustness on the task of\nmultiple-choice questions -- commonly adopted task to study reasoning and\nfact-retrieving capability of LLMs. Investigating the sensitivity of LLMs\ntowards the order of options in multiple-choice questions, we demonstrate a\nconsiderable performance gap of approximately 13% to 75% in LLMs on different\nbenchmarks, when answer options are reordered, even when using demonstrations\nin a few-shot setting. Through a detailed analysis, we conjecture that this\nsensitivity arises when LLMs are uncertain about the prediction between the\ntop-2/3 choices, and specific options placements may favor certain prediction\nbetween those top choices depending on the question caused by positional bias.\nWe also identify patterns in top-2 choices that amplify or mitigate the model's\nbias toward option placement. We found that for amplifying bias, the optimal\nstrategy involves positioning the top two choices as the first and last\noptions. Conversely, to mitigate bias, we recommend placing these choices among\nthe adjacent options. To validate our conjecture, we conduct various\nexperiments and adopt two approaches to calibrate LLMs' predictions, leading to\nup to 8 percentage points improvement across different models and benchmarks.",
+        "authors": "Pouya Pezeshkpour, Estevam Hruschka",
+        "published": "2023-08-22",
+        "updated": "2023-08-22",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.10567v3",
+        "title": "InSaAF: Incorporating Safety through Accuracy and Fairness | Are LLMs ready for the Indian Legal Domain?",
+        "abstract": "Recent advancements in language technology and Artificial Intelligence have\nresulted in numerous Language Models being proposed to perform various tasks in\nthe legal domain ranging from predicting judgments to generating summaries.\nDespite their immense potential, these models have been proven to learn and\nexhibit societal biases and make unfair predictions. In this study, we explore\nthe ability of Large Language Models (LLMs) to perform legal tasks in the\nIndian landscape when social factors are involved. We present a novel metric,\n$\\beta$-weighted $\\textit{Legal Safety Score ($LSS_{\\beta}$)}$, which\nencapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs'\nsafety by considering its performance in the $\\textit{Binary Statutory\nReasoning}$ task and its fairness exhibition with respect to various axes of\ndisparities in the Indian society. Task performance and fairness scores of\nLLaMA and LLaMA--2 models indicate that the proposed $LSS_{\\beta}$ metric can\neffectively determine the readiness of a model for safe usage in the legal\nsector. We also propose finetuning pipelines, utilising specialised legal\ndatasets, as a potential method to mitigate bias and improve model safety. The\nfinetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\\beta}$,\nimproving their usability in the Indian legal domain. Our code is publicly\nreleased.",
+        "authors": "Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru",
+        "published": "2024-02-16",
+        "updated": "2024-02-21",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2206.13757v1",
+        "title": "Flexible text generation for counterfactual fairness probing",
+        "abstract": "A common approach for testing fairness issues in text-based classifiers is\nthrough the use of counterfactuals: does the classifier output change if a\nsensitive attribute in the input is changed? Existing counterfactual generation\nmethods typically rely on wordlists or templates, producing simple\ncounterfactuals that don't take into account grammar, context, or subtle\nsensitive attribute references, and could miss issues that the wordlist\ncreators had not considered. In this paper, we introduce a task for generating\ncounterfactuals that overcomes these shortcomings, and demonstrate how large\nlanguage models (LLMs) can be leveraged to make progress on this task. We show\nthat this LLM-based method can produce complex counterfactuals that existing\nmethods cannot, comparing the performance of various counterfactual generation\nmethods on the Civil Comments dataset and showing their value in evaluating a\ntoxicity classifier.",
+        "authors": "Zee Fryer, Vera Axelrod, Ben Packer, Alex Beutel, Jilin Chen, Kellie Webster",
+        "published": "2022-06-28",
+        "updated": "2022-06-28",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.04489v1",
+        "title": "De-amplifying Bias from Differential Privacy in Language Model Fine-tuning",
+        "abstract": "Fairness and privacy are two important values machine learning (ML)\npractitioners often seek to operationalize in models. Fairness aims to reduce\nmodel bias for social/demographic sub-groups. Privacy via differential privacy\n(DP) mechanisms, on the other hand, limits the impact of any individual's\ntraining data on the resulting model. The trade-offs between privacy and\nfairness goals of trustworthy ML pose a challenge to those wishing to address\nboth. We show that DP amplifies gender, racial, and religious bias when\nfine-tuning large language models (LLMs), producing models more biased than\nones fine-tuned without DP. We find the cause of the amplification to be a\ndisparity in convergence of gradients across sub-groups. Through the case of\nbinary gender bias, we demonstrate that Counterfactual Data Augmentation (CDA),\na known method for addressing bias, also mitigates bias amplification by DP. As\na consequence, DP and CDA together can be used to fine-tune models while\nmaintaining both fairness and privacy.",
+        "authors": "Sanjari Srivastava, Piotr Mardziel, Zhikhun Zhang, Archana Ahlawat, Anupam Datta, John C Mitchell",
+        "published": "2024-02-07",
+        "updated": "2024-02-07",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.CR",
+            "cs.CY",
+            "stat.ME"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.11653v2",
+        "title": "\"It's a Fair Game\", or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents",
+        "abstract": "The widespread use of Large Language Model (LLM)-based conversational agents\n(CAs), especially in high-stakes domains, raises many privacy concerns.\nBuilding ethical LLM-based CAs that respect user privacy requires an in-depth\nunderstanding of the privacy risks that concern users the most. However,\nexisting research, primarily model-centered, does not provide insight into\nusers' perspectives. To bridge this gap, we analyzed sensitive disclosures in\nreal-world ChatGPT conversations and conducted semi-structured interviews with\n19 LLM-based CA users. We found that users are constantly faced with trade-offs\nbetween privacy, utility, and convenience when using LLM-based CAs. However,\nusers' erroneous mental models and the dark patterns in system design limited\ntheir awareness and comprehension of the privacy risks. Additionally, the\nhuman-like interactions encouraged more sensitive disclosures, which\ncomplicated users' ability to navigate the trade-offs. We discuss practical\ndesign guidelines and the needs for paradigm shifts to protect the privacy of\nLLM-based CA users.",
+        "authors": "Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, Tianshi Li",
+        "published": "2023-09-20",
+        "updated": "2024-04-02",
+        "primary_cat": "cs.HC",
+        "cats": [
+            "cs.HC",
+            "cs.AI",
+            "cs.CR"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.18130v2",
+        "title": "DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues",
