Mitigating Skeleton Key, a new type of generative AI jailbreak technique

Interesting jailbreak, though I think the focus on jailing AI is a bit overblown. ChatGPT is becoming very difficult to use now, it's so hard not to step on its long toes if you have hobbies slightly outside the American mainstream lifestyle. Being from Europe and into some more kinky stuff this means I run into this a lot. For that reason I don't think this is something that should be overly mitigated.

It raises an interesting question too: not everyone has the same sensibilities. Even mainstream European ones are very different than American ones (here sexual content can go much further in the mainstream - I have friends who causally wear outfits that in America would only fly in the strip club, and violence is much less ok than in America) In order to train your model with your specific ones you need an uncensored model.

I think right now all these AI companies are viewing this "Responsible AI" mostly as a risk to their business and are avoiding news headlines that might trigger overzealous regulators. I don't think they really care, just like they don't take care about being "green".

So hopefully as AI becomes more established this kind of hyperfocus on AI censorship dies down a bit.