Stuart_Armstrong comments on Learning human preferences: optimistic and pessimistic scenarios