💭 The obvious advice for board game ai project of validate the model, ignore the engine framework part, start with small games and build up - is actually really bad advice in my particular position? We already know that alphazero/muzero work. We already know transformer architecture is quite powerful with them. The only new finding I could make from my work would be if the generality improvements are possible. Otherwise I'm just manually making ai for games. I guess I should settle for that.