Study shows AI image-generators being trained on explicit photos of children::Hidden inside the foundation of popular artificial intelligence image-generators are thousands of images of child sexual abuse, according to a new report that urges companies to take action to address a harmful flaw in the technology they built
False headlines about this took no time at all.
One dataset found suspected images, comprising approximately 0% of examples. Out of a bajillion. And immediately called the cops. But the headline acts like “scientific proof all AIs are fed these images!!!” Which has been the fantasy peddled by people who know less than nothing about this technology, and god fucking dammit, we’re gonna be explaining this forever.
An AI does not need pictures of Shrek riding a unicycle, to combine the concept of “Shrek” and “unicycle.” Satisfying multiple arbitrary labels is kinda the whole point. The fact it can combine “child” and “porn” is never going to stop being a thing, unless you completely scrub all examples of both those unrelated concepts.
And even that might not work.