论文
38AI 评分(两次独立评分的平均)

METR:AI智能体难验证任务判断力逊于专家

原标题:AI agents are improving even on our hardest-to-verify tasks, but appear to have worse judgment than human experts on tasks without clear,...

METR@METR_Evals发布时间 (北京时间)

AI智能体即使在我们最难验证的任务上也在进步,但在缺乏清晰、快速反馈的任务上,其判断力似乎不如人类专家。 例如,智能体现在能为我们构建勉强有用的评估任务,尽管并非我们认为特别有信息量的那些。

中文译文尚未完成,先为你展示原文。

AI agents are improving even on our hardest-to-verify tasks, but appear to have worse judgment than human experts on tasks without clear, fast feedback.

For example, agents can now build borderline-useful evaluation tasks for us, though not ones we consider especially informative.

Agent 智能体论文研究评测基准#论文/研究#Agent#评测/基准