💭 interesting thing with the board game ai project is I think I've largely filled all the unknowns (that I know of). Every hard thing between me and the goal I have, at minimum, an established paper known to solve the problem. The substance of the gap between me and success is just a massive gulf of careful engineering and tuning. I'm used to such situations being a speedrun, but this is huge and complex enough that if I move forward with any shaky foundation I'll basically need to restart later. Eg I need an easy to write for game engine which creates definitions that can be automatically converted into optimized ml setups. A change to any part of that pipeline could cause problems in the other parts.