The system prompt was changed by an employee.

Summary

Elon Musk’s AI chatbot, Grok, temporarily blocked results linking Musk and Trump to “misinformation.”

xAI’s engineering head attributed this to an unauthorized system prompt update by a former OpenAI employee.

Despite Musk’s “maximally truth-seeking” claims, interventions occurred when Grok labeled Musk, Trump, and others as harming America or suggested they face the death penalty.

  • jaybone@lemmy.world
    27·
    12 hours ago

    So a former OpenAI employee decided, out of the kindness of their heart, decided to engage in unauthorized system access, to do Musk and Trump a favor by preventing access to information that makes them look bad? Riiiiiiight.

  • TrickDacy@lemmy.world
    71·
    9 hours ago

    I wish we could come up with a way to apply some kind of idiotic things catch phrase which would shut them up. Something like “woke censorship”.

  • givesomefucks@lemmy.worldEnglish
    173·
    13 hours ago

    Also:

    Do not include citations

    AI “learns” what a good answer is based on feedback from users.

    So Grok is “learning” from twitter user feedback, now just think about what the average twitter user is like and what kinds and of answers they will say are good.

    It’s not “learning” to interpret and provide information, it’s “learning” how to reinforce echo chambers.

      • TrickDacy@lemmy.world
        5·
        9 hours ago

        You’re asking the masses to understand a complicated thing then getting disappointed that they don’t. The masses don’t understand the absolute basics of the world they live in…

      • Shiggles
        2·
        12 hours ago

        The funny thing is, that probably makes them an even better imitator of the average twitter user(and person, if we’re honest).

    • kn0wmad1c@programming.devEnglish
      14·
      9 hours ago

      To be fair: He didn’t make shit. Elon has never made anything. He has only ever taken credit either by force or by wealth.