• Victor@lemmy.world
    link
    fedilink
    English
    arrow-up
    9
    ·
    11 hours ago

    They know whatever science they were trained on. They could be trained on 1 + 1 = 3 and report that as the truth.

    • whalesnrainbows@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      51 minutes ago

      They could be trained on 1 + 1 = 3 and report that as the truth.

      And they do, chatbots are dogshit at math and measurements.

      I know a parent who asked a chatbot how much Tylenol to give her toddler, and it recommended that she give her toddler six times the amount actually prescribed for toddlers.

          • willington@lemmy.dbzer0.com
            link
            fedilink
            English
            arrow-up
            3
            arrow-down
            1
            ·
            54 minutes ago

            Eh? Whatever.

            What you said wasn’t wrong but it serves to shut down the conversation. It redirects the attention into a tangent. So you are not disputing anything your interlocutor says but simply attempt to remove the oxygen from the thoughtstream.

            It’s a ruthless technique of manipulation.

            I don’t give that much of a shit online in this context, but if I saw that behavior among my social contacts, I would instantly correct all my prior assessments of the ruthless technique user.

            • Nouvellalia@lemmy.world
              link
              fedilink
              English
              arrow-up
              1
              arrow-down
              1
              ·
              19 minutes ago

              Oh, you’re
              you’re digging deeper huh?
              Ok

              You are wildly reactionary, and clearly trapped by your concept of your own intelligence, a hell of a prison, because you never want to leave.

              I was saying that the problem isn’t intractable, and in fact we already know the solution. “The solution is to educate/train people/AI on truthful information” would have been an easy continuation.

              The problem isn’t that you didn’t see this. Single sentence responses are not information dense. The problem is, well, just look at what you decided to post instead.

              If I was one of your social contacts, it would clearly be because of some sort of desire not to embarrass you socially, and I would be relieved at your ghosting.

              Again, how embarrassing 🫂

      • Victor@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 hour ago

        You don’t even have to manipulate all the senses of people to make them say all kinds of crazy shit.

        • Nouvellalia@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          arrow-down
          1
          ·
          1 hour ago

          Not even close. You literally only have to feed them words, and they will deny everything else they experience.

  • Humanius@lemmy.world
    link
    fedilink
    English
    arrow-up
    93
    ·
    1 day ago

    This has been a consistent problem with AI chatbots. How do people keep being surprised by this?

    Is climate change real?
    - Yes climage change is real, and here is the proof.
    I don’t think that’s right.
    - You’re absolutely right! It’s all a hoax and a conspiracy!

    • Fallibilist@feddit.uk
      link
      fedilink
      English
      arrow-up
      7
      arrow-down
      1
      ·
      15 hours ago

      That used to be the case but in my experience hasn’t been for years.

      Not too long ago Grok scolded me for racial stereotypes when I mentioned the unfortunate coincidence that my first ever Gypsy customer is also the only one ever to refuse to pay the invoice. Even Mecha Hitler refuses to go along with racism even if you ask it to. It has repeatedly told me I’m wrong. Even when it agrees it usually brings in more nuance to demonstrate that it’s not that simple.

      • KairuByte@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        7 hours ago

        To be fair, I don’t think most mean “if you say it’s wrong it agrees” it’s more along the lines of “if you throw bullshit “science” at it and tell it it is wrong, it will go along with it”

      • Zarobi@aussie.zone
        link
        fedilink
        English
        arrow-up
        2
        arrow-down
        1
        ·
        10 hours ago

        I tend to find that it is weak to emotional manipulation. If you say things like “trust me” or even something like “it hurts me when you don’t believe me - why aren’t you on my side?” you get immediate capitulation. Though I felt kind of bad being a narcissist even to an LLM that deserves it

    • expatriado@lemmy.world
      link
      fedilink
      English
      arrow-up
      43
      arrow-down
      1
      ·
      1 day ago

      are going to agree with me always? that a toxic trait

      yes, you’re right, i shouldn’t do that

  • DarkCloud@lemmy.world
    link
    fedilink
    English
    arrow-up
    62
    arrow-down
    3
    ·
    1 day ago

    It’s almost like they’re stocastic parrots just being directed into different pools of content depending on what users say - or something.

  • gandalf_der_13te@feddit.org
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    7
    ·
    9 hours ago

    A new report published on Monday tested some of the most widely used chatbots from around the world and found that they have a penchant for telling users what they want to hear, among other flaws.

    has any of you actually experimentally checked this? because it’s not true

  • TrackenDaKracken@lemmy.world
    link
    fedilink
    English
    arrow-up
    27
    ·
    1 day ago

    Better to link to the source, than NYT:

    https://unesdoc.unesco.org/ark:/48223/pf0000400048?posInSet=13&queryId=N-c7820124-26f6-40cc-8c66-0375e501acaf

    Tested on a single measure, such as whether a carbon tax will reduce projected warming, the models generally answer correctly, reaching up to 94.5% accuracy. Yet when two policies are combined, accuracy collapses to 33.8%. Asked whether two combined policies achieve more, less, or the same, models scored, identical to random guessing.

    Even the largest frontier systems show no improvement on this task. Since meaningful climate action relies on overlapping policies working together, AI outputs cannot replace validated simulation models or expert analysis.

  • ratjefe@lemmy.today
    link
    fedilink
    English
    arrow-up
    11
    arrow-down
    2
    ·
    1 day ago

    yeah so I can remember using chatgpt when it was a brand new thing … it was too real for actual people and they got angry … so they’ve been pushed to censor it over time to placate average people … now, you can pay for access to the realness or you can enjoy the pedantic sycophancy … remember when this all happened to TV? boy, that was fun