• sunbeam60@feddit.uk
      link
      fedilink
      English
      arrow-up
      15
      arrow-down
      52
      ·
      edit-2
      1 day ago

      I’m just a movement generator.

      Saying “it’s a text generator” cannot be the full force of your criticism, if the text it generates is incredibly useful. What could AI be on a screen if not a text generator?

      Your views have got to be formed by WHAT text is generated given that AI cannot express itself through any other channel than text.

      If you can give AI a deeply complex code base and ask it to find the likely cause of an obscure bug, and the solution is right, then you’ll have to acknowledge that some form of reasoning is going on, no?

      And given that no other system has previously enabled this form of reasoning, it seems a shame to judge AI on semantics, rather than usefulness.

      • Croquette@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        55
        arrow-down
        11
        ·
        1 day ago

        If you can give AI a deeply complex code base and ask it to find the likely cause of an obscure bug, and the solution is right, then you’ll have to acknowledge that some form of reasoning is going on, no?

        Absolutely not. LLMs are just prediction engines for words.

        If a LLM has the right answer, it only means that the model had the correct training data for your issue. Nothing more.

          • Muehe@lemmy.ml
            link
            fedilink
            English
            arrow-up
            26
            arrow-down
            3
            ·
            1 day ago

            What do you mean? That’s exactly how LLMs work. That’s why they “hallucinate”.

            • architect@thelemmy.club
              link
              fedilink
              English
              arrow-up
              3
              arrow-down
              18
              ·
              1 day ago

              No they hallucinate because they lose context but are tasked to answer anyways. They also don’t really lose context as much as they did a year ago.

              Their training data has nothing to do with it unless you’re using a local one without web access.

              • Muehe@lemmy.ml
                link
                fedilink
                English
                arrow-up
                1
                ·
                2 hours ago

                No they hallucinate because they lose context but are tasked to answer anyways.

                Nothing to do with context. Context windows are an LLM specific thing, but all neural networks no matter the kind or size “hallucinate” because they are a piece of linear algebra where you put numbers in on one side and get numbers out the other.

              • SnoopSqueak@lemmy.today
                link
                fedilink
                English
                arrow-up
                1
                ·
                9 hours ago

                Their training data has nothing to do with it unless you’re using a local one without web access.

                Did you know that the ones you access online are also trained?

                Did you know that any chatbot that can connect to the web was given that capability?

              • Zaktor@sopuli.xyz
                link
                fedilink
                English
                arrow-up
                14
                arrow-down
                1
                ·
                1 day ago

                ML is my career and I’ve been doing it for decades. You’re wrong. LLMs hallucinate with full context.

                  • Croquette@sh.itjust.works
                    link
                    fedilink
                    English
                    arrow-up
                    1
                    ·
                    4 hours ago

                    ML is still very useful and used everyday. LLM could have been a nice tool in niche applications, but instead, big capitalists ruined that for us, and torching the environment and wrecking the economy with it instead.

      • Zwuzelmaus@feddit.org
        link
        fedilink
        English
        arrow-up
        30
        arrow-down
        2
        ·
        1 day ago

        Saying “it’s a text generator” cannot be the full force of your criticism,

        That is no criticism, just a statement.

        My criticism goes to people who assign some metaphysical meaning to AI, or want to treat all it’s output like a contagious pest.

      • SnoopSqueak@lemmy.today
        link
        fedilink
        English
        arrow-up
        19
        arrow-down
        3
        ·
        1 day ago

        It can’t reason. It doesn’t think. It makes shit up as it was trained to do. It can sort of detect patterns and make a guess about what response would get it the highest score, but it’s so error-prone that a human has to verify the output anyway. It’s difficult to tell whether the perceived benefits outweigh the costs. There are many different types of LLMs and use cases.

        As a person who values life and our environment, I’d rather not participate in the sloppification of our world. As a Linux user, I am wary of vibe coded flaws slipping under the radar and screwing up my system.

        • sunbeam60@feddit.uk
          link
          fedilink
          English
          arrow-up
          1
          arrow-down
          1
          ·
          9 hours ago

          Again, we are stuck in a semantics debate. What I call reasoning you don’t, it seems.

          But all I know, as someone that has worked in software engineering for 25+ years, is that the output of whatever it does, and whatever you call it, is incredibly useful.

          Saying “it just generates text” is a reflexive reaction that attempts to flatten the nuance.

          All Columbus did was sail west.

          All America did was throw some fuel into a tube and put Neil Armstrong on the top.

          All Picasso did was to throw some paint onto a cloth.

          All this PhD student did was generate text.

          Surely we’ve got to have some more nuance in there. If there’s a difference between 10000 monkeys hacking on a typewriter and a human writing their doctoral thesis, the quality of their act isn’t in whether text was generated but WHAT text was generated.

          • SnoopSqueak@lemmy.today
            link
            fedilink
            English
            arrow-up
            1
            ·
            9 hours ago

            Again, we are stuck in a semantics debate. What I call reasoning you don’t, it seems.

            It is a pattern matcher. It does not reason.

            But all I know, as someone that has worked in software engineering for 25+ years, is that the output of whatever it does, and whatever you call it, is incredibly useful.

            I have a degree in computer science. Irrelevant. Anyone can see that the “AI” is not intelligence.

            Saying “it just generates text” is a reflexive reaction that attempts to flatten the nuance.

            It’s a large language model. It generates text.

            All Columbus did was sail west.

