“The state-enumeration approach used to build superhuman poker AIs breaks down when applied to games with a much larger number of hidden states, such as Omaha poker or Stratego.”