AI的自我修炼：如何让机器拥有“反思”的能力？

[CL] Direct Reasoning Optimization：LLMs Can Reward And Refine Their Own Reasoning for Open-Ended Tasks

[Microsoft]

arxiv.org