Efficient prestige hypothesis — LessWrong