论文: GEPA: Reflective Prompt Evolution Can Outperform Reinforcement LearningarXiv: 2507.19457会议: ICLR 2026 (Oral)作者: Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alexandros G. Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, Omar Khattab代码: https://github.com/gepa-ai/gepa相关工具:optimize_anything()API (2026-02-18 发布)一、一个反直觉的结论：说话比打分更高效强化学习（RL）训练 LLM 的基本逻辑是：让模型试很多次 → 用标量 reward（0/1，或 0~1 之间的一个数）告诉它"这次好还是不好" → 通过 policy gradi