That’s not exactly how generative AI works. Some does have CSAM in its training material, which I said they should be held accountable. Early Stable Diffusion had it, and so does Grok (and I doubt Musk ever removed it).
But generative AIs can create things not in its data set.
Like you said, its a remix machine of sorts, to put it simply. It doesn’t need actual CSAM to generate that, it just needs nude adults, clothed children, and sex images to extrapolate from to then generate pedophilic content. That’s why they can’t easily stop generators from generating such content even on the ones they’ve confirmed don’t have it on their data set, and have to rely instead on keyword blocking.
You could perhaps go through a data set and remove all images of children, but then it’s likely someone would still find a way to loophole it (using keywords like gnomes or cherubs perhaps), or you could remove all nudity but it could also break the gen AIs capability to depict humans at all (as was the case with one of the later Stable Diffusion models, 3.0 I think).
You can chill on the witch hunting. I get you are understandably passionate on the subject but accusing anyone trying to find solutions of doing the opposite or being disingenous won’t help either.
I work in tech.
That’s not exactly how generative AI works. Some does have CSAM in its training material, which I said they should be held accountable. Early Stable Diffusion had it, and so does Grok (and I doubt Musk ever removed it).
But generative AIs can create things not in its data set.
Like you said, its a remix machine of sorts, to put it simply. It doesn’t need actual CSAM to generate that, it just needs nude adults, clothed children, and sex images to extrapolate from to then generate pedophilic content. That’s why they can’t easily stop generators from generating such content even on the ones they’ve confirmed don’t have it on their data set, and have to rely instead on keyword blocking.
You could perhaps go through a data set and remove all images of children, but then it’s likely someone would still find a way to loophole it (using keywords like gnomes or cherubs perhaps), or you could remove all nudity but it could also break the gen AIs capability to depict humans at all (as was the case with one of the later Stable Diffusion models, 3.0 I think).
You can chill on the witch hunting. I get you are understandably passionate on the subject but accusing anyone trying to find solutions of doing the opposite or being disingenous won’t help either.