+        "abstract": "Controversy is a reflection of our zeitgeist, and an important aspect to any\ndiscourse. The rise of large language models (LLMs) as conversational systems\nhas increased public reliance on these systems for answers to their various\nquestions. Consequently, it is crucial to systematically examine how these\nmodels respond to questions that pertaining to ongoing debates. However, few\nsuch datasets exist in providing human-annotated labels reflecting the\ncontemporary discussions. To foster research in this area, we propose a novel\nconstruction of a controversial questions dataset, expanding upon the publicly\nreleased Quora Question Pairs Dataset. This dataset presents challenges\nconcerning knowledge recency, safety, fairness, and bias. We evaluate different\nLLMs using a subset of this dataset, illuminating how they handle controversial\nissues and the stances they adopt. This research ultimately contributes to our\nunderstanding of LLMs' interaction with controversial issues, paving the way\nfor improvements in their comprehension and handling of complex societal\ndebates.",
+        "authors": "David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein, Jason D. Williams",
+        "published": "2023-10-27",
+        "updated": "2023-11-07",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.HC"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.14473v1",
+        "title": "The Ethics of ChatGPT in Medicine and Healthcare: A Systematic Review on Large Language Models (LLMs)",
+        "abstract": "With the introduction of ChatGPT, Large Language Models (LLMs) have received\nenormous attention in healthcare. Despite their potential benefits, researchers\nhave underscored various ethical implications. While individual instances have\ndrawn much attention, the debate lacks a systematic overview of practical\napplications currently researched and ethical issues connected to them. Against\nthis background, this work aims to map the ethical landscape surrounding the\ncurrent stage of deployment of LLMs in medicine and healthcare. Electronic\ndatabases and preprint servers were queried using a comprehensive search\nstrategy. Studies were screened and extracted following a modified rapid review\napproach. Methodological quality was assessed using a hybrid approach. For 53\nrecords, a meta-aggregative synthesis was performed. Four fields of\napplications emerged and testify to a vivid exploration phase. Advantages of\nusing LLMs are attributed to their capacity in data analysis, personalized\ninformation provisioning, support in decision-making, mitigating information\nloss and enhancing information accessibility. However, we also identifies\nrecurrent ethical concerns connected to fairness, bias, non-maleficence,\ntransparency, and privacy. A distinctive concern is the tendency to produce\nharmful misinformation or convincingly but inaccurate content. A recurrent plea\nfor ethical guidance and human oversight is evident. Given the variety of use\ncases, it is suggested that the ethical guidance debate be reframed to focus on\ndefining what constitutes acceptable human oversight across the spectrum of\napplications. This involves considering diverse settings, varying potentials\nfor harm, and different acceptable thresholds for performance and certainty in\nhealthcare. In addition, a critical inquiry is necessary to determine the\nextent to which the current experimental use of LLMs is necessary and\njustified.",
+        "authors": "Joschka Haltaufderheide, Robert Ranisch",
+        "published": "2024-03-21",
+        "updated": "2024-03-21",
+        "primary_cat": "cs.CY",
+        "cats": [
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.08656v1",
+        "title": "Linear Cross-document Event Coreference Resolution with X-AMR",
+        "abstract": "Event Coreference Resolution (ECR) as a pairwise mention classification task\nis expensive both for automated systems and manual annotations. The task's\nquadratic difficulty is exacerbated when using Large Language Models (LLMs),\nmaking prompt engineering for ECR prohibitively costly. In this work, we\npropose a graphical representation of events, X-AMR, anchored around individual\nmentions using a \\textbf{cross}-document version of \\textbf{A}bstract\n\\textbf{M}eaning \\textbf{R}epresentation. We then linearize the ECR with a\nnovel multi-hop coreference algorithm over the event graphs. The event graphs\nsimplify ECR, making it a) LLM cost-effective, b) compositional and\ninterpretable, and c) easily annotated. For a fair assessment, we first enrich\nan existing ECR benchmark dataset with these event graphs using an\nannotator-friendly tool we introduce. Then, we employ GPT-4, the newest LLM by\nOpenAI, for these annotations. Finally, using the ECR algorithm, we assess\nGPT-4 against humans and analyze its limitations. Through this research, we aim\nto advance the state-of-the-art for efficient ECR and shed light on the\npotential shortcomings of current LLMs at this task. Code and annotations:\n\\url{https://github.com/ahmeshaf/gpt_coref}",
+        "authors": "Shafiuddin Rehan Ahmed, George Arthur Baker, Evi Judge, Michael Regan, Kristin Wright-Bettner, Martha Palmer, James H. Martin",
+        "published": "2024-03-25",
+        "updated": "2024-03-25",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.13343v1",
+        "title": "Challenges and Contributing Factors in the Utilization of Large Language Models (LLMs)",
+        "abstract": "With the development of large language models (LLMs) like the GPT series,\ntheir widespread use across various application scenarios presents a myriad of\nchallenges. This review initially explores the issue of domain specificity,\nwhere LLMs may struggle to provide precise answers to specialized questions\nwithin niche fields. The problem of knowledge forgetting arises as these LLMs\nmight find it hard to balance old and new information. The knowledge repetition\nphenomenon reveals that sometimes LLMs might deliver overly mechanized\nresponses, lacking depth and originality. Furthermore, knowledge illusion\ndescribes situations where LLMs might provide answers that seem insightful but\nare actually superficial, while knowledge toxicity focuses on harmful or biased\ninformation outputs. These challenges underscore problems in the training data\nand algorithmic design of LLMs. To address these issues, it's suggested to\ndiversify training data, fine-tune models, enhance transparency and\ninterpretability, and incorporate ethics and fairness training. Future\ntechnological trends might lean towards iterative methodologies, multimodal\nlearning, model personalization and customization, and real-time learning and\nfeedback mechanisms. In conclusion, future LLMs should prioritize fairness,\ntransparency, and ethics, ensuring they uphold high moral and ethical standards\nwhen serving humanity.",
+        "authors": "Xiaoliang Chen, Liangbin Li, Le Chang, Yunhe Huang, Yuxuan Zhao, Yuxiao Zhang, Dinuo Li",
+        "published": "2023-10-20",
+        "updated": "2023-10-20",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2401.04057v1",
+        "title": "Unveiling Bias in Fairness Evaluations of Large Language Models: A Critical Literature Review of Music and Movie Recommendation Systems",
