The problem is not one of intelligence, but of consistency. Even human operators would struggle with this sort of stuff. Though at least with them, they have em

The problem is not one of intelligence, but of consistency. Even human operators would struggle with this sort of stuff. Though at least with them, they have empathy and personal-bias which just makes the outcomes swing one way or the other even if they don't makes sense. AI struggles with this problem because it's a vague and poorly-defined one. That's why you have to take the only way out, either ban it all, or allow it all. For the latter, I'd echo what someone else in this discussion mentioned, let them self-censor what they don't like. Rather than trying to come up with one size fits all classification rule that really doesn't cater for anyone really. I would liken it to a multi-dimensional classification problem. And each person has their own classification. We're essentially trying to pigeon-hole a single delineation between acceptable/non-acceptable topics onto an entire population. It's borderline stupid behavior, and I have no idea why anyone thinks it'll please everyone all the time.