cross-posted from: https://sopuli.xyz/post/48033292

“Role tags were a formatting trick that became the security architecture and the cognitive scaffolding of modern LLMs,” the authors explain in a blog post. “We’ve shown that this architecture doesn’t survive into the model’s actual representations, and that such role confusion is linked to prompt injection.”

    • OwOarchist@pawb.social
      link
      fedilink
      English
      arrow-up
      5
      ·
      1 month ago

      The important part is they were easily able to trick the LLM into doing something it was told not to do.

      Which has serious implications for any LLM that’s ever put in charge of anything important or given access to any sensitive data it shouldn’t divulge. (Such as Instagram’s customer service AI, which hackers repeatedly prompt-injected into giving them access to other people’s accounts.)

      • southsamurai@sh.itjust.works
        link
        fedilink
        arrow-up
        3
        ·
        1 month ago

        Yeah, sorry. Was making a joke and it really needed a /s or /j because of the way I phrased it.

        But, thank you very much for the explanation! One of the beautiful things about lemmy is people like you stepping up with good information. Completely wonderful thing to do :)

      • southsamurai@sh.itjust.works
        link
        fedilink
        arrow-up
        1
        ·
        1 month ago

        Sure!

        Coca-Cola is made by a recipe that has been kept secret pretty much since its creation. It is colloquially referred to by just the word “Coke”.

        Cocaine is a hell of a drug. It is sometimes referred to by the word “coke”. However, the recipe for making it is dead simple and well known among the populace of the world, and very well known in areas where the plant it is made from is grown.

        At one point, Coca-Cola included the same plant product in its recipe, and is where the “Coca” part of its name comes from. Thus, the joke also referred to that shared history as a neat little package.

        Not really the best joke, tbh. But they can’t all be winners ¯\_(ツ)_/¯

        • ToastedRavioli@midwest.social
          link
          fedilink
          arrow-up
          2
          ·
          1 month ago

          Coca Cola actually still contains coca in it for the flavor, it just has the cocaine drug portion removed from it. They do this so that the drink still has the same flavor as it originally did, just without the extra kick

          They remove it and then sell it to pharmaceutical companies, making Coca Cola the largest producer and distributor of cocaine in the entire world.

          • southsamurai@sh.itjust.works
            link
            fedilink
            arrow-up
            2
            ·
            1 month ago

            I thought they had synthesized the flavoring! Kinda cool, but also kinda weird that they’re now coke dealers in both senses (again) lol.