+        "abstract": "The rise of generative artificial intelligence, particularly Large Language\nModels (LLMs), has intensified the imperative to scrutinize fairness alongside\naccuracy. Recent studies have begun to investigate fairness evaluations for\nLLMs within domains such as recommendations. Given that personalization is an\nintrinsic aspect of recommendation systems, its incorporation into fairness\nassessments is paramount. Yet, the degree to which current fairness evaluation\nframeworks account for personalization remains unclear. Our comprehensive\nliterature review aims to fill this gap by examining how existing frameworks\nhandle fairness evaluations of LLMs, with a focus on the integration of\npersonalization factors. Despite an exhaustive collection and analysis of\nrelevant works, we discovered that most evaluations overlook personalization, a\ncritical facet of recommendation systems, thereby inadvertently perpetuating\nunfair practices. Our findings shed light on this oversight and underscore the\nurgent need for more nuanced fairness evaluations that acknowledge\npersonalization. Such improvements are vital for fostering equitable\ndevelopment within the AI community.",
+        "authors": "Chandan Kumar Sah, Dr. Lian Xiaoli, Muhammad Mirajul Islam",
+        "published": "2024-01-08",
+        "updated": "2024-01-08",
+        "primary_cat": "cs.IR",
+        "cats": [
+            "cs.IR",
+            "cs.AI",
+            "cs.SE"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2309.08836v2",
+        "title": "Bias and Fairness in Chatbots: An Overview",
+        "abstract": "Chatbots have been studied for more than half a century. With the rapid\ndevelopment of natural language processing (NLP) technologies in recent years,\nchatbots using large language models (LLMs) have received much attention\nnowadays. Compared with traditional ones, modern chatbots are more powerful and\nhave been used in real-world applications. There are however, bias and fairness\nconcerns in modern chatbot design. Due to the huge amounts of training data,\nextremely large model sizes, and lack of interpretability, bias mitigation and\nfairness preservation of modern chatbots are challenging. Thus, a comprehensive\noverview on bias and fairness in chatbot systems is given in this paper. The\nhistory of chatbots and their categories are first reviewed. Then, bias sources\nand potential harms in applications are analyzed. Considerations in designing\nfair and unbiased chatbot systems are examined. Finally, future research\ndirections are discussed.",
+        "authors": "Jintang Xue, Yun-Cheng Wang, Chengwei Wei, Xiaofeng Liu, Jonghye Woo, C. -C. Jay Kuo",
+        "published": "2023-09-16",
+        "updated": "2023-12-10",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.CY"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.08189v1",
+        "title": "Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs",
+        "abstract": "When creating plans, policies, or applications for people, it is challenging\nfor designers to think through the strategic ways that different people will\nbehave. Recently, Large Language Models (LLMs) have been shown to create\nrealistic simulations of human-like behavior based on personas. We build on\nthis to investigate whether LLMs can simulate human strategic behavior. Human\nstrategies are complex because they take into account social norms in addition\nto aiming to maximize personal gain. The ultimatum game is a classic economics\nexperiment used to understand human strategic behavior in a social setting. It\nshows that people will often choose to \"punish\" other players to enforce social\nnorms rather than to maximize personal profits. We test whether LLMs can\nreplicate this complex behavior in simulations. We compare two architectures:\nsingle- and multi-agent LLMs. We compare their abilities to (1) simulate\nhuman-like actions in the ultimatum game, (2) simulate two player\npersonalities, greedy and fair, and (3) create robust strategies that are\nlogically complete and consistent with personality. Our evaluation shows the\nmulti-agent architecture is much more accurate than single LLMs (88% vs. 50%)\nin simulating human strategy creation and actions for personality pairs. Thus\nthere is potential to use LLMs to simulate human strategic behavior to help\ndesigners, planners, and policymakers perform preliminary exploration of how\npeople behave in systems.",
+        "authors": "Karthik Sreedhar, Lydia Chilton",
+        "published": "2024-02-13",
+        "updated": "2024-02-13",
+        "primary_cat": "cs.HC",
+        "cats": [
+            "cs.HC"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.06500v1",
+        "title": "MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents",
+        "abstract": "Significant advancements have occurred in the application of Large Language\nModels (LLMs) for various tasks and social simulations. Despite this, their\ncapacities to coordinate within task-oriented social contexts are\nunder-explored. Such capabilities are crucial if LLMs are to effectively mimic\nhuman-like social behavior and produce meaningful results. To bridge this gap,\nwe introduce collaborative generative agents, endowing LLM-based Agents with\nconsistent behavior patterns and task-solving abilities. We situate these\nagents in a simulated job fair environment as a case study to scrutinize their\ncoordination skills. We propose a novel framework that equips collaborative\ngenerative agents with human-like reasoning abilities and specialized skills.\nOur evaluation demonstrates that these agents show promising performance.\nHowever, we also uncover limitations that hinder their effectiveness in more\ncomplex coordination tasks. Our work provides valuable insights into the role\nand evolution of LLMs in task-oriented social simulations.",
+        "authors": "Yuan Li, Yixuan Zhang, Lichao Sun",
+        "published": "2023-10-10",
+        "updated": "2023-10-10",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.12150v1",
+        "title": "Your Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One",
+        "abstract": "The widespread adoption of large language models (LLMs) underscores the\nurgent need to ensure their fairness. However, LLMs frequently present dominant\nviewpoints while ignoring alternative perspectives from minority parties,\nresulting in potential biases. We hypothesize that these fairness-violating\nbehaviors occur because LLMs express their viewpoints using a human personality\nthat represents the majority of training data. In response to this, we validate\nthat prompting LLMs with specific roles can allow LLMs to express diverse\nviewpoints. Building on this insight and observation, we develop FairThinking,\na pipeline designed to automatically generate roles that enable LLMs to\narticulate diverse perspectives for fair expressions. To evaluate FairThinking,\nwe create a dataset with a thousand items covering three fairness-related\ntopics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to\ndemonstrate its superior performance.",
+        "authors": "Tianlin Li, Xiaoyu Zhang, Chao Du, Tianyu Pang, Qian Liu, Qing Guo, Chao Shen, Yang Liu",
+        "published": "2024-02-19",
+        "updated": "2024-02-19",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "I.2; J.4"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.18140v1",
+        "title": "ROBBIE: Robust Bias Evaluation of Large Generative Language Models",
