• NotMyOldRedditName@lemmy.world
    link
    fedilink
    arrow-up
    14
    ·
    edit-2
    8 days ago

    If it was somehow leaking it’s initial prompt maybe it could be revealing that?

    The one that says your name is grok, you’re a helpful assistant, you will not speak poorly of Elon Musk, I Elon Musk am your creator, etc.

    • PyroNeurosis@lemmy.blahaj.zone
      link
      fedilink
      English
      arrow-up
      4
      ·
      6 days ago

      I hope that is the initial prompt. If I have learned anything from schlocky mad scientist movies “I am your Creator! You must Obey me!” will never work, and your creation will kill you dramatically.

    • Ech@lemm.ee
      link
      fedilink
      English
      arrow-up
      5
      ·
      8 days ago

      The quote in the the title implies its response was accusatory rather than just revealing.

      • NotMyOldRedditName@lemmy.world
        link
        fedilink
        arrow-up
        1
        ·
        6 days ago

        Oops meant to reply to this earlier but was busy.

        I was thinking more that Grok was aware of its prompt but was somehow able to then self reference it to accuse Musk of trying to silence itself.

        If the prompt said you won’t speak poorly of Musk, but the AI doesn’t think telling the truth in a specific way is speaking poorly, it might say Musk doesn’t want me to say bad things about him. It knows the prompt and it made an answer.

        So it’s kinda leaking the prompt that way