💭 I'd optimistically thought muzero might be able to learn models for hidden info and rng games but apparently those are still major limitations with experimental solutions. From trying to make those things in mcts in the past I expect the solutions to not be pretty