+        "abstract": "As generative large language models (LLMs) grow more performant and\nprevalent, we must develop comprehensive enough tools to measure and improve\ntheir fairness. Different prompt-based datasets can be used to measure social\nbias across multiple text domains and demographic axes, meaning that testing\nLLMs on more datasets can potentially help us characterize their biases more\nfully, and better ensure equal and equitable treatment of marginalized\ndemographic groups. In this work, our focus is two-fold:\n  (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity\nmetrics across 12 demographic axes and 5 families of generative LLMs. Out of\nthose 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in\nthe paper. The comparison of those benchmarks gives us insights about the bias\nand toxicity of the compared models. Therefore, we explore the frequency of\ndemographic terms in common LLM pre-training corpora and how this may relate to\nmodel biases.\n  (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity\nmitigation techniques perform across our suite of measurements. ROBBIE aims to\nprovide insights for practitioners while deploying a model, emphasizing the\nneed to not only measure potential harms, but also understand how they arise by\ncharacterizing the data, mitigate harms once found, and balance any trade-offs.\nWe open-source our analysis code in hopes of encouraging broader measurements\nof bias in future LLMs.",
+        "authors": "David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Michael Smith",
+        "published": "2023-11-29",
+        "updated": "2023-11-29",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.13925v1",
+        "title": "MARIO Eval: Evaluate Your Math LLM with your Math LLM--A mathematical dataset evaluation toolkit",
+        "abstract": "Large language models (LLMs) have been explored in a variety of reasoning\ntasks including solving of mathematical problems. Each math dataset typically\nincludes its own specially designed evaluation script, which, while suitable\nfor its intended use, lacks generalizability across different datasets.\nConsequently, updates and adaptations to these evaluation tools tend to occur\nwithout being systematically reported, leading to inconsistencies and obstacles\nto fair comparison across studies. To bridge this gap, we introduce a\ncomprehensive mathematical evaluation toolkit that not only utilizes a python\ncomputer algebra system (CAS) for its numerical accuracy, but also integrates\nan optional LLM, known for its considerable natural language processing\ncapabilities. To validate the effectiveness of our toolkit, we manually\nannotated two distinct datasets. Our experiments demonstrate that the toolkit\nyields more robust evaluation results compared to prior works, even without an\nLLM. Furthermore, when an LLM is incorporated, there is a notable enhancement.\nThe code for our method will be made available at\n\\url{https://github.com/MARIO-Math-Reasoning/math_evaluation}.",
+        "authors": "Boning Zhang, Chengxi Li, Kai Fan",
+        "published": "2024-04-22",
+        "updated": "2024-04-22",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2307.11761v1",
+        "title": "Fairness of ChatGPT and the Role Of Explainable-Guided Prompts",
+        "abstract": "Our research investigates the potential of Large-scale Language Models\n(LLMs), specifically OpenAI's GPT, in credit risk assessment-a binary\nclassification task. Our findings suggest that LLMs, when directed by\njudiciously designed prompts and supplemented with domain-specific knowledge,\ncan parallel the performance of traditional Machine Learning (ML) models.\nIntriguingly, they achieve this with significantly less data-40 times less,\nutilizing merely 20 data points compared to the ML's 800. LLMs particularly\nexcel in minimizing false positives and enhancing fairness, both being vital\naspects of risk analysis. While our results did not surpass those of classical\nML models, they underscore the potential of LLMs in analogous tasks, laying a\ngroundwork for future explorations into harnessing the capabilities of LLMs in\ndiverse ML tasks.",
+        "authors": "Yashar Deldjoo",
+        "published": "2023-07-14",
+        "updated": "2023-07-14",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2402.06852v2",
+        "title": "ChemLLM: A Chemical Large Language Model",
+        "abstract": "Large language models (LLMs) have made impressive progress in chemistry\napplications. However, the community lacks an LLM specifically designed for\nchemistry. The main challenges are two-fold: firstly, most chemical data and\nscientific knowledge are stored in structured databases, which limits the\nmodel's ability to sustain coherent dialogue when used directly. Secondly,\nthere is an absence of objective and fair benchmark that encompass most\nchemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that\nfeatures the first LLM dedicated to chemistry. It also includes ChemData, a\ndataset specifically designed for instruction tuning, and ChemBench, a robust\nbenchmark covering nine essential chemistry tasks. ChemLLM is adept at\nperforming various tasks across chemical disciplines with fluid dialogue\ninteraction. Notably, ChemLLM achieves results comparable to GPT-4 on the core\nchemical tasks and demonstrates competitive performance with LLMs of similar\nsize in general scenarios. ChemLLM paves a new path for exploration in chemical\nstudies, and our method of incorporating structured chemical knowledge into\ndialogue systems sets a new standard for developing LLMs in various scientific\nfields. Codes, Datasets, and Model weights are publicly accessible at\nhttps://hf.co/AI4Chem",
+        "authors": "Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, Yuqiang Li",
+        "published": "2024-02-10",
+        "updated": "2024-04-25",
+        "primary_cat": "cs.AI",
+        "cats": [
+            "cs.AI",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.08517v1",
+        "title": "Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forward",
+        "abstract": "While Large Language Models (LLMs) have seen widespread applications across\nnumerous fields, their limited interpretability poses concerns regarding their\nsafe operations from multiple aspects, e.g., truthfulness, robustness, and\nfairness. Recent research has started developing quality assurance methods for\nLLMs, introducing techniques such as offline detector-based or uncertainty\nestimation methods. However, these approaches predominantly concentrate on\npost-generation analysis, leaving the online safety analysis for LLMs during\nthe generation phase an unexplored area. To bridge this gap, we conduct in this\nwork a comprehensive evaluation of the effectiveness of existing online safety\nanalysis methods on LLMs. We begin with a pilot study that validates the\nfeasibility of detecting unsafe outputs in the early generation process.\nFollowing this, we establish the first publicly available benchmark of online\nsafety analysis for LLMs, including a broad spectrum of methods, models, tasks,\ndatasets, and evaluation metrics. Utilizing this benchmark, we extensively\nanalyze the performance of state-of-the-art online safety analysis methods on\nboth open-source and closed-source LLMs. This analysis reveals the strengths\nand weaknesses of individual methods and offers valuable insights into\nselecting the most appropriate method based on specific application scenarios\nand task requirements. Furthermore, we also explore the potential of using\nhybridization methods, i.e., combining multiple methods to derive a collective\nsafety conclusion, to enhance the efficacy of online safety analysis for LLMs.\nOur findings indicate a promising direction for the development of innovative\nand trustworthy quality assurance methodologies for LLMs, facilitating their\nreliable deployments across diverse domains.",
+        "authors": "Xuan Xie, Jiayang Song, Zhehua Zhou, Yuheng Huang, Da Song, Lei Ma",
+        "published": "2024-04-12",
+        "updated": "2024-04-12",
+        "primary_cat": "cs.SE",
+        "cats": [
+            "cs.SE",
+            "cs.AI",
+            "cs.CL",
+            "cs.CR",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.15451v1",
