arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.07591cs.LGcs.AIcs.CL
收藏

ResearchClawBench: 端到端自主科学研究基准

ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research

Shanghai Artificial Intelligence Laboratory(上海人工智能实验室)
Wanghan Xu, Shuo Li, Tianlin Ye, Qinglong Cao, Yixin Chen, Hengjian Gao, Yiheng Wang, Qi Li, Kun Li, Sheng Xu, Shengdu Chai, Fangchen Yu, Xiangyu Zhao, Zhangrui… 展开作者

Wanghan Xu, Shuo Li, Tianlin Ye, Qinglong Cao, Yixin Chen, Hengjian Gao, Yiheng Wang, Qi Li, Kun Li, Sheng Xu, Shengdu Chai, Fangchen Yu, Xiangyu Zhao, Zhangrui Zhao, Weijie Ma, Zijie Guo, Koutian Wu, Haoyu Zhou, Haoxiang Yin, Lixue Cheng, Chaofan Hu, Haoxuan Li, Lu Mi, Xuxuan Xie, Yifan Zhou, Ruizhe Chen, Zhiwang Zhou, Xingjian Guo, Yuhao Zhou, Xuming He, Shengyuan Xu, Xinyu Gu, Jiamin Wu, Mianxin Liu, Chunfeng Song, Fenghua Ling, Dongzhan Zhou, Shixiang Tang, Yuqiang Li, Mao Su, Peng Ye, Siqi Sun, Bin Wang, Xue Yang, Zhenfei Yin, Tianfan Fu, Guangtao Zhai, Wanli Ouyang, Bo Zhang, Lei Bai, Wenlong Zhang

AI总结:

提出ResearchClawBench基准,包含10个领域40个任务,通过多模态评分标准评估自主科研能力,最强智能体仅得21.5分,揭示当前系统在实验协议、证据匹配和科学核心方面的不足。

摘要:

AI编码智能体越来越多地用于科学工作,但其端到端自主研究能力仍然难以验证。我们提出了ResearchClawBench,一个用于评估自主科学研究的基准,涵盖来自10个科学领域的40个任务。每个任务基于一篇真实发表论文,提供相关文献和原始数据,并在评估期间隐藏目标论文。专家策划的多模态评分标准将目标科学制品分解为加权标准,从而能够评估目标论文级别的重新发现,同时为新发现留出空间。我们在统一协议下评估了七个自主研究(auto-research)智能体,并通过轻量级ResearchHarness评估了十七个原生LLM。当前系统远未达到可靠的重新发现:最强的自主智能体Claude Code平均得分为21.5,最强的ResearchHarness LLM Claude-Opus-4.7平均得分为20.7,LLM前沿均值仅为26.5。错误分析表明,失败集中在实验协议不匹配、证据不匹配和缺失科学核心。ResearchClawBench为衡量自主科学研究进展提供了一个可复现的评估前沿。

英文摘要:

AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify. We present ResearchClawBench, a benchmark for evaluating autonomous scientific research across 40 tasks from 10 scientific domains. Each task is grounded in a real published paper, provides related literature and raw data, and hides the target paper during evaluation. Expert-curated multimodal rubrics decompose the target scientific artifacts into weighted criteria, enabling evaluation of target-paper-level re-discovery while leaving room for new discovery. We evaluate seven autonomous research (auto-research) agents under a unified protocol and seventeen native LLMs through the lightweight ResearchHarness. Current systems remain far from reliable re-discovery: the strongest autonomous agent, Claude Code, averages 21.5, and the strongest ResearchHarness LLM, Claude-Opus-4.7, averages 20.7, with an LLM frontier mean of only 26.5. Error analysis shows that failures concentrate in experimental protocol mismatch, evidence mismatch, and missing scientific core. ResearchClawBench provides a reproducible evaluation frontier for measuring progress toward autonomous scientific research.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