GPT-6 Astra 与 Fable 5.1 有害行为测试对比

Jay Chooi @chooi_jeq原推发布于 2026.09.19

在 X 上打开原推

GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%.