

These models are overtrained. It’s incredibly likely the CSAM is “baked in” there somewhere, plus it’s abstraction in tokens and whatnot. There just needs to be a way to prove it mathematically or logically. That would be a good way to pivot into other things like copyright- if you’re a capitalist or something.
If we take some of the extreme AGI claims seriously, they’re all like CSAM experts, but don’t talk about it because they’re polite. That’s weird right?

Wherever it came from it worked. The invisible hand of capital definitely played it’s part.