arXiv · Computation and Language· Yingfeng Luo, Shaowei Wei, Daixin Wang, Dingyang Lin, Kaiyan Chang, Weiqiao Shan, Tong Zheng, Zhiqiang Zhang, Jingbo Zhu, Tong Xiao·· 7 hr agoSelectedAI score66
终端智能体能信任自己的验证吗？诊断与改进自验证
Can Terminal Agents Trust Their Own Verification? Diagnosing and Improving Self-Verification
AI brief
论文提出诊断框架，系统评估终端智能体在 TerminalBench2.1 上的自验证行为，发现验证几乎总会触发，但错误检出率仅 61.43%、修复率仅 49.36%，主要瓶颈在检测与修复而非启动验证。
Why it matters
诊断了终端智能体自验证的漏检与修复率，并给出蒸馏方法提升修复效果。
Source: arXiv · Computation and Language · arxiv.org