马东锡 NLP· @dongxi_nlp · X·· 2026-08-01AI 评分29
AI 导读
学习可以反向进行:从专家行为出发,发现其背后的推理过程,然后教会模型正向推理。 论文: LeAct:从专家行为中学习推理
正文 · 原文
Learning can run backward: begin with an expert action, discover the reasoning behind it, then teach the model to reason forward.
Paper:
LeAct: Learning to Reason from Expert Actions
来源:马东锡 NLP · x.com