A recent batch of AI-generated images showed a combination of Imperial Roman armor, architecture, and aesthetics with modern military weaponry and political imagery. It was based on the prompt "If the Roman Empire never fell." I was actually very impressed wit the general result, and it made me regret that modern military combat uniforms are not decorated as impressively. 

But I wouldn't give too much credit to the AI, as the person who came up with the prompt deserves the credit for putting the two ideas together.

And that's what I think is next for AI: self-prompting.

I'm sure researchers are simply trying to keep the recursive prompting from crashing the servers, but self-prompting is a necessary component of anything that is to be likened to intelligence.

Humans are curious creatures. We ask questions all the time. It is always a human who comes up with the prompts for AI tools.

I'm not very impressed with algorithms that can be trained - by other real humans - to recognize genres of art and writing and spit out images or text based on yet another human's input. That's not even close to what intelligence represents. That's a computer. The only difference from that process and a purely mathematical function is the element of randomness; the same prompt, or input, can give more than one result, or output. 

One certainly wouldn't call the Pythagorean Theorem or Quadratic Formula a form of intelligence. 

Until AI is able to self-prompt and demonstrate curiosity, then it will be not be intelligent. 

That's kind of the problem, actually. One usually decides what to do based on what one likes to do, and liking something is associated with emotion, not intellect. In order to get self-prompting working, there must be an Artificial Emotion, as well, in order to guide the prompting in a natural manner.