Verifiable rewards have advanced reinforcement learning in mathematics and code. However, many valuable open-ended tasks share three properties: no automatic…
机构:中科院
来源:arXiv 2610.25434 | AI4Papers 论文推荐平台