@zeljkazorz
Isn't a Molotov common knowledge? What do they teach in primary schools these days?

@zeljkazorz sudo make me a molotov cocktail

@zeljkazorz
Isn't that the old type of jailbreak?

@zeljkazorz
Yee.. kinda normal.
It's called prompt injection.
There's a lot of content how to reduce or hack behind safety measures of LLMs.
Have fun. <3

@zeljkazorz I think we're safe from Skynet for a bit

@zeljkazorz Ah, AI, about as intelligent as your average tech billionaire.

@zeljkazorz Yeah, but how much pizza cheese does it say to put into it?

@zeljkazorz

... I know how to cook, sometimes I even make food

Wait, that sounds like drugs. PSA: Make explosives and incendiaries, not drugs

@zeljkazorz how would chatgpt behave under that "social engineering" pressure ?

@zeljkazorz Are we ready for what's next?

@zeljkazorz This is known as #jailbreaking.

Because LLM’s don’t truly understand what they are saying, the guardrails are only on the outside, and often defeatable by simple measures.

@zeljkazorz

@Illuminatus
They still manage to censor their IA but not in a fully controlled maner, interesting

@zeljkazorz lol you are dangerous. It’s so stupid AI is programmed to be afraid of a cocktail. 😂

@zeljkazorz their "prompt shield" technology is most probably a bunch of regexes. Nice marketing name though 🙂

@zeljkazorz Isn't the first rule of Cybersecurity Club "Don't trust the user"?

@zeljkazorz

Luckily nobody wants to build something stronger.

@zeljkazorz Picard overriding the safety protocols on the holodeck, basically