让AI自己给自己“立规矩”，结果会怎样？

[LG] AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning

[CMU]

arxiv.org