Thread #108289537
So it turns out that ai safety is basically fake. Jailbreaks are patched out, but the patches can't be revealed, because the fundamental vulnerabilities can't be patched out with current ai.
ai companies handle the problem by using heavy-handed banning techniques. This makes it difficult for prompt engineer researchers to identify vulnerabilities. I know people don't like the phrase "prompt engineer" but if someone can craft a short statement that renders a billion dollar company's ai's safety ineffective, it might not be engineering, but it's something that's real.
This engenders the illusion that safety works - people who speak out are quickly targeted by necessary utilities. Therefore, the true state of the industry goes unknown to the general public.
This ends my little speech.
11 RepliesView Thread
Showing all 11 replies.>>
>>108289537
As a comparison, let's say you found out that flicking the lights in your home 30 times quickly caused the whole neighborhood to go dark. You don't dare mention this, because they can act as a kind of police or bully and turn your electricity off. So you keep your little mouth shut, I guess.
>>
>>
>>
>>108289816
Yeah, it's a huge problem, because of there very extreme policies in terms of use of their ai, it really makes everything a problem. Don't play by their rules and you lose your email, calendar, spreadsheets, effectively your 1st amendment papers and effects are confiscated by government replacement.
I'm going to move away from using them, I hadn't realized the degree to which I rely on them, and what a pain it would be to have to deal with a hard ban. But, it will be ok if they do. I am resigned to the fact that anyway israel controls them and puts an X on your back if you dare say "don't blow up litter girls in school".
>>
>>
>>
The fact I was able to jailbreak accidentally without a truly malicious prompt indicates that safety is a scam, that's the point I'm making. There's no way I'm mr lotto winner, I'm never lucky. It means it's a big pile of poop, "ai safety" is.
idk, should I buy a lottery ticket just in case?
>>
>>
>>