The human doesn't just predict. It predicts based upon simulations that it runs. These LLMs do not work like this.