Comment by freejazz

> They aren't references to internal concepts, the model is not aware that it's doing anything so how could it "explain itself"?

You should read OpenAI's brief on the issue of fair use in its cases. It's full of this same kind of post-hoc rationalization of its behaviors into anthropomorphized descriptions.