BTW the models for how you'd do moderation if you weren't allowed to let bad stuff through are porn sites. They tend to have constraints like the ones Nadine apparently wants to put on ever… (View Post)

I think that makes my point? It's easy to make something that works some of the time against an adversary who isn't trying to defeat it, but it's impossible to make something that work… (View Post)

The dishwasher problem is easier because the dishes aren't constantly inventing subtle new behaviours to avoid being washed. (View Post)

No such AI exists. It's basically an impossible problem because language and culture evolve in real time to avoid censorship. What you can do that can mostly work is run everything through a mod… (View Post)

I don't think it's just civility, it's also about incentives. If you let murdering politicians potentially change policy, you encourage people to murder politicians. Politicians prefer… (View Post)