I recall an article from, like, a year or two back, where an AI exec talked about how it was basically impossible to keep CSAM out of their training data, and I feel like that should've been a bigger red flag at the time.
add a skeleton here at some point
5 months ago