Large language models (LLMs) have advanced rapidly in recent years, driven by scale, abundant high-quality training data, and reinforcement learning. Yet this progress faces a fundamental bottleneck: the need for ever more data from which models can continue to learn. In this work, we propose a reinforcement learning approach that removes this dependency by enabling models to improve without additional data. Our method leverages a game-theoretic framework of self-play, where a model's capabilities are cast as performance in a competitive game and stronger policies emerge by having the model play against itself-a process we call Language Self-Play (LSP). Experiments with Llama-3.2-3B-Instruct on instruction-following, mathematics, and coding benchmarks show that pretrained models can be effectively improved with self-play alone.
@article{arxiv.2509.07414,
title = {Language Self-Play For Data-Free Training},
author = {Jakub Grudzien Kuba and Mengting Gu and Qi Ma and Yuandong Tian and Vijai Mohan and Jason Chen},
journal= {arXiv preprint arXiv:2509.07414},
year = {2025}
}