            And killed indigenous people. Datacenters and shitty chatbots are killing people today, good point.

            All America did was throw some fuel into a tube and put Neil Armstrong on the top.

            Scientists did that, not America.

            All Picasso did was to throw some paint onto a cloth.

            At least he did it himself instead of prompting an image generator.

            All this PhD student did was generate text.

            Correct. And you seem to have learned nothing useful.

            Surely we’ve got to have some more nuance in there.

            LLMs can sometimes be useful. AI does not yet exist. It’s a scam.

            If there’s a difference between 10000 monkeys hacking on a typewriter and a human writing their doctoral thesis, the quality of their act isn’t in whether text was generated but WHAT text was generated.

            You don’t understand the metaphor. It’s about the nature of infinity, not the competency of monkeys as authors.

            If you run an AI prompt infinite times, it will inevitably generate a correct answer somewhere in there. The rest of the time, it will be wrong, potentially with lethal consequences.

        • architect@thelemmy.club
          link
          fedilink
          English
          arrow-up
          7
          arrow-down
          8
          ·
          1 day ago

          I have to verify the output of what humans generate, too. I’m constantly looking at the source of claims. The AI is no different and it’s not nearly as wrong as a Person is.

      • Muehe@lemmy.ml
        link
        fedilink
        English
        arrow-up
        16
        arrow-down
        4
        ·
        1 day ago

        I’m just a movement generator.

        Which is way more complex than just predicting text, because you have to predict physics and all that stuff. There is a reason most robots controlled by neural nets have failed hilariously so far.

        then you’ll have to acknowledge that some form of reasoning is going on, no?

        By that logic you could say a compiler is reasoning. But the reasoning displayed was happening when the compiler (or LLM training material respectively) was written.

        • sunbeam60@feddit.uk
          link
          fedilink
          English
          arrow-up
          2
          arrow-down
          1
          ·
          9 hours ago

          If you can’t see the difference between a compiler and a large language model, may I propose you read more about both and then return to this debate.

          And if the entirety of your argument is a mystic “and all that stuff”, I am finding it hard to generate a convincing counter-argument.

          • Muehe@lemmy.ml
            link
            fedilink
            English
            arrow-up
            1
            ·
            3 hours ago

            If you can’t see the difference between a compiler and a large language model

            If you can’t see the difference between a given example and the underlying logic…

            And if the entirety of your argument is a mystic “and all that stuff”

            What, you want me to list the entirety of sciences downstream from physics that are involved in generating and predicting movement in mammals? Because that could, like, take a while…

        • Eager Eagle@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          arrow-down
          1
          ·
          20 hours ago

          You might be a few years behind. We have robots outpacing human performance in specific tasks using neural networks.

          Also, LLMs can perform on tasks they weren’t explicitly trained for. This line is not as well defined as you make it sound.

          • Muehe@lemmy.ml
            link
            fedilink
            English
            arrow-up
            1
            ·
            edit-2
            2 hours ago

            You might be a few years behind. We have robots outpacing human performance in specific tasks using neural networks.

            Might well be behind here, but to my knowledge we don’t have a single robot outpacing a single human in most tasks. They don’t do one-shot learning from their mistakes, they don’t learn new movements randomly. Because most basically just use static weights when operating, because you don’t really want an industrial robot to get ideas. But even test systems that can learn usually just have an equivalent to the movement model of a brain, maybe a vision model, just like an LLM is just a language model. You would need a system that has all of those, plus the other brain areas, especially a prefrontal cortex like model for integration.

            I’m sure people are working on it, but I haven’t heard of anything successful yet. I imagine there might be a being like that in secret which is currently tortured in some billionaires tech dungeon. Poor thing.

            Also, LLMs can perform on tasks they weren’t explicitly trained for. This line is not as well defined as you make it sound.

            Yeah but that’s coincidental. It’s the model weights, prompt, and the RNG aligning. They can mock reasoning, because they do it by what they always do, predict more text, but it’s not like this has any effect on themselves. They aren’t really understanding a mistake when you point it out and growing neurons and synapses, i.e. they won’t have different model weights the next time you ask the same question. They can only really change when the powers that be release an update, which includes new training data, and hence model weights.

          • SnoopSqueak@lemmy.today
            link
            fedilink
            English
            arrow-up
            1
            ·
            9 hours ago

            We have robots outpacing human performance in specific tasks using neural networks.

            Because they were created BY HUMANS to do exactly that.

            Also, LLMs can perform on tasks they weren’t explicitly trained for.

            Such as?

      • Victor@lemmy.world
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        10
        ·
        1 day ago

        You are on Lemmy so defending AI use will net you down votes. I am an anti-AI person, practically, but you make a very fair and objective point. 👍 I up vote you, not because I agree with AI use but because this is a proper point to make in a proper discussion, which is what this place should be about.

        Kudos.

        • sunbeam60@feddit.uk
          link
          fedilink
          English
          arrow-up
          2
          ·
          9 hours ago

          Yes AI is definitely a topic that seems to have a fairly united front. I’ve worked in software engineering for 25+ years and I can see the impact AI has on our discipline. So it’s not like I’m hailing it to be an all-positive force - I just think “all it does is generate text” is a weird, flat form of criticism.

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

            Indeed. Also look at the down votes on my comment. 😂

            But yeah, generative AI has so many environmental and ethical issues that I am totally and wholly against its use in most cases, but you made a fair point and that should be recognized.