Where Hindsight Credit Can Reside: A Signed-Capacity View of Token Updates in RLVR 事件

Where Hindsight Credit Can Reside: A Signed-Capacity View of Token Updates in RLVR arXiv:2604.11056v2 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) improves the reasoning ability of Large Language Models (LLMs), but sparse outcome rewards make token-level credit assignment difficult. We study token-level credit as a reward-conditioned shift from the behavior policy to a hindsight posterior. In autoregressive RLVR, this shift can be expressed th