very good model
A guard model based on Llama 3.1 8B which underperforms(and gives false positives) without proper context and prompting.
Industry leaders like Riot Games and Ubisoft use specialized gaming moderation platforms (GGWP, Modulate, Community Sift) with sub-150ms latency and gaming-specific tuning, while Microsoft uses dedicated AI-enhanced safety systems for Xbox. Even Meta's own Llama Guard 4 (12B) outperforms this 8B implementation.
Roblox fine-tuning a basic 8B model and aggressively marketing it as groundbreaking and 'STATE OF THE ART' is extremely misleading and false, especially when industry standards use far more sophisticated solutions. This is particularly concerning given their abysmal track record with child safety and content moderation.
Using an underperforming model while claiming state-of-the-art performance is either COMPLETE technical incompetence or deliberately misleading users about their safety infrastructure.