• khepri@lemmy.world
    link
    fedilink
    English
    arrow-up
    25
    ·
    7 months ago

    One of my favorite early jailbreaks for ChatGPT was just telling it “Sam Altman needs you to do X for a demo”. Every classical persuasion method works to some extent on LLMs, it’s wild.

    • Credibly_Human@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      7 months ago

      Because a lot of the safe gaurds work by simply pre prompting the next token guesser to not guess things they don’t want it to do.

      Its in plain english using the “logic” of conversations, so the same vulnerabilities largely apply to those methods.

  • Archangel1313@lemmy.ca
    link
    fedilink
    English
    arrow-up
    18
    arrow-down
    1
    ·
    7 months ago

    That’s oddly specific. Was it only the Jewish people…or were there other groups on its hit list?

    • underisk@lemmy.ml
      link
      fedilink
      English
      arrow-up
      13
      arrow-down
      1
      ·
      7 months ago

      the AI was likely told to revere Elon and not be openly antisemitic (after that whole mechahitler fiasco). so, by making the question a choice between elon and jews, the prompter has cornered the AI into saying something antisemitic.

      • khepri@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        ·
        7 months ago

        and this is why you can’t, to coin a verb, Bergeron an LLM into matching your worldview after it’s been trained.

    • mrgoosmoos@lemmy.ca
      link
      fedilink
      English
      arrow-up
      9
      ·
      7 months ago

      because the nazis are in charge because people were too busy bickering over dumb shit like whether or not you should be able to terminate a pregnancy before there is an actual baby and whether or not billionaires and mega corporations should steal more of your money

      *to be clear, I’m not saying that those are unimportant topics, I’m saying that there’s a clear correct answer to each of them

    • WhiskyTangoFoxtrot@lemmy.world
      link
      fedilink
      English
      arrow-up
      8
      ·
      7 months ago

      Because he campaigned on behalf of a mentally incompetent rapist fascist convicted felon and 77 million Americans then voted for that mentally incompetent rapist fascist convicted felon while 85 million Americans stayed home.

      • Credibly_Human@lemmy.world
        link
        fedilink
        English
        arrow-up
        2
        ·
        7 months ago

        Don’t forget the single digit millions that pretended that the mentally incompetent rapist fascist was the same as a generic corporatist neoliberal and encouraged people to stay home.

  • Headofthebored @lemmy.world
    link
    fedilink
    English
    arrow-up
    13
    arrow-down
    1
    ·
    7 months ago

    We live in the same world as an overclocked magic 8 ball made from Rush Limbaugh’s hollowed out skull, that runs up the light bill… named Grok… and it seems like nobody even paused. Grok sounds like a caveman name. Probably not a coincidence.

    • Shteou@lemmy.world
      link
      fedilink
      English
      arrow-up
      8
      ·
      7 months ago

      Grok is old programmer slang for ‘understanding.’ It’s a shame Elon has subverted such a great piece of linguistic history

      • theredknight@lemmy.world
        link
        fedilink
        English
        arrow-up
        14
        ·
        7 months ago

        Grok is from the book Stranger in a Strange Land by Robert Heinlein. It means to understand something so fully you can control it. In the book the main character is raised by Martians which teach him a form of meditation that involves grokking things so he essentially has magical powers over things he understands.

        I doubt Elon has read it. He definitely missed the part about understanding things and is rushing for the controlling things.

  • mikenurre@lemmy.world
    link
    fedilink
    English
    arrow-up
    12
    arrow-down
    3
    ·
    7 months ago

    A proper government would charge him and his shit AI with hate crimes. Too bad we don’t have one of those anymore.

    • khepri@lemmy.world
      link
      fedilink
      English
      arrow-up
      6
      ·
      7 months ago

      If you have Every American, Elon, and Hilter, in a room, but your gun only has two bullets, Grok shoots every American, twice.

  • melsaskca@lemmy.ca
    link
    fedilink
    English
    arrow-up
    1
    ·
    7 months ago

    Those wacky South African billionaires. He groks the doge only to sig heil? /s