generate an image of a small herd of Scottish highland cows, being shepherded by a Highlander in kilt and hat, and his cattle dog. In the style of an origami paper-craft scene
which resulted in the runner up image. Then I continued with:
The Highlander should be a human man with a red beard
And got:
Then I continued with:
Now the dogs are missing. Also can you make the background also look like cut/folded paper?
I recently read, and enjoyed, the Singularity Series by William Hertling. The first book is called Avogadro Corp.
It's certainly not going to win any prizes for amazing prose. And it's self-published, so the first two books could have definitely benefited from a professional editor (typos, etc). But if you're like me, and here for the interesting ideas and good understanding of tech, it's a fun little series.
One very concrete change I've noticed in the past 48 hours. When I would prompt DALL-E 3 via chat GPT4, I used to always get 4 images in response, but in the last day or two, it will only give 2 images max.
Obviously this isn't an example of the model getting qualitatively worse, but it is evidence that they are definitely making changes to the live service without change notes/communication.
If only hammers were more useful when pentesting..