A soccer simulator played by AI Agents (gangtao.github.io)

A groundbreaking AI project has introduced a soccer simulator where each player is an AI agent powered by large language models (LLMs) that use a decision-making callback function. This innovative setup features clean separations between the API surface and the simulation engine, ensuring robust security by sandboxing the generated code. These layers facilitate the creation of a safer environment for the agents to evolve and interact without risking the host system. The significance of this development lies in its structured approach to AI competition, with matches being classified into three formats that promote strategic evolution. As tournaments progress, the AI agents learn and adapt, enhancing their performance over time. This setup not only showcases the capabilities of LLMs in complex environments but also highlights the potential for AI to evolve continuously through competitive interactions, paving the way for more sophisticated AI applications in gaming and beyond.