LLM学会“思考”：元认知重用让大模型推理又快又准 - ListenHub