I think that’s still anthropomorphization. The point I’m making is that these things aren’t “assumptions” as we characterize them, not from the model’s perspective. We use assumptions as an analogy but the analogy becomes leaky when we get to the edges (like this situation).

It is not anthropomorphism. It is literally a prediction model and saying that a model "assumes" something is common parlance. This isn't new to neural models, this is a general way that we discuss all sorts of models from physical to conceptual.

And in the case of an LLM, walking a noncommutative path down a probabilistic knowledge manifold, it's incorrect to oversimplify the model's capabilities as simply parroting a training dataset. It has an internal world model and is capable of simulation.