Good Reasoning Makes Good Demonstrations: Implicit Reasoning Quality Supervision via In-Context Reinforcement Learning

TL;DR

This paper introduces In-Context RLVR, a method that uses the model's own reasoning ability to implicitly evaluate and prioritize high-quality demonstrations, leading to improved reasoning and accuracy in large language models.

Contribution

The paper proposes In-Context RLVR, a novel training approach that leverages the model's demonstration utility to implicitly reweight training traces based on reasoning quality.

Findings

Improved accuracy on mathematical benchmarks.

Enhanced reasoning quality over standard RLVR.

Effective implicit reweighting without external evaluators.

Abstract

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.