Misspecification in Inverse Reinforcement Learning - Part II — LessWrong