Proper value learning through indifference — LessWrong