Show HN: Teaching a 30ms local model, play browser games with LLM written rules (github.com)

A new project, IronBee Gamer, has emerged that leverages Local Language Models (LLMs) to enable automated play of browser games without any game-specific APIs or hooks. By reading and interpreting a game's state through a custom JSON format, IronBee Gamer's decision engine, Laya, acts within a 30ms response time to decide every game move. It promotes an innovative approach for training an LLM to generate rules and strategies for various games directly based on their displayed content, significantly simplifying the process of game automation and AI training. This development is particularly significant for the AI/ML community as it showcases the potential for creating flexible, generalized systems capable of learning game dynamics without extensive manual coding. The architecture allows for easy integration of new games by defining them through a basic setup with the coding-agent CLI. Moreover, players can interactively observe decisions made by the model in real-time, enhancing transparency and learning opportunities. With built-in functionalities for training and distilling models, IronBee Gamer stands to streamline the process of AI training in gaming environments, encouraging wider experimentation and innovation in AI game playing strategies.