That is the big question with LLM's. How can we tell what is being fed in is original content or just the output fed back in like a recursive fractal?