Right, it seems incorrect to claim that "The model can generate text that it couldn’t possibly have seen in the training data" when it's been trained on all man

Right, it seems incorrect to claim that "The model can generate text that it couldn’t possibly have seen in the training data" when it's been trained on all manner of humans writing about all manner of things. It seems likely that the training data included text involving " self-serving bias, metaphor, statistical syllogism and common-knowledge physics". So, when given a prompt asking for these things; the model is able to 'Stochastic Parrot' an appropriate response. The graph theory work is clever, but the assumptions about the Models, and about Language, seem incorrect.