x
Learning human preferences: optimistic and pessimistic scenarios — LessWrong