Nudging Beyond the Comfort Zone: Efficient Strategy-Guided Exploration for RLVR

Excerpt

arXiv:2605.15726v2 Announce Type: replace Abstract: Reinforcement learning with verifiable rewards (RLVR) is a scalable paradigm for improving the mathematical reasoning of large language models, but it is fundamentally limited by exploration: the policy can only improve on trajectories it has already sampled. Sampling more rollouts alleviates this at prohibitive compute cost, while objective-level modifications offer little control over what is explored. We propose NudgeRL, a framework for stru