• Kecessa
    link
    fedilink
    arrow-up
    23
    arrow-down
    6
    ·
    edit-2
    2 months ago

    Pigeon = edible bird

    Cleaning a bird > preparing a bird after killing it (hunting term)

    AI figured the “rescued” part was either a mistake or that the person wanted to eat a bird they rescued

    If you make a research for “how to clean a dirty bird” you give it better context and it comes up with a better reply

    • DannyBoy
      link
      fedilink
      English
      arrow-up
      27
      arrow-down
      4
      ·
      edit-2
      2 months ago

      The context is clear to a human. If an LLM is giving advice to everybody who asks a question in Google, it needs to do a much better job at giving responses.

        • HighlyRegardedArtist@lemmy.world
          link
          fedilink
          arrow-up
          3
          arrow-down
          2
          ·
          2 months ago

          I have to disagree with that. To quote the comment I replied to:

          AI figured the “rescued” part was either a mistake or that the person wanted to eat a bird they rescued

          Where’s the “turn of phrase” in this, lol? It could hardly read any more clearly that they assume this “AI” can “figure” stuff out, which is simply false for LLMs. I’m not trying to attack anyone here, but spreading misinformation is not ok.

          • Kecessa
            link
            fedilink
            arrow-up
            4
            ·
            edit-2
            2 months ago

            I’ll be the first one to explain to people that AI as we know it is just pattern recognition, so yeah, it was a turn of phrase, thanks for your concern.

            • HighlyRegardedArtist@lemmy.world
              link
              fedilink
              arrow-up
              1
              arrow-down
              2
              ·
              2 months ago

              Ok, great to know. Nuance doesn’t cross internet well, so your intention wasn’t clear, given all the uninformed hype & grifters around AI. Being somewhat blunt helps getting the intended point across better. ;)

          • Rivalarrival@lemmy.today
            link
            fedilink
            English
            arrow-up
            2
            ·
            edit-2
            2 months ago

            My point wasn’t that LLMs are capable of reasoning. My point was that the human capacity for reasoning is grossly overrated.

            The core of human reasoning is simple pattern matching: regurgitating what we have previously observed. That’s what LLMs do well.

            LLMs are basically at the toddler stage of development, but with an extraordinary vocabulary.

      • iAmTheTot
        link
        fedilink
        arrow-up
        7
        ·
        2 months ago

        I don’t think they are really “making excuses”, just explaining how the search came up with those steps, which what the OP is so confused about.

      • lunarul@lemmy.world
        link
        fedilink
        arrow-up
        5
        ·
        2 months ago

        I thought AI was great at picking up context?

        I don’t know why you thought that. LLMs split your question into separate words and assigns scores to those words, then looks up answers relevant to those words. It has no idea of how those words are relevant to each other. That’s why LLMs couldn’t answer how many "r"s are in “strawberry”. They assigned the word “strawberry” a lower relevancy score in that question. The word “rescue” is probably treated the same way here.

    • huginn@feddit.it
      link
      fedilink
      arrow-up
      4
      arrow-down
      1
      ·
      2 months ago

      Let me take the tag off my bird then snap it’s wings back together

      • bluewing@lemm.ee
        link
        fedilink
        arrow-up
        3
        ·
        2 months ago

        Bought in a grocery store - see squab - they are usually clean and prepped for cooking. So while the de-boning instructions were not good, the AI wasn’t technically wrong.

        But while a human can make the same mistake and many here just assume the question was about how to wash a rescued pigeon - maybe that’s not the original intent - what human can do that AI cannot is to ask for clarification to the original question and intent of the question. We do this kind of thing every day.

        At the very best, AI can only supply multiple different answers if a poorly worded question is asked or it misunderstands something in the original question, (they seem to be very bad at even that or simply can’t do it at all). And we would need to be able to choose the correct answer from several provided.