Well it's the prediction part that is complicated. How that works is a mystery. But even our LLMs are for a certain part a mystery.