Dan Hendrycks：可解释性工具脆弱性新发现
More findings that interpretability tools are fragile or worse than simple baselines.
AI 导读
更多发现表明，可解释性工具很脆弱，甚至不如简单基线。
查看原帖A good explanation of a model's behavior should help you make predictions in related situations. We turn this into an eval, with thousands of real behaviors found in the wild. Can interp tools help here? On average, no. 🧵 https://t.co/yN3t8G4n40
正文
当前显示原文，中文全文译文尚未就绪。
More findings that interpretability tools are fragile or worse than simple baselines.