GEPA optimizes LLMs without costly reinforcement learning

Moving beyond the slow, costly trial-and-error of RL, GEPA teaches AI systems to learn and improve using natural language.