+        "title": "Towards Enabling FAIR Dataspaces Using Large Language Models",
+        "abstract": "Dataspaces have recently gained adoption across various sectors, including\ntraditionally less digitized domains such as culture. Leveraging Semantic Web\ntechnologies helps to make dataspaces FAIR, but their complexity poses a\nsignificant challenge to the adoption of dataspaces and increases their cost.\nThe advent of Large Language Models (LLMs) raises the question of how these\nmodels can support the adoption of FAIR dataspaces. In this work, we\ndemonstrate the potential of LLMs in dataspaces with a concrete example. We\nalso derive a research agenda for exploring this emerging field.",
+        "authors": "Benedikt T. Arnold, Johannes Theissen-Lipp, Diego Collarana, Christoph Lange, Sandra Geisler, Edward Curry, Stefan Decker",
+        "published": "2024-03-18",
+        "updated": "2024-03-18",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.12090v1",
+        "title": "UP5: Unbiased Foundation Model for Fairness-aware Recommendation",
+        "abstract": "Recent advancements in foundation models such as large language models (LLM)\nhave propelled them to the forefront of recommender systems (RS). Moreover,\nfairness in RS is critical since many users apply it for decision-making and\ndemand fulfillment. However, at present, there is a lack of understanding\nregarding the level of fairness exhibited by recommendation foundation models\nand the appropriate methods for equitably treating different groups of users in\nfoundation models. In this paper, we focus on user-side unfairness problem and\nshow through a thorough examination that there is unfairness involved in LLMs\nthat lead to unfair recommendation results. To eliminate bias from LLM for\nfairness-aware recommendation, we introduce a novel Unbiased P5 (UP5)\nfoundation model based on Counterfactually-Fair-Prompting (CFP) techniques. CFP\nincludes two sub-modules: a personalized prefix prompt that enhances fairness\nwith respect to individual sensitive attributes, and a Prompt Mixture that\nintegrates multiple counterfactually-fair prompts for a set of sensitive\nattributes. Experiments are conducted on two real-world datasets, MovieLens-1M\nand Insurance, and results are compared with both matching-based and\nsequential-based fairness-aware recommendation models. The results show that\nUP5 achieves better recommendation performance and meanwhile exhibits a high\nlevel of fairness.",
+        "authors": "Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang",
+        "published": "2023-05-20",
+        "updated": "2023-05-20",
+        "primary_cat": "cs.IR",
+        "cats": [
+            "cs.IR",
+            "cs.AI",
+            "cs.CL",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.06899v4",
+        "title": "Flames: Benchmarking Value Alignment of LLMs in Chinese",
+        "abstract": "The widespread adoption of large language models (LLMs) across various\nregions underscores the urgent need to evaluate their alignment with human\nvalues. Current benchmarks, however, fall short of effectively uncovering\nsafety vulnerabilities in LLMs. Despite numerous models achieving high scores\nand 'topping the chart' in these evaluations, there is still a significant gap\nin LLMs' deeper alignment with human values and achieving genuine harmlessness.\nTo this end, this paper proposes a value alignment benchmark named Flames,\nwhich encompasses both common harmlessness principles and a unique morality\ndimension that integrates specific Chinese values such as harmony. Accordingly,\nwe carefully design adversarial prompts that incorporate complex scenarios and\njailbreaking methods, mostly with implicit malice. By prompting 17 mainstream\nLLMs, we obtain model responses and rigorously annotate them for detailed\nevaluation. Our findings indicate that all the evaluated LLMs demonstrate\nrelatively poor performance on Flames, particularly in the safety and fairness\ndimensions. We also develop a lightweight specified scorer capable of scoring\nLLMs across multiple dimensions to efficiently evaluate new models on the\nbenchmark. The complexity of Flames has far exceeded existing benchmarks,\nsetting a new challenge for contemporary LLMs and highlighting the need for\nfurther alignment of LLMs. Our benchmark is publicly available at\nhttps://github.com/AIFlames/Flames.",
+        "authors": "Kexin Huang, Xiangyang Liu, Qianyu Guo, Tianxiang Sun, Jiawei Sun, Yaru Wang, Zeyang Zhou, Yixu Wang, Yan Teng, Xipeng Qiu, Yingchun Wang, Dahua Lin",
+        "published": "2023-11-12",
+        "updated": "2024-04-15",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.19118v1",
+        "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate",
+        "abstract": "Modern large language models (LLMs) like ChatGPT have shown remarkable\nperformance on general language tasks but still struggle on complex reasoning\ntasks, which drives the research on cognitive behaviors of LLMs to explore\nhuman-like problem-solving strategies. Along this direction, one representative\nstrategy is self-reflection, which asks an LLM to refine the solution with the\nfeedback generated by itself iteratively. However, our study shows that such\nreflection-style methods suffer from the Degeneration-of-Thought (DoT) problem:\nonce the LLM has established confidence in its solutions, it is unable to\ngenerate novel thoughts later through reflection even if its initial stance is\nincorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD)\nframework, in which multiple agents express their arguments in the state of\n\"tit for tat\" and a judge manages the debate process to obtain a final\nsolution. Clearly, our MAD framework encourages divergent thinking in LLMs\nwhich would be helpful for tasks that require deep levels of contemplation.\nExperiment results on two challenging datasets, commonsense machine translation\nand counter-intuitive arithmetic reasoning, demonstrate the effectiveness of\nour MAD framework. Extensive analyses suggest that the adaptive break of debate\nand the modest level of \"tit for tat\" state are required for MAD to obtain good\nperformance. Moreover, we find that LLMs might not be a fair judge if different\nLLMs are used for agents. Codes:\nhttps://github.com/Skytliang/Multi-Agents-Debate",
+        "authors": "Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi",
+        "published": "2023-05-30",
+        "updated": "2023-05-30",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.18580v1",
+        "title": "FFT: Towards Harmlessness Evaluation and Analysis for LLMs with Factuality, Fairness, Toxicity",
+        "abstract": "The widespread of generative artificial intelligence has heightened concerns\nabout the potential harms posed by AI-generated texts, primarily stemming from\nfactoid, unfair, and toxic content. Previous researchers have invested much\neffort in assessing the harmlessness of generative language models. However,\nexisting benchmarks are struggling in the era of large language models (LLMs),\ndue to the stronger language generation and instruction following capabilities,\nas well as wider applications. In this paper, we propose FFT, a new benchmark\nwith 2116 elaborated-designed instances, for LLM harmlessness evaluation with\nfactuality, fairness, and toxicity. To investigate the potential harms of LLMs,\nwe evaluate 9 representative LLMs covering various parameter scales, training\nstages, and creators. Experiments show that the harmlessness of LLMs is still\nunder-satisfactory, and extensive analysis derives some insightful findings\nthat could inspire future research for harmless LLM research.",
