The Bayesian Agent — LessWrong