1000% this. I would only add this has been demonstrated explicitly with chess: https://adamkarvonen.github.io/machine_learning/2024/01/03/c...