My use of AI tends to fall into three categories:

• Generative Art
• Failed attempts to use AI as a coding assistant
• Collecting examples for my soon-to-be-published “Artificial Stupidity” page.

There is significant overlap between the latter two categories.

Generative Art

As can be seen on my Generative Illustrations category, I do a lot of work with Generative AI art.

Most often, believe it or not, I use Bing’s instance of Dall-E, because it’s fast and free. Unfortunately earlier this year they moved from Dall-E 2.5 to Dall-E 3, which is significantly hobbled in its ability to produce abstraction, and I’m in search of a replacement. As a demonstration: A Helluva Band, one of my favorite early galleries, was done in Dall-E 2.5, hence it’s more impressionistic, almost symbolist imagery, while Dream Jobs was an early experiment with Dall-E 3, and show’s its greater emphasis in crisp illustrations and concrete subjects, in just a few styles that it tends to prefer: either a sort of cartoonish, hand-drafted style with a lot of filigree and earthy colors; an airbrushed, digital-illustration type style reminiscent of drawings made using digital illustration tools; and a hyperrealistic photography style featuring skinny people who appear to be in their early 20s with hollow cheeks, strong cheekbones and narrow chins. Getting it to break out of these defaults is difficult and I’m pretty disappointed that I can’t get the lurid, indefinite forms of Dall-E 2.5 anymore.

I also occasionally wander into other generative art tools like Stable Diffusion or running old VQGAN on Google Code. But I keep coming back to Bing because I’ve gotten to know it, it’s free, and easy.

Failed attempts to use AI as a coding assistant

I have a subscription to Quora’s poe.com which gives me access to the latest popular models for one price. I have often tried to use Anthropic’s Claude 3 Opus, Claude 3.5 Sonnet, or ChatGPT 4o to help solve coding problems or automate repetitive coding tasks. I also for a few months subscribed to GitHub’s Copilot service and used it in the VS Code or Cursor IDEs.

The results of many months of such efforts were, I think I once did get one of them to produce working code in a reasonable amount of time. Much of the remainder will fall into the next section.

I recently canceled my subscription to Copilot. It was a waste of money.

Collecting examples for my soon-to-be-published “Artificial Stupidity” page

I have privately saved many glaring examples of obnoxiously unintelligent behavior from so-called “AIs”, which will eventually be organized and published in an “Artificial Stupidity” section somewhere on this site.

Until then, here’s a typical screenshot. Here, Claude 3.5 Sonnet, Anthropic’s most advanced model at present writing, gives me code with an obvious big mistake in it. When I point out the mistake, it claims never to have given me any code, and when I point out that it just did give me code, it arguest, insisting that it hadn’t and that I must have gotten confused.

This is going to be a fun gallery when I get all these posted.