GEPA demonstrates that reflective evolution can outperform reinforcement learning. A revolutionary approach to AI system optimization through natural language feedback.
Agrawal, Lakshya A., et al. GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning. arXiv:2507.19457, July 2025, https://arxiv.org/abs/2507.19457.