What's the point in DALL·E 2?

DALL·E 2 is the latest blockbuster AI model from OpenAI. The model can generate fairly convincing images from a short text prompt.

Dave Orr has been playing with the model and has written about it on his blog. Certain types of cartoonish images work well, but most realistic images, particularly with people or text, seem to get things dramatically wrong.

I don't know much about these types of model. They are fun and are very impressive. But I can't see (at this stage) any practical application for them. Can anybody else?