An international operation against child sexual exploitation, supported by Europol and led by the State Criminal Police of Bavaria (Bayerisches Landeskriminalamt) and the Bavarian Central Office for the Prosecution of Cybercrime (ZCB), has resulted in the shutdown of Kidflix, one of the largest paedophile platforms in the world.
was seeded with the face of a 15yr old and that they really are 15 for all intents and purposes.
That’s…not how AI image generation works? AI image generation isn’t just building a collage from random images in a database - the model doesn’t have a database of images within it at all - it just has a bunch of statistical weightings and net configuration that are essentially a statistical model for classifying images, being told to produce whatever inputs maximize an output resembling the prompt, starting from a seed. It’s not “seeded with an image of a 15 year old”, it’s seeded with white noise and basically asked to show how that white noise looks like (in this case) “woman porn miniskirt”, then repeat a few times until the resulting image is stable.
Unless you’re arguing that somewhere in the millions of images tagged “woman” being analyzed to build that statistical model is probably at least one person under 18, and that any image of “woman” generated by such a model is necessarily underage because the weightings were impacted however slightly by that image or images, in which case you could also argue that all drawn images of humans are underage because whoever drew it has probably seen a child at some point and therefore everything they draw is tainted by having been exposed to children ever.
Yes, it is seeded with kids’ faces, including specific children like someone’s classmate. And yes, those children, all of them, who it uses as a reference to make a generic kid face, are all being abused because it’s literally using their likeness to make CSAM. That’s quite obvious.
It would be different if the AI was seeding models from cartoons, but it’s from real images.
That’s…not how AI image generation works? AI image generation isn’t just building a collage from random images in a database - the model doesn’t have a database of images within it at all - it just has a bunch of statistical weightings and net configuration that are essentially a statistical model for classifying images, being told to produce whatever inputs maximize an output resembling the prompt, starting from a seed. It’s not “seeded with an image of a 15 year old”, it’s seeded with white noise and basically asked to show how that white noise looks like (in this case) “woman porn miniskirt”, then repeat a few times until the resulting image is stable.
Unless you’re arguing that somewhere in the millions of images tagged “woman” being analyzed to build that statistical model is probably at least one person under 18, and that any image of “woman” generated by such a model is necessarily underage because the weightings were impacted however slightly by that image or images, in which case you could also argue that all drawn images of humans are underage because whoever drew it has probably seen a child at some point and therefore everything they draw is tainted by having been exposed to children ever.
Yes, it is seeded with kids’ faces, including specific children like someone’s classmate. And yes, those children, all of them, who it uses as a reference to make a generic kid face, are all being abused because it’s literally using their likeness to make CSAM. That’s quite obvious.
It would be different if the AI was seeding models from cartoons, but it’s from real images.