LLMs are just a fancy way of saying "expensive automaton that's only as smart as the data it was trained on