Author name: Fengwei Teng, Jinyi Bai, Xinhao Yao, Demi Ruohan Wang, Jiahao Zhao, Zhijiang Guo

arXiv:2604.08690v1 Announce Type: new
Abstract: Group Relative Policy Optimization (GRPO) has proven effective in RLVR by using outcome-based rewards. While fine-grained dense rewards can theoretically improve performance, we reveal that under practic…