I try to sharpen this by e.g. making a point to distinguish between LLMs as technology and LLMs as product. Studying LLMs can tell us some cool structural stuff about language (was Chomsky right?), but that's a very different use-case than letting them run DOGE or do people's homework