Learning to Hint for Reinforcement Learning

2026-04-01

AI Summary

This work addresses the issue of advantage collapse in reinforcement learning, which arises when all trajectories receive identical group-level rewards—particularly problematic when the reasoning agent fails on challenging tasks, leading to a lack of informative learning signals. To mitigate this, the authors propose HiLL, a framework that jointly trains a prompt generator and a reasoning agent via an adaptive online prompting mechanism. HiLL introduces a prompt dependency metric and theoretically demonstrates that prompts with lower dependency better enhance the performance of the unprompted policy. Leveraging this insight, a transfer-aware reward weighting scheme is devised. Combined with conditional prompt generation and Group Relative Policy Optimization (GRPO), HiLL enables end-to-end joint training. Experimental results across multiple benchmarks show that HiLL significantly outperforms existing methods, validating the efficacy of adaptive, transfer-aware prompt learning in enhancing reinforcement learning performance.

Abstract

Group Relative Policy Optimization (GRPO) is widely used for reinforcement learning with verifiable rewards, but it often suffers from advantage collapse: when all rollouts in a group receive the same reward, the group yields zero relative advantage and thus no learning signal. For example, if a question is too hard for the reasoner, all sampled rollouts can be incorrect and receive zero reward. Recent work addresses this issue by adding hints or auxiliary scaffolds to such hard questions so that the reasoner produces mixed outcomes and recovers a non-zero update. However, existing hints are usually fixed rather than adapted to the current reasoner, and a hint that creates learning signal under the hinted input does not necessarily improve the no-hint policy used at test time. To this end, we propose Hint Learning for Reinforcement Learning (HiLL), a framework that jointly trains a hinter policy and a reasoner policy during RL. For each hard question, the hinter generates hints online conditioned on the current reasoner's incorrect rollout, allowing hint generation to adapt to the reasoner's evolving errors. We further introduce hint reliance, which measures how strongly correct hinted trajectories depend on the hint. We derive a transferability result showing that lower hint reliance implies stronger transfer from hinted success to no-hint success, and we use this result to define a transfer-weighted reward for training the hinter. Therefore, HiLL favors hints that not only recover informative GRPO groups, but also produce signals that are more likely to improve the original no-hint policy. Experiments across multiple benchmarks show that HiLL consistently outperforms GRPO and prior hint-based baselines, demonstrating the value of adaptive and transfer-aware hint learning for RL. The code is available at https://github.com/Andree-9/HiLL.

Problem

• advantage collapse
• reinforcement learning
• hinting
• policy transfer
• GRPO

Innovation

• adaptive hinting
• hint reliance
• transfer-aware reinforcement learning
• relative policy optimization
• joint policy training