In defence of epistemic modesty — LessWrong