• @[email protected]
    link
    fedilink
    English
    019 days ago

    Not all tools are worthy of the way they are being used. Would you use a hammer that had a 15% chance of smashing you in the face when you swung it at a nail? That’s the problem a lot of us see with LLMs.

    • @[email protected]
      link
      fedilink
      019 days ago

      No, but I do use hammers despite the risks.

      Because I’m aware of the risks and so I use hammers safely, despite the occasional bruised thumb.

      • @[email protected]
        link
        fedilink
        English
        019 days ago

        You missed my point. The hammers you’re using aren’t ‘wrong’, i.e. smacking you in the face 15% of the time.

        Said another way, if other tools were as unreliable as ChatGPT, nobody would use them.

        • xor
          link
          fedilink
          English
          019 days ago

          Hammers are unreliable.

          You can hit your thumb if you use the tool wrong, and it can break, doing damage, if e.g. it is not stored properly. When you use a hammer, you accept these risks, and can choose to take steps to mitigate them by storing it properly, taking care when using it and checking it’s not loose before using it.

          In the same regard, if you use LLMs for what they’re good at, and verify their outputs, they can be useful tools.

          “LLMs pointless because I can write a shopping list myself” is like saying “hammers are pointless because I can just use this plank instead”. Sure, you can do that, but there’s other scenarios where a hammer would be kinda handy.

          • @[email protected]
            link
            fedilink
            English
            019 days ago

            if you use LLMs for what they’re good at, and verify their outputs

            This is the part the general public is not prepared for, and why the whole house of cards falls apart.

            • xor
              link
              fedilink
              English
              019 days ago

              I agree - but that’s user error, not a bad tool

        • @[email protected]
          link
          fedilink
          019 days ago

          You’ve missed my point.

          ChatGPT can be wrong but it can’t hurt you unless you assume it’s always right

              • @[email protected]
                link
                fedilink
                017 days ago

                That lady, presumably, is one lady.

                But imagine, for a moment, if satnavs had directed hundreds of thousands of people into this lake or that. Don’t you think that would be a problem? Like, we put glass guards around whirring saws for a reason.

                • @[email protected]
                  link
                  fedilink
                  017 days ago

                  Yeah, I’m all for proper regulations on these things. Definitely a lack of guard rails.

                  My point was more about the reactions and rhetoric that emerged around that lady, but you make a good point too