RL for LLM Reasoning Is Sparse Policy Selection, Not Capability Learning

2 points | by BlackGlory 8 hours ago