@swyx· @swyx · X·2026-06-03 14:33· 2026-06-03AI 评分16阅读原文 AI 导读这可能是我见过的用于推理效率的最好的奖励函数 @eliebakouch：长度惩罚其实非常优雅且简单正文probably the best reward function for reasoning efficiency i've seen引用elie (@eliebakouch)@eliebakouchlength penalty is very elegant and simple tbh在 X 查看被引用的帖子 来源：@swyx · x.com#大佬观点#推理