Training large language models for complex reasoning tasks relies on two complementary signals: sparse, outcome-grounded reinforcement learning (RL) for corr…
机构：上交
来源：arXiv 2610.30796 | AI4Papers 论文推荐平台