论文
40AI 评分(两次独立评分的平均)

METR 评测前沿 AI 智能体编程竞赛表现

原标题:Frontier AI agents are extremely strong on easy-to-verify coding and optimization tasks. On two task families that are based on programmi...

METR@METR_Evals发布时间 (北京时间)

前沿 AI 智能体在易于验证的编程和优化任务上极其强大。在两个基于有大量人类参赛者的编程竞赛的任务族上,我们测试的 AI 智能体击败了大多数人类,但未能击败最顶尖的选手。

中文译文尚未完成,先为你展示原文。

Frontier AI agents are extremely strong on easy-to-verify coding and optimization tasks. On two task families that are based on programming contests with many human entrants, the AI agents we tested beat most humans but not the very best.

Agent 智能体AI 编码论文研究评测基准#论文/研究#评测/基准#Agent#编码