Bruh. (Gaming)

by stabbim , Des Moines, IA, USA, Friday, March 27, 2026, 10:34 (1 hours, 38 minutes ago) @ Kermit
edited by stabbim, Friday, March 27, 2026, 10:38

Life-saving benefits? I think maybe you don't mean the same thing by generative AI that I would.

I bet you're right.

So, when I say generative AI, I mean AI which generates content. Image, video, and audio generators, and basically any chat bot or similar - the text output being the generated content. As an example of the other case, I wouldn't consider a model trained on X-Rays to identify spots on lungs as generative AI. Flagging X-Rays as potential cancer isn't generating content, to me.

The reason that I default to being negative about them is that, for the most part, the models were trained on just... whatever content they could find on the internet. As much text, image, audio or video as they could get their hands on. Whether it's copyrighted, public domain, or somewhere in between, probably wasn't taken into account and in my estimation that's theft. It might not be intended to directly duplicate the content, but it's still using it without the creator's consent, usually as part of a product to make money for people who are not those creators (worth considering, at least in the early days basically all of the training data would have been created before generative AI existed or was widely known, and thus the authors couldn't possibly have consented). And sometimes it does duplicate the content. Despite the name, AI is fundamentally dumb and lacks the judgement or awareness to realize when it has made a copy, so occasionally it'll happen.

Edit/addition: I don't tend to have these concerns about the medical diagnostic stuff, for example. I would assume most of that work is being done by academic institutions and/or medical technology companies and they probably have authorization to use the medical data for research. That seems completely fair, and I would think any concerns about that content would be more about privacy than copyright/intellectual property.

Now, if every company would say what their models were trained on, we as consumers could at least make a judgement call, although IMO very few companies would choose to limit their training data because more is almost always better (unless the data in question is poor quality). But they almost never say anything, so for those of us who have these concerns, we just have to assume it's all poison by default most of the time.