That’s not what’s happening.

The question is revealing that the model has a model of language but not of reality. It knows what words go together, but not real-world concepts.