PAG: Multi-Turn Reinforced LLM Self-Correction with Policy as Generative Verifier figure
AlphaXiv 中文概览(可滚动查看)