"Self-pretending" is not as useful as we think — LessWrong