Idealized Factored Cognition — LessWrong