Good Reasoning Makes Good Demonstrations: Implicit Reasoning Quality Supervision via In-Context Reinforcement Learning
This addresses the issue of improving reasoning quality in large language models for tasks like mathematics, though it is incremental as it builds upon existing RLVR methods.
The paper tackles the problem of reinforcement learning for reasoning in large language models, which often reinforces flawed reasoning traces that yield correct answers by chance, by introducing In-Context RLVR that implicitly reweights rewards based on reasoning quality, leading to improvements in accuracy and reasoning quality on mathematical benchmarks.
Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that get correct answers by chance. We observe that better reasoning are better teachers: high-quality solutions serve as more effective demonstrations than low-quality ones. We term this teaching ability Demonstration Utility, and show that the policy model's own in-context learning ability provides an efficient way to measure it, yielding a quality signal termed Evidence Gain. To employ this signal during training, we introduce In-Context RLVR. By Bayesian analysis, we show that this objective implicitly reweights rewards by Evidence Gain, assigning higher weights to high-quality traces and lower weights to low-quality ones, without requiring costly computation or external evaluators. Experiments on mathematical benchmarks show improvements in both accuracy and reasoning quality over standard RLVR.