How do we know the training data did not contain child sexual assault and that’s what’s used to gene...

afavour • last Monday at 8:49 PM • 2 replies • view on HN

How do we know the training data did not contain child sexual assault and that’s what’s used to generate “accurate” images?

“Just a multi million array of floats” could be applied to anything once it’s digitally encoded.

po1nt • yesterday at 2:13 AM

Someone could draw a 100% realistic depiction of a CSAM from memory. Would he be liable?

nekusar • last Monday at 9:18 PM

Its the courts job to determine burden of proof of that claim.

I'm also not a SCOTUS nor federal judge.

And if they deem it poisoned and illegal, then I guess we have more illegal numbers to deal with.

alt Hacker News