A: all models are trained on something
2, you’re building your own straw man here. You’ve set up an extremely narrow condition under which this particular type of pedophilia is acceptable. Prove to me that that’s the norm, that it’s a typical use scenario, and that people looking at that crap are exclusively looking at loli, and not images meant to look like real people, and there’s a debate to be had there. But if you think any of that is true you’re lying to yourself. Sexualization of others is not going to happen in a vacuum under sterile conditions, it’s going to bleed in to real life.
Prove to me that removing this will not bleed into real life even more than it is not? You can’t either.
What I can prove is that Japan has csam cartoons for decades and they have less CSA per Capita than the USA. Is it possible that the Japanese know something we don’t? Who knows.
Can you prove to me that the AI trained models were done with real csam materials? If so, not reporting this to the FBI seems irresponsible.
Generative models does not work like that, if it were so, how do you explain that I can generate a picture of a purple six legged cat throwing lasers from the eyes in space?
In a very very very simplified way, the models are trained that from noise it de noises it until the image is “restored”. A part of the model learns to remove noise until a drawing of a child is restored, another learns to restore the image of a drawing of a nude woman. Basically you say to the model that from noise it has to restore the drawing of a nude child it combines the two proceses (also it is trained to combine things in a way that makes sense).
