It has no way of judging the quality of it's own answer, I see this trait as it's biggest hurdle to most practical uses.