Dall-e must generate an image, even for a random prompt of letters, because it has been trained to do only that. Take a human, hold a gun to their head, and force them to paint an image for a random string of letters, and I guarantee they will make you something.

Retrain dall-e and give it a choice whether it generates an image or does something else, and you will get a different outcome.

The argument boils down to this: is a human brain nothing but a mapping of inputs onto outputs that loops back on itself? If so the dall-e / gpt-3 approach can scale up to the same complexity. If not, why not?