Why doesn't a computer program that predicts the next word in a sentence thereby predict what happens next in the physical world?

Answer: Word order captures how people talk about events, not the physical rules that decide what actually happens

Reality is built from grammar rules, so sentences and objects follow exactly the same order — Grammar is a set of rules for arranging words, and it is tempting to imagine the world arranged the same way. But a cup falls because of gravity and a slipped grip, not because a sentence put 'falls' after 'cup'. Physical events obey forces and constraints that have nothing to do with sentence structure.

Word order captures how people talk about events, not the physical rules that decide what actually happens ✓ — A program trained on text learns which words tend to follow which, and that reflects how people describe the world. Describing a cup falling and predicting a cup falling are different tasks: the second needs gravity, support, timing and cause-and-effect, which the text only mentions in passing. Such a program may pick up some physical regularities from what people have written, but good word prediction does not by itself amount to a model of what will physically happen.

Objects move the way we describe them, so writing the right sentence makes the next event happen — A sentence can set up expectations, and a clear description helps us plan. But writing 'the cup stays put' does not hold the cup in place; it falls or not according to physics, however it is described. Prediction has to follow the world, and the world does not follow our wording.