1
OpenSIR: Open-Ended Self-Improving Reasoner
提出一种开放式自我改进推理器OpenSIR，通过自博弈突破标注数据限制，有望超越人类水平推理。
arXiv:2511.00602v3 Announce Type: replace Abstract: Recent advances in large language model (LLM) reasoning through reinforcement learning rely on ann…