Learning to Self-Evolve

TL;DR

This paper presents Learning to Self-Evolve (LSE), a reinforcement learning framework enabling large language models to iteratively improve their own contexts at test time, leading to better performance on tasks like Text-to-SQL and question answering.

Contribution

LSE introduces a novel RL-based approach to train models for self-evolution, explicitly optimizing context refinement to enhance downstream task performance.

Findings

LSE-trained models outperform GPT-5 and Claude Sonnet 4.5 in self-evolution tasks.

LSE improves performance on Text-to-SQL and question answering benchmarks.

The approach transfers to other models without additional training.

Abstract

We introduce Learning to Self-Evolve (LSE), a reinforcement learning framework that trains large language models (LLMs) to improve their own contexts at test time. We situate LSE in the setting of test-time self-evolution, where a model iteratively refines its context from feedback on seen problems to perform better on new ones. Existing approaches rely entirely on the inherent reasoning ability of the model and never explicitly train it for this task. LSE reduces the multi-step evolution problem to a single-step RL objective, where each context edit is rewarded by the improvement in downstream performance. We pair this objective with a tree-guided evolution loop. On Text-to-SQL generation (BIRD) and general question answering (MMLU-Redux), a 4B-parameter model trained with LSE outperforms self-evolving policies powered by GPT-5 and Claude Sonnet 4.5, as well as prompt optimization methods including GEPA and TextGrad, and transfers to guide other models without additional training. Our results highlight the effectiveness of treating self-evolution as a learnable skill.