AI Papers Daily

#Medical LLMs

1 篇

Quantifying and Mitigating Premature Closure in Frontier LLMs

量化与缓解前沿大语言模型的过早闭合行为

🧭 Eval ★ 深度解读

Stanford University

在医疗场景中，前沿LLM普遍在信息不足时给出不安全答案，安全提示可部分缓解但无法根除

Eval | 2026-05-14 | cs.CL | score 8/10

Premature ClosureMedical LLMsEvalPrompt EngineeringAbstentionSafety