I played around with a few images to see how the AI worked, and I think what struck me the most is how just a slight change (i.e., a misspelled word) drastically alters the image panel put together by the algorithm. In one scenario, I typed in “Art Decco” (the incorrect spelling), and was presented with this image:
In another, I used the correct spelling (Art Deco) and was presented with the following:
I assumed the AI would be able to “guess” what I wanted in the first image, but I was surprised by how off it was from what I was expecting. It might have been thrown off because Decco doesn’t mean anything, and so the algorithm was confused.