Going back to a very old school notion of "AI" : just updated a demo I did a few years back on efficiently computing cellular automata forward steps using convolution operations. Added a huggingface demo too.
<a href="https://github.com/thearn/game-of-life">https://github.com/thearn/game-of-life</a><p>for fun, I let users swap rules dynamically in the app, try letting it run Conway for a short bit, pause the simulation, switch to Maze and start from there.