Yep that's nice, although it seems to be proprietary which isn't ideal, it's the last thing we need now. Companies exploited the hell out of everything, now companies/universities exploiting the solutions too, when there's absolutely nothing stopping it from being open.

It's interesting but can't work outside of a lab. The example they gave was watermarking a picture of a cow with a purse. If every cow picture has a hidden purse watermark, an AI will be trained into categorizing a cow as a purse.

But to make that work in the real world would require every artist and especially stock photo sites to agree to watermark every cow with a purse. If everyone doesn't pick a consistent watermark of a purse for a cow, then it becomes noise that is trained out. Just like training a to identify a cow sometimes has a farmhouse in the picture, other times grass, other times birds. It learns cow because that's the consistent part. Without a purse watermarked into the majority of every cow photo everywhere, the ai will learn cow.

Well that's what the project I linked does, although I'm not sure it solves all of the issues right now, it's definitely a start.

I'm going to link this at the current revision, so that it makes sense in the future: https://en.wikipedia.org/w/index.php?title=Transformer_%28deep_learning%29&oldid=1333135164

Read the first line from the link, I'll add it here if you're lazy: "In deep learning, the transformer is an artificial neural network..."

Do you know what "GPT" stands for? "Generative Pre-trained Transformers"

What were you thinking LLMs use? They're literally just neural networks stacked as much as possible. That's why they require all of those data centers, because their only solution to the problem is adding more neural nets and more data which means more hardware, at this point it's borderline brute forcing. Sure, you can mention the """clever""" tricks they use to "tokenize" words at the beginning, but that's still a neural net in itself. Don't get confused by their terminology, every single bit of the "technology" has impressive sounding names until you see how they actually work and smack your forehead so hard it leaves a mark forever.

Cool, but again, seems proprietary, which is not ideal. Also isn't it a bit backwards to add artifacts, instead of look for ways to detect artifacts in generated images, so that we catch early and avoid AI in the first place?