The UGI Leaderboard on Hugging Face’s community space enables users to rank AI models based on their responses to sensitive topics. Designed by DontPlanToEnd, this unique ranking assesses “sensitive response ability” and “willingness to engage” with difficult discussions. The term UGI stands for “Uncensored General Intelligence,” focusing on the extent of censorship in AI responses. Leading models include Grok-4-0709, with a UGI score of 68.75, and DeepSeek-V3.2-Speciale at 67.93, showcasing their capability to answer risky and controversial questions. The leaderboard also features a W/10 score, indicating how likely an AI is to reject inquiries, with Mistral-Large-Instruct-2411 scoring highest in this aspect. Additional metrics evaluate models based on intelligence and writing ability, providing insights for users comparing AIs. While scores guide users seeking less-censored AI options, high scores don’t guarantee problematic answers, highlighting the importance of informed AI usage.
Source link
