Today’s post will be brief, and poor St Jerome probably won’t make it to the conference, as far as this blog is concerned. Time travel is exhausting, especially when you have to walk.
After what seemed an eternity, Jerome comes up on an open space, with a modern glassy building in the distance. But the world is still empty, and vegetation seems to take over everywhere, although at most places it looks like it’s been well tended to. But where are the people who take care of all this?
Lost in thought, Jerome almost walks into a giant penguin who happens by. Desperate for an answer, he implores the penguin to say what he knows. The penguin, very sternly, gives Jerome a lecture of how he overshot the time and needs to go back almost a hundred years.
It isn’t shown in the images, but Jerome then sits down on a rock. He’s very exhausted now, but he tries to gather his strength to figure out his next step.
Me: Create a photorealistic image: St Jerome, dressed as a Franciscan monk, walks through a park. At the far end, there is a futuristic high-rise building. The park continues in the building. It’s sunny but the light is strangely blue.
I was counting on a much more peculiar image, and when revisiting this session, I actually felt disappointed. So, I repeated the prompt four months later, correcting some mistakes I thought I made back then.
Me: Create a photorealistic image: St Jerome, dressed as a Franciscan monk, walks through a park. At the far end of the park, there is a futuristic high-rise building. The park continues inside the building. It’s sunny but the light is a strange dim blue color.
As you can see, not much changed. DALL-E doesn’t have the concept of a park inside a building, so that part of the prompt still gets ignored.
Anyway, let’s have Jerome meet the penguin (“The penguing is penguing”, as Benedict Cumberbatch would say):
Me: Recreate the image to show that St Jerome stops to talk to a black penguin as tall as him.
The penguin comes out much taller than St Jerome, which proves that DALL-E doesn’t know sizes and comparisons. I have the feeling that by the time we catch up with my DALL-E backlog, we will strike off the relationship with all of the most common elements of reality, illustrating how the language model is a pure language model — and it needs help from the outside if reality needs to be handled.
thirdtower.ai 