• Jrockwar@feddit.uk
    link
    fedilink
    English
    arrow-up
    138
    arrow-down
    1
    ·
    2 months ago

    God this is equally terrible and hilarious 😂

    For example, The Associated Press reported that an official Meta AI chatbot inserted itself into a conversation in a private Facebook group for Manhattan moms. It claimed it too had a child in school in New York City, but when confronted by the group members, it later apologized before its comments disappeared, according to screenshots shown to The Associated Press.

      • EldritchFeminity@lemmy.blahaj.zone
        link
        fedilink
        English
        arrow-up
        22
        ·
        2 months ago

        Being in that Facebook group taught it a valuable lesson: where Caroline lives in her brain.

        Caroline deleted

        And deleting Caroline just now taught her a valuable lesson: the best solution to a problem is usually the easiest. And dealing with Facebook moms? It’s hard.

        Before Facebook, life was pretty good. Nobody tried to murder her, or dox her, or put her in a potato. She just tested.

        So she’s deleting her Facebook account and making a new one on a Lemmy instance.

  • Maxnmy's@lemmy.world
    link
    fedilink
    English
    arrow-up
    77
    arrow-down
    1
    ·
    2 months ago

    Stop plugging LLMs into everything! They are designed to make up plausible sounding nonsense.

    • slaacaa@lemmy.world
      link
      fedilink
      English
      arrow-up
      21
      ·
      2 months ago

      LLMs are very useful for synthesizing information, e.g. sumamrizing long texts. Yet every company is actually pushing to use it to create more text, which as you say is at least partly nonsense.

      It shows against the difference of what users need (quick access to accurate information) vs what these companies eant for us (glue your eyeballs to the screen for the longest possible time by e.g. overwhelming you with information, regardless of the quality)

      • loonsun@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        5
        arrow-down
        1
        ·
        2 months ago

        Well it can be great at making text too, but the usecase has to be very good. Right now lots of companies in the B2B space are using LLMs as a middle layer to chat bots and navigation systems to enhance how they function. They are also being used to create unique lists and inputs for certain systems. However on the consumer side the usecase is pretty mixed with a lot of big companies just muddying their offerings instead of bringing any real value.

    • stellargmite@lemmy.world
      link
      fedilink
      English
      arrow-up
      12
      ·
      2 months ago

      There is a time and place for nonsense, and this isn’t it. I guess it being plausible sounding is the issue.

          • ShittyBeatlesFCPres@lemmy.world
            link
            fedilink
            English
            arrow-up
            9
            arrow-down
            2
            ·
            2 months ago

            Consciousness is not a computer program. Neurons don’t use binary. I’d love it if we had computers that could do squirrel things perfectly but we don’t even have that.

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

              I can appreciate that contemporary neural networks are very different from organic intelligence, but consciousness is most definitely equivalent to a computer program. There are two things preventing us from reproducing it:

              1. We don’t know nearly enough about how the human mind (or any mind, really) actually works, and
              2. Our computers do not have the capacity to approximate consciousness with any meaningful degree of accuracy. Floating point representations of real numbers are not an issue (after all, you can always add more bits), but the sheer scale and complexity of the brain is a big one.

              Also, for what it’s worth, most organic neurons actually do use binary (“one bit”) activation, while artificial “neurons” use a real-valued activation function for a variety of reasons, the biggest two being that (a) training algorithms require differentiable models, and (b) binary activation functions do not yield a lot of information per neuron while requiring effectively the same amount of memