When All Rollouts Fail: Policy-Compatible Joint Decoding for Hard-Problem Reasoning
Reinforcement Learning with Verifiable Rewards (RLVR) derives its learning signal from binary rewards via on-policy exploration and becomes a powerful paradigm for improving the reasoning of large language models. Yet it faces a learning cliff on problems beyond model's current ability: when every rollout fails, the samples reveal nothing about a correct solution, and group-relative estimators…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
