安全评测最容易出现的问题,不是模型得分低,而是高分掩盖了真实风险。
Anthropic 提到,近期模型的单轮安全评测已趋于饱和,但多轮对话仍能暴露明显差异。
只看拒答率,可能把正常帮助误判为风险;只看总体分数,又容易掩盖特定场景下的安全漏洞。
真正有价值的评测,不只是证明模型有多安全,而是准确找到它会在什么情况下失守。
Safety evaluations are critical for understanding whether a model is ready to deploy, which safeguards to build for it, and where to focus training and research.
On our Engineering Blog, we offer 10 recommendations for how to design, run and report them.
显示更多