arXiv · Artificial Intelligence· Abinitha Gourabathina, Haoran Zhang, Yuexing Hao, Walter Gerych, Marzyeh Ghassemi·2026-10-01 12:00· 3 小时前AI 评分52

Sense and Sensitivity：基于医生专家基准评估LLM临床分诊推荐

Sense and Sensitivity: Benchmarking LLM Clinical Triage Recommendations with Physician Experts

AI 导读

论文提出覆盖6000+临床场景、7000+医生标注和225000条模型响应的临床分诊基准，发现LLM比医生更倾向推荐不必要护理，且对性别和语气扰动更敏感。