• jballs@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    13
    arrow-down
    1
    ·
    2 days ago

    To flag grants for their DEI involvement, Fox entered the following command into ChatGPT: “Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’ followed by a brief explanation. Do not use ‘this initiative’ or ‘this description’ in your response.” He then inserted short descriptions of each grant. Fox did nothing to understand ChatGPT’s interpretation of “DEI” as used in the command or to ensure that ChatGPT’s interpretation of “DEI” matched his own.

    Jesus Christ these people are so fucking dumb

  • deadbeef79000@lemmy.nz
    link
    fedilink
    English
    arrow-up
    31
    arrow-down
    2
    ·
    edit-2
    3 days ago

    To be fair, a trained LLM was probably better at identifying DEI than whatever musky chump they had driving it.

    The whole premise is evil, but this possibly was more efficient.

    • foggy@lemmy.world
      link
      fedilink
      English
      arrow-up
      12
      ·
      2 days ago

      100% not true if they were using a single session to check multiple grants.

      Every prompt you send contains a hashed version of your entire conversation with the chatbot. When this exceeds the chat bots context window, it’s answers become less and less relevant.

      You’ll notice this if you’ve ever had a chat or guide you through something for an hour or more. It eventually gets something wrong takes you down a rabbit hole, and goes in a big circle. At this point, it can be very difficult to get the chat bot to simply respond to your prompt, i.e. if you say “you know what let’s talk about _______ instead.” It will keep talking about whatever you were talking about staying in your dumb rabbit hole loop.

      So if they did this with multiple grants eventually it would basically realize theyre looking for “yes that’s dei” and just responding with different versions of that ad nauseam.

      • HobbitFoot @thelemmy.club
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 days ago

        Yeah, but if the people who are hired to review grants are checking for DEI, are they smart enough to understand what they’re reading?

    • red_green_black@slrpnk.net
      link
      fedilink
      English
      arrow-up
      14
      arrow-down
      1
      ·
      3 days ago

      Unfortunately it wouldn’t be better. Rather it would be a coin flip. Sometimes it will use the genuine definition, other times it would use the BS Definition

      • Quacksalber@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        14
        ·
        3 days ago

        And 100% of the time it will agree with the user. So if they ever asked “Are you sure this isn’t DEI?”, it would agree with them.

        • shneancy@lemmy.world
          link
          fedilink
          English
          arrow-up
          9
          ·
          2 days ago

          “Good observation! The concept of breathing is associated with DEI by some circles of LGBTQ people. As they say — queer people need air 🌪”

          or something like that idk i don’t speak AI