That's really funny when you consider that generative models also don't hallucinate if you check up on them on every token generated?