Learning to Self-Evolve
This addresses the challenge of enhancing model performance through self-evolution for users in natural language processing, representing a novel method rather than an incremental improvement.
The paper tackles the problem of enabling large language models to improve their own contexts at test time by introducing Learning to Self-Evolve (LSE), a reinforcement learning framework that trains models for this task, resulting in a 4B-parameter model outperforming self-evolving policies powered by GPT-5 and Claude Sonnet 4.5 on Text-to-SQL generation and general question answering benchmarks.
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.