Comment by onion2k

That implies you think the output of an LLM is only a rewording of the prompt, and that it doesn't add anything from it's corpus of training data. That's obviously untrue.