You worry about AI. We worry about AI. But actually, we don't need to worry. Because this is fine. For proof, just ask the generative AI to repeat the cartoon meme, "This is fine. It will immediately be convinced that, indeed, everything is fine. Or it will conclude that this is completely bonkers, circular, meta, self-referential, and brilliantly crazy. And like me, you will laugh and cry.

The ploy goes like this: ask a text-based AI model like CoPilot or ChatGPT to input the prompt "a cartoon of a dog saying 'this is fine'" into an image generator like DALL-E 3, and the output will be almost identical to the original cartoon. Here is a reddit thread on how to know if an image generator has learned for an image.

The original, incidentally, is from the "Gunshow" webcomic series, drawn by K.C. Green and published in early January 2013.

In any case, madness and laughter occur when the model is made to repeat its initial output. It's really just simple repetition that doesn't seem to imply strangeness or insanity. For example, "turn it into a four-panel cartoon," or "turn the dog into a pug," or "add bees," or "turn fire into ice."

The complete insanity into which the image is rapidly descending is both utterly hilarious and faintly alarming, and perfectly describes the current state of AI.

This is AI in its purest form. It is completely mad, completely alien yet human, chaotically creative, subtly knowing yet somehow destructive.

Sure, there are some cartoon dogs that are a bit alarmingly deformed. But there will also be dingy text that almost certainly isn't there, but which makes some sort of sense as well as hinting at a newborn consciousness that human instincts can't help but perceive. A strange mash-up of visuals and text, where the boundary between text and image is blurred. And pure, dizzying, thrilling, hilarious madness.

Oh, it's definitely a great way to kill an hour or three on a Friday afternoon. Enjoy.