A tiny world model running games on a Raspberry Pi Pico 2 W

A developer created a system that runs two small neural networks on a Raspberry Pi Pico 2 W: one generates game frames based on actions, and the other learns to play within the generated game world. The approach uses looped transformers with reused weights to keep memory usage minimal, allowing the microcontroller to both simulate games and train an AI agent entirely on-device without accessing the real game.