News for hackers. Stuff that matters.
Re:How 'overworked, underpaid' humans train Google's AI to s... by michaelt on Saturday September 13, 2025 @01:22PM (#45233732)
Nobody's distributing a free dataset of child abuse, animal torture and terror beheading images, for obvious reasons.

There are some open-weights NSFW detectors [1] but even if your detector is 99.9% accurate, you still need an appeals/review mechanism. And someone's got to look at the appeals.

[1] https://github.com/yahoo/open_nsfw

Re:How 'overworked, underpaid' humans train Google's AI to s... by mallowdram on Saturday September 13, 2025 @01:59PM (#45234018)
All of this is so dystopian (flowers/beheadings) it makes K Dick look like a golden-age Hollywood musical. Are the engineers so unaware of the essential primate forces underneath this that cannot be sanitized from the events? You can unearth our extinction from this value dichotomy.
Re:How 'overworked, underpaid' humans train Google's AI to s... by alasarmas on Sunday September 14, 2025 @07:34AM (#45239066)
I mean, yes, my assumption is there exists an image / video normalization algorithm that can be followed by hashing the normalized value. There’s a CSAM scanning tool that exists that I believe uses a similar approach