Reflect, Retry, Reward: Self-Improving LLMs via Reinforcement Learning

TL;DR

This paper introduces a self-improving framework for large language models that uses self-reflection and reinforcement learning to enhance performance on complex tasks with limited feedback.

Contribution

It presents a novel two-stage self-reflection and reinforcement learning method enabling models to self-improve without synthetic data or detailed feedback.

Findings

Up to 34.7% improvement in math tasks

Up to 18.1% improvement in function calling

Smaller models outperform larger ones in fine-tuning

Abstract

We explore a method for improving the performance of large language models through self-reflection and reinforcement learning. By incentivizing the model to generate better self-reflections when it answers incorrectly, we demonstrate that a model's ability to solve complex, verifiable tasks can be enhanced even when generating synthetic data is infeasible and only binary feedback is available. Our framework operates in two stages: first, upon failing a given task, the model generates a self-reflective commentary analyzing its previous attempt; second, the model is given another attempt at the task with the self-reflection in context. If the subsequent attempt succeeds, the tokens generated during the self-reflection phase are rewarded. Our experimental results show substantial performance gains across a variety of model architectures, as high as 34.7% improvement at math equation writing and 18.1% improvement at function calling. Notably, smaller fine-tuned models (1.5 billion to 7 billion parameters) outperform models in the same family that are 10 times larger. Our novel paradigm is thus an exciting pathway to more useful and reliable language models that can self-improve on challenging tasks with limited external feedback.