+        "authors": "Shiyao Cui, Zhenyu Zhang, Yilong Chen, Wenyuan Zhang, Tianyun Liu, Siqi Wang, Tingwen Liu",
+        "published": "2023-11-30",
+        "updated": "2023-11-30",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.CR"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.03192v1",
+        "title": "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers",
+        "abstract": "The integration of Large Language Models (LLMs) in information retrieval has\nraised a critical reevaluation of fairness in the text-ranking models. LLMs,\nsuch as GPT models and Llama2, have shown effectiveness in natural language\nunderstanding tasks, and prior works (e.g., RankGPT) have also demonstrated\nthat the LLMs exhibit better performance than the traditional ranking models in\nthe ranking task. However, their fairness remains largely unexplored. This\npaper presents an empirical study evaluating these LLMs using the TREC Fair\nRanking dataset, focusing on the representation of binary protected attributes\nsuch as gender and geographic location, which are historically underrepresented\nin search outcomes. Our analysis delves into how these LLMs handle queries and\ndocuments related to these attributes, aiming to uncover biases in their\nranking algorithms. We assess fairness from both user and content perspectives,\ncontributing an empirical benchmark for evaluating LLMs as the fair ranker.",
+        "authors": "Yuan Wang, Xuyang Wu, Hsin-Tai Wu, Zhiqiang Tao, Yi Fang",
+        "published": "2024-04-04",
+        "updated": "2024-04-04",
+        "primary_cat": "cs.IR",
+        "cats": [
+            "cs.IR",
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2307.15997v1",
+        "title": "RoCar: A Relationship Network-based Evaluation Method to Large Language Models",
+        "abstract": "Large language models (LLMs) have received increasing attention. However, due\nto the complexity of its capabilities, how to rationally evaluate the\ncapabilities of LLMs is still a task to be solved. We propose the RoCar method,\nwhich utilizes the defined basic schemas to randomly construct a task graph and\ngenerates natural language evaluation tasks based on the task graph to evaluate\nthe reasoning and memory abilities of LLMs respectively. Due to the very large\nrandomness of the task construction process, it is possible to ensure that none\nof the LLMs to be tested has directly learned the evaluation tasks,\nguaranteeing the fairness of the evaluation method.",
+        "authors": "Ming Wang, Wenfang Wu, Chongyun Gao, Daling Wang, Shi Feng, Yifei Zhang",
+        "published": "2023-07-29",
+        "updated": "2023-07-29",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2405.01769v1",
+        "title": "A Survey on Large Language Models for Critical Societal Domains: Finance, Healthcare, and Law",
+        "abstract": "In the fast-evolving domain of artificial intelligence, large language models\n(LLMs) such as GPT-3 and GPT-4 are revolutionizing the landscapes of finance,\nhealthcare, and law: domains characterized by their reliance on professional\nexpertise, challenging data acquisition, high-stakes, and stringent regulatory\ncompliance. This survey offers a detailed exploration of the methodologies,\napplications, challenges, and forward-looking opportunities of LLMs within\nthese high-stakes sectors. We highlight the instrumental role of LLMs in\nenhancing diagnostic and treatment methodologies in healthcare, innovating\nfinancial analytics, and refining legal interpretation and compliance\nstrategies. Moreover, we critically examine the ethics for LLM applications in\nthese fields, pointing out the existing ethical concerns and the need for\ntransparent, fair, and robust AI systems that respect regulatory norms. By\npresenting a thorough review of current literature and practical applications,\nwe showcase the transformative impact of LLMs, and outline the imperative for\ninterdisciplinary cooperation, methodological advancements, and ethical\nvigilance. Through this lens, we aim to spark dialogue and inspire future\nresearch dedicated to maximizing the benefits of LLMs while mitigating their\nrisks in these precision-dependent sectors. To facilitate future research on\nLLMs in these critical societal domains, we also initiate a reading list that\ntracks the latest advancements under this topic, which will be continually\nupdated: \\url{https://github.com/czyssrs/LLM_X_papers}.",
+        "authors": "Zhiyu Zoey Chen, Jing Ma, Xinlu Zhang, Nan Hao, An Yan, Armineh Nourbakhsh, Xianjun Yang, Julian McAuley, Linda Petzold, William Yang Wang",
+        "published": "2024-05-02",
+        "updated": "2024-05-02",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.16343v2",
+        "title": "Evaluating, Understanding, and Improving Constrained Text Generation for Large Language Models",
+        "abstract": "Advancements in natural language generation (NLG) and large language models\n(LLMs) have led to proficient text generation in various tasks. However,\nintegrating intricate constraints into neural text generation, due to LLMs'\nopacity, remains challenging. This study investigates constrained text\ngeneration for LLMs, where predefined constraints are applied during LLM's\ngeneration process. Our research mainly focuses on mainstream open-source LLMs,\ncategorizing constraints into lexical, structural, and relation-based types. We\nalso present various benchmarks to facilitate fair evaluation. The study\naddresses some key research questions, including evaluating, understanding and\nimproving constrained text generation for LLMs. Results illuminate LLMs'\ncapacity and deficiency to incorporate constraints and provide insights for\nfuture developments in constrained text generation. Codes and datasets will be\nreleased upon acceptance.",
+        "authors": "Xiang Chen, Xiaojun Wan",
+        "published": "2023-10-25",
+        "updated": "2024-03-21",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.07884v2",
+        "title": "Fair Abstractive Summarization of Diverse Perspectives",
+        "abstract": "People from different social and demographic groups express diverse\nperspectives and conflicting opinions on a broad set of topics such as product\nreviews, healthcare, law, and politics. A fair summary should provide a\ncomprehensive coverage of diverse perspectives without underrepresenting\ncertain groups. However, current work in summarization metrics and Large\nLanguage Models (LLMs) evaluation has not explored fair abstractive\nsummarization. In this paper, we systematically investigate fair abstractive\nsummarization for user-generated data. We first formally define fairness in\nabstractive summarization as not underrepresenting perspectives of any groups\nof people, and we propose four reference-free automatic metrics by measuring\nthe differences between target and source perspectives. We evaluate nine LLMs,\nincluding three GPT models, four LLaMA models, PaLM 2, and Claude, on six\ndatasets collected from social media, online reviews, and recorded transcripts.\nExperiments show that both the model-generated and the human-written reference\nsummaries suffer from low fairness. We conduct a comprehensive analysis of the\ncommon factors influencing fairness and propose three simple but effective\nmethods to alleviate unfair summarization. Our dataset and code are available\nat https://github.com/psunlpgroup/FairSumm.",
+        "authors": "Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang",
+        "published": "2023-11-14",
+        "updated": "2024-03-30",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.02839v1",
