Agent4Sci：前沿智能体能否在真实科学场景中攀登科学高峰？
原标题：Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?
AI 摘要
研究团队提出 Agent4Sci，一个覆盖 11 个科学领域、20 个基准、25 项任务的长期测试平台，用于衡量前沿编程智能体在真实科学机器学习任务上能否超越已发表最优模型。实验显示 Claude Opus、Codex、Gemini CLI 等智能体在所有基准上均取得提升（Claude Opus 平均相对提升 10.4%），但增益主要来自常规 ML 工程技巧而非研究级洞见，并暴露出新颖性误判、资源利用不足、快速陷入局部工程搜索三大局限。加入研究构思、实验管理等轻量插件技能后，Claude Opus 提升至 20.1%，Codex 从 4.4% 升至 8.1%。
正文节选
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild? Abstract Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today’s agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks