• Humanius@lemmy.world
    link
    fedilink
    English
    arrow-up
    71
    ·
    5 days 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!

  • LeftReddit2@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    4 days ago

    Yes, these things are just algorithms at their core. They prioritize what they think you want to hear, not the truth.

    This is grossly oversimplyfying it, but still.

  • DarkCloud@lemmy.world
    link
    fedilink
    English
    arrow-up
    42
    arrow-down
    3
    ·
    5 days ago

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

  • Victor@lemmy.world
    link
    fedilink
    English
    arrow-up
    14
    ·
    4 days 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
      5
      arrow-down
      1
      ·
      4 days 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.

      • Victor@lemmy.world
        link
        fedilink
        English
        arrow-up
        2
        ·
        4 days ago

        Yeah I almost took the AI answer for the truth once when I (inadvertently) asked it to convert between different units of speed. The AI was way off, by like double.

        It seems here like it converted 60,000 feet into kilometers, but then multiplied that by 100 instead of by 60 to go from per minute to per hour.

        Love the little disclaimer at the bottom of the image. 🙃

      • Rezoie@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        4 days ago

        Tbh maths and measurements is a simple problem to solve, nowadays every llm uses python to calculate stuff instead of simply telling you the answer from memory and the same goes for measurements, it can just google it.

      • Gpennyhigh@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        4 days ago

        For something that important they should check other sources, and the bot should cite sources. For important things people should know how to use tools. Don’t assume any web result is right unchecked, whether old style or llm. It’s awareness any adult should have, what tool to use and when, and to double check for important things. The med bottle probably says something about children as well. Misinformed people use things wrong all the time and adults should understand all this, especially if a child is in their care. Should need to pass tests like such basic knowledge of living in the world to have a child, but don’t and see the damage done.

  • TrackenDaKracken@lemmy.world
    link
    fedilink
    English
    arrow-up
    20
    ·
    5 days 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.

  • gandalf_der_13te@feddit.org
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    5
    ·
    4 days 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