Why RL Feedback Fails Language Models (And What ERL Fixes)

ERL adds a reflection step to reinforcement learning: attempt, feedback, explanation, refined attempt. The result: faster learning, higher reward, same inference cost.