Remix.run Logo
▲ dcl 5 hours ago

I do not expect to get far, but I'm having a bit of fun using AI to produce a performant Rust engine for 4 player EDH designed with reinforcement learning in mind and also thinking about RL modelling strategies for such a complex game.

I've learned a bit about the game itself trying to figure out good ways to represent and model things. There's so much we do 'automatically' (perhaps mistakenly) when playing as a human, that a is hard to efficiently represent to an agent or a bot (or even to get the engine to accurately reflect the gamestate).

I note that Fable 5.2 improved our throughput on game engine mechanics and playable cards dramatically (mostly through redesigning processes), but I worry the Anthropic and OpenAI models will crap out when I begin asking them to help out with RL.

Out of curiosity:

- When a new set comes out, how fast do the existing engines 'catch up' with the new rules and cards?

- What engine is considered the most accurate and bug free at the moment?