The Bayesian Tyrant — LessWrong