The circular problem of epistemic irresponsibility — LessWrong