Weak-Driven Learning: How Weak Agents make Strong Agents Stronger

TL;DR

This paper introduces WMSS, a post-training method that uses weak model checkpoints to identify and recover learning gaps, enabling large language models to surpass saturation limits and improve performance without extra inference costs.

Contribution

The paper presents WMSS, a novel post-training paradigm that leverages weak checkpoints to enhance model performance beyond traditional saturation points.

Findings

Improves mathematical reasoning and code generation performance.

Achieves performance gains without additional inference costs.

Effectively recovers learning gaps through entropy-based analysis.

Abstract

As post-training optimization becomes central to improving large language models, we observe a persistent saturation bottleneck: once models grow highly confident, further training yields diminishing returns. While existing methods continue to reinforce target predictions, we find that informative supervision signals remain latent in models' own historical weak states. Motivated by this observation, we propose WMSS (Weak Agents Can Make Strong Agents Stronger), a post-training paradigm that leverages weak checkpoints to guide continued optimization. By identifying recoverable learning gaps via entropy dynamics and reinforcing them through compensatory learning, WMSS enables strong agents to improve beyond conventional post-training saturation. Experiments on mathematical reasoning and code generation datasets show that agents trained with our approach achieve effective performance improvements, while incurring zero additional inference cost.