3 ms·
That's one of the few good aspects of a legendary fuck-up like this: you learn about people defending it. People defending CSAM should go to hell, fast. But ar
by firebaze 5y ago
That's one of the few good aspects of a legendary fuck-up like this: you learn about people defending it.
People defending CSAM should go to hell, fast. But are we already done destroying all low-hanging fruits? Did we stop Johnny Savile? Did we put all clerical actors behind bars? Did we extinguish the child porn network behind Marc Dutroux (https://en.wikipedia.org/wiki/Marc_Dutroux https://en.wikipedia.org/wiki/Marc_Dutroux)?
And even if we did, would that be enough of an excuse to implicitly accuse anyone? My spouses' family (well-off, so using iDevices) took photos of their young age kids playing, partially naked at the sea. They are now frightened if their photos could be stolen by someone and marketed as child porn.
So unbelievable.
- notJim 5y ago> They are now frightened if their photos could be stolen by someone and marketed as child porn. That sounds bad, someone high up at Apple should do an interview clarifying that that's not what's happening!
- cwkoss 5y agoIf you don't think this is possible, you are not understanding the technical implementation they announced. Perceptual hashes on chunks of images will yield false positives.
- notJim 5y agoI'm not sure what you mean by chunks of images. My understanding is that it's a perceptual hash meant to determine if it's the same image, with minor adjustments like cropping or adjusting the colors. There is no AI/ML classification running. Apple says false positives occur at a rate of one in a trillion or something like that. Someone elsewhere in the thread calculated that this would not be frequent enough for false positives to trigger a manual review.
- cwkoss 5y agoThe perceptual hashing are being applied to a chunk of the image - not the whole image but some part of it. These chunks contain less entropy than the whole image so necessarily have higher chance of triggering false positives than whole-image comparisons.