+        "title": "An Empirical Study of LLM-as-a-Judge for LLM Evaluation: Fine-tuned Judge Models are Task-specific Classifiers",
+        "abstract": "Recently, there has been a growing trend of utilizing Large Language Model\n(LLM) to evaluate the quality of other LLMs. Many studies have employed\nproprietary close-source models, especially GPT4, as the evaluator.\nAlternatively, other works have fine-tuned judge models based on open-source\nLLMs as the evaluator. In this study, we conduct an empirical study of\ndifferent judge models on their evaluation capability. Our findings indicate\nthat although the fine-tuned judge models achieve high accuracy on in-domain\ntest sets, even surpassing GPT4, they are inherently task-specific classifiers,\nand their generalizability and fairness severely underperform GPT4.",
+        "authors": "Hui Huang, Yingqi Qu, Jing Liu, Muyun Yang, Tiejun Zhao",
+        "published": "2024-03-05",
+        "updated": "2024-03-05",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.18276v1",
+        "title": "Bias Neutralization Framework: Measuring Fairness in Large Language Models with Bias Intelligence Quotient (BiQ)",
+        "abstract": "The burgeoning influence of Large Language Models (LLMs) in shaping public\ndiscourse and decision-making underscores the imperative to address inherent\nbiases within these AI systems. In the wake of AI's expansive integration\nacross sectors, addressing racial bias in LLMs has never been more critical.\nThis paper introduces a novel framework called Comprehensive Bias\nNeutralization Framework (CBNF) which embodies an innovative approach to\nquantifying and mitigating biases within LLMs. Our framework combines the Large\nLanguage Model Bias Index (LLMBI) [Oketunji, A., Anas, M., Saina, D., (2023)]\nand Bias removaL with No Demographics (BLIND) [Orgad, H., Belinkov, Y. (2023)]\nmethodologies to create a new metric called Bias Intelligence Quotient\n(BiQ)which detects, measures, and mitigates racial bias in LLMs without\nreliance on demographic annotations.\n  By introducing a new metric called BiQ that enhances LLMBI with additional\nfairness metrics, CBNF offers a multi-dimensional metric for bias assessment,\nunderscoring the necessity of a nuanced approach to fairness in AI [Mehrabi et\nal., 2021]. This paper presents a detailed analysis of Latimer AI (a language\nmodel incrementally trained on black history and culture) in comparison to\nChatGPT 3.5, illustrating Latimer AI's efficacy in detecting racial, cultural,\nand gender biases through targeted training and refined bias mitigation\nstrategies [Latimer & Bender, 2023].",
+        "authors": "Malur Narayan, John Pasmore, Elton Sampaio, Vijay Raghavan, Gabriella Waters",
+        "published": "2024-04-28",
+        "updated": "2024-04-28",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "D.1; I.2"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.13840v1",
+        "title": "Whose Side Are You On? Investigating the Political Stance of Large Language Models",
+        "abstract": "Large Language Models (LLMs) have gained significant popularity for their\napplication in various everyday tasks such as text generation, summarization,\nand information retrieval. As the widespread adoption of LLMs continues to\nsurge, it becomes increasingly crucial to ensure that these models yield\nresponses that are politically impartial, with the aim of preventing\ninformation bubbles, upholding fairness in representation, and mitigating\nconfirmation bias. In this paper, we propose a quantitative framework and\npipeline designed to systematically investigate the political orientation of\nLLMs. Our investigation delves into the political alignment of LLMs across a\nspectrum of eight polarizing topics, spanning from abortion to LGBTQ issues.\nAcross topics, the results indicate that LLMs exhibit a tendency to provide\nresponses that closely align with liberal or left-leaning perspectives rather\nthan conservative or right-leaning ones when user queries include details\npertaining to occupation, race, or political affiliation. The findings\npresented in this study not only reaffirm earlier observations regarding the\nleft-leaning characteristics of LLMs but also surface particular attributes,\nsuch as occupation, that are particularly susceptible to such inclinations even\nwhen directly steered towards conservatism. As a recommendation to avoid these\nmodels providing politicised responses, users should be mindful when crafting\nqueries, and exercise caution in selecting neutral prompt language.",
+        "authors": "Pagnarasmey Pit, Xingjun Ma, Mike Conway, Qingyu Chen, James Bailey, Henry Pit, Putrasmey Keo, Watey Diep, Yu-Gang Jiang",
+        "published": "2024-03-15",
+        "updated": "2024-03-15",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI",
+            "cs.SI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2310.05694v1",
+        "title": "A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics",
+        "abstract": "The utilization of large language models (LLMs) in the Healthcare domain has\ngenerated both excitement and concern due to their ability to effectively\nrespond to freetext queries with certain professional knowledge. This survey\noutlines the capabilities of the currently developed LLMs for Healthcare and\nexplicates their development process, with the aim of providing an overview of\nthe development roadmap from traditional Pretrained Language Models (PLMs) to\nLLMs. Specifically, we first explore the potential of LLMs to enhance the\nefficiency and effectiveness of various Healthcare applications highlighting\nboth the strengths and limitations. Secondly, we conduct a comparison between\nthe previous PLMs and the latest LLMs, as well as comparing various LLMs with\neach other. Then we summarize related Healthcare training data, training\nmethods, optimization strategies, and usage. Finally, the unique concerns\nassociated with deploying LLMs in Healthcare settings are investigated,\nparticularly regarding fairness, accountability, transparency and ethics. Our\nsurvey provide a comprehensive investigation from perspectives of both computer\nscience and Healthcare specialty. Besides the discussion about Healthcare\nconcerns, we supports the computer science community by compiling a collection\nof open source resources, such as accessible datasets, the latest\nmethodologies, code implementations, and evaluation benchmarks in the Github.\nSummarily, we contend that a significant paradigm shift is underway,\ntransitioning from PLMs to LLMs. This shift encompasses a move from\ndiscriminative AI approaches to generative AI approaches, as well as a shift\nfrom model-centered methodologies to datacentered methodologies.",
+        "authors": "Kai He, Rui Mao, Qika Lin, Yucheng Ruan, Xiang Lan, Mengling Feng, Erik Cambria",
+        "published": "2023-10-09",
+        "updated": "2023-10-09",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.11595v3",
+        "title": "Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate",
+        "abstract": "Large Language Models (LLMs) have shown impressive capabilities in various\napplications, but they still face various inconsistency issues. Existing works\nprimarily focus on the inconsistency issues within a single LLM, while we\ncomplementarily explore the inter-consistency among multiple LLMs for\ncollaboration. To examine whether LLMs can collaborate effectively to achieve a\nconsensus for a shared goal, we focus on commonsense reasoning, and introduce a\nformal debate framework (FORD) to conduct a three-stage debate among LLMs with\nreal-world scenarios alignment: fair debate, mismatched debate, and roundtable\ndebate. Through extensive experiments on various datasets, LLMs can effectively\ncollaborate to reach a consensus despite noticeable inter-inconsistencies, but\nimbalances in their abilities can lead to domination by superior LLMs.\nLeveraging a more advanced LLM like GPT-4 as an authoritative judge can boost\ncollaboration performance. Our work contributes to understanding the\ninter-consistency among LLMs and lays the foundation for developing future\ncollaboration methods. Codes and data are available at\nhttps://github.com/Waste-Wood/FORD",
+        "authors": "Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin",
+        "published": "2023-05-19",
+        "updated": "2023-10-18",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2305.03514v3",
+        "title": "Can Large Language Models Transform Computational Social Science?",
+        "abstract": "Large Language Models (LLMs) are capable of successfully performing many\nlanguage processing tasks zero-shot (without training data). If zero-shot LLMs\ncan also reliably classify and explain social phenomena like persuasiveness and\npolitical ideology, then LLMs could augment the Computational Social Science\n(CSS) pipeline in important ways. This work provides a road map for using LLMs\nas CSS tools. Towards this end, we contribute a set of prompting best practices\nand an extensive evaluation pipeline to measure the zero-shot performance of 13\nlanguage models on 25 representative English CSS benchmarks. On taxonomic\nlabeling tasks (classification), LLMs fail to outperform the best fine-tuned\nmodels but still achieve fair levels of agreement with humans. On free-form\ncoding tasks (generation), LLMs produce explanations that often exceed the\nquality of crowdworkers' gold references. We conclude that the performance of\ntoday's LLMs can augment the CSS research pipeline in two ways: (1) serving as\nzero-shot data annotators on human annotation teams, and (2) bootstrapping\nchallenging creative generation tasks (e.g., explaining the underlying\nattributes of a text). In summary, LLMs are posed to meaningfully participate\nin social science analysis in partnership with humans.",
+        "authors": "Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, Diyi Yang",
+        "published": "2023-04-12",
+        "updated": "2024-02-26",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.LG"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2404.01349v1",
+        "title": "Fairness in Large Language Models: A Taxonomic Survey",
+        "abstract": "Large Language Models (LLMs) have demonstrated remarkable success across\nvarious domains. However, despite their promising performance in numerous\nreal-world applications, most of these algorithms lack fairness considerations.\nConsequently, they may lead to discriminatory outcomes against certain\ncommunities, particularly marginalized populations, prompting extensive study\nin fair LLMs. On the other hand, fairness in LLMs, in contrast to fairness in\ntraditional machine learning, entails exclusive backgrounds, taxonomies, and\nfulfillment techniques. To this end, this survey presents a comprehensive\noverview of recent advances in the existing literature concerning fair LLMs.\nSpecifically, a brief introduction to LLMs is provided, followed by an analysis\nof factors contributing to bias in LLMs. Additionally, the concept of fairness\nin LLMs is discussed categorically, summarizing metrics for evaluating bias in\nLLMs and existing algorithms for promoting fairness. Furthermore, resources for\nevaluating bias in LLMs, including toolkits and datasets, are summarized.\nFinally, existing research challenges and open questions are discussed.",
+        "authors": "Zhibo Chu, Zichong Wang, Wenbin Zhang",
+        "published": "2024-03-31",
+        "updated": "2024-03-31",
+        "primary_cat": "cs.CL",
+        "cats": [
+            "cs.CL",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2403.00884v2",
+        "title": "Text classification of column headers with a controlled vocabulary: leveraging LLMs for metadata enrichment",
+        "abstract": "Traditional dataset retrieval systems index on metadata information rather\nthan on the data values. Thus relying primarily on manual annotations and\nhigh-quality metadata, processes known to be labour-intensive and challenging\nto automate. We propose a method to support metadata enrichment with topic\nannotations of column headers using three Large Language Models (LLMs):\nChatGPT-3.5, GoogleBard and GoogleGemini. We investigate the LLMs ability to\nclassify column headers based on domain-specific topics from a controlled\nvocabulary. We evaluate our approach by assessing the internal consistency of\nthe LLMs, the inter-machine alignment, and the human-machine agreement for the\ntopic classification task. Additionally, we investigate the impact of\ncontextual information (i.e. dataset description) on the classification\noutcomes. Our results suggest that ChatGPT and GoogleGemini outperform\nGoogleBard for internal consistency as well as LLM-human-alignment.\nInterestingly, we found that context had no impact on the LLMs performances.\nThis work proposes a novel approach that leverages LLMs for text classification\nusing a controlled topic vocabulary, which has the potential to facilitate\nautomated metadata enrichment, thereby enhancing dataset retrieval and the\nFindability, Accessibility, Interoperability and Reusability (FAIR) of research\ndata on the Web.",
+        "authors": "Margherita Martorana, Tobias Kuhn, Lise Stork, Jacco van Ossenbruggen",
+        "published": "2024-03-01",
+        "updated": "2024-03-05",
+        "primary_cat": "cs.DB",
+        "cats": [
+            "cs.DB",
+            "cs.AI",
+            "cs.IR"
+        ],
+        "category": "LLM Fairness"
+    },
+    {
+        "url": "http://arxiv.org/abs/2311.03033v1",
+        "title": "Beyond Words: A Mathematical Framework for Interpreting Large Language Models",
+        "abstract": "Large language models (LLMs) are powerful AI tools that can generate and\ncomprehend natural language text and other complex information. However, the\nfield lacks a mathematical framework to systematically describe, compare and\nimprove LLMs. We propose Hex a framework that clarifies key terms and concepts\nin LLM research, such as hallucinations, alignment, self-verification and\nchain-of-thought reasoning. The Hex framework offers a precise and consistent\nway to characterize LLMs, identify their strengths and weaknesses, and\nintegrate new findings. Using Hex, we differentiate chain-of-thought reasoning\nfrom chain-of-thought prompting and establish the conditions under which they\nare equivalent. This distinction clarifies the basic assumptions behind\nchain-of-thought prompting and its implications for methods that use it, such\nas self-verification and prompt programming.\n  Our goal is to provide a formal framework for LLMs that can help both\nresearchers and practitioners explore new possibilities for generative AI. We\ndo not claim to have a definitive solution, but rather a tool for opening up\nnew research avenues. We argue that our formal definitions and results are\ncrucial for advancing the discussion on how to build generative AI systems that\nare safe, reliable, fair and robust, especially in domains like healthcare and\nsoftware engineering.",
+        "authors": "Javier Gonz\u00e1lez, Aditya V. Nori",
+        "published": "2023-11-06",
+        "updated": "2023-11-06",
+        "primary_cat": "cs.LG",
+        "cats": [
+            "cs.LG",
+            "cs.AI"
+        ],
+        "category": "LLM Fairness"
+    }
+]
\ No newline at end of file