Papers17:56Chinese contentLLM验证新框架:选择性精修提升医疗问答准确率这篇论文教你怎么用弃权信号做推理精修,不用建知识图谱,医疗问答准确率直接涨9.6个点,GPT-5.5和DeepSeek-R1都试过。#GPT-5.5#DeepSeek-R1#MedQA#MedReasonaarXiv: OpenAI@Uma Ranjan et al. (12)SourceRead laterReadWorth followingUsefulFollow GPT-5.5