Copilot and ChatGPT iterating 'This is fine' memes perfectly captures the current state of AI: faintly alarming hilarious madness

pcgamer.com:

You're worried about AI. We're worried about AI. But there's actually no need to worry. Because this is fine. For proof, all you need to do is ask generative AI to iterate on the «This is fine» cartoon meme. Pretty quickly, you'll agree that it is indeed all fine. Either that or you conclude this is completely bonkers, circular, meta, self-referential and beautifully deranged. And you'll be weeping with laughter like me.

The ruse goes like this. Ask a text-based AI model like CoPilot or ChatGPT to feed a prompt for «a comic of a dog saying 'this is fine'» into an image generator such as DALL-E 3 and the output will likely be pretty much exactly the same as the original cartoon. This is indeed where my favourite new AI rabbit hole has come from: a reddit thread about how to know whether an image generator has trained on an image.

That original, for the record, was from the webcomic series Gunshow illustrated by K.C. Green and published in early January 2013.

Anyway, it's when you ask the model to iterate on that initial output where the madness and hilarity ensues. You know, really simple iterations that you wouldn't think imply any weirdness or madness. Like, «make it into a four-pane cartoon» or «have the dog be a Pug», «add bees» or «turn the fire into ice».

The utter insanity into which the images rapidly descends is, all at the same time, totally hilarious and faintly alarming, but also perfectly encapsulating of the current state of AI.

This is AI in its purest form. Totally mad, utterly alien and yet disarmingly human, chaotically creative, subtly knowing and somehow destructively oblivious.

Sure, you'll get some slightly alarmingly deformed cartoon dogs. But you'll also get gibberish text that sort of makes sense at the same time as hinting at a nascent consciousness that almost certainly isn't there but which our human instincts can't help but perceive. And weird visual-to-text mashups where the delineation between text characters and images are blurred. And