💭 I have a concern that a key obstacle to developing strong game AI is compute, which I don't have and don't want to pay much for. Figuring out generalizable algorithms thatd work for diverse complex games kind of presumes training AIs for those game as validation. Its very unlikely everything will work in few attempts. Might get to a point where I have a research direction I feel good about and then shelve it because I don't want to pay that much gpu money to go the rest of the way. Also rl is painful stuff.