LLMs use historic data to help create useful current data. It works well sometimes.

I find that a human is able to solve a P=NP situation, and an LLM can’t quite yet do that. When they can the game changes.