That’s a good definition: it’s a model of a model.

It seems the debate seems to center around whether language models are meta-models (in the category sense) or mere encodings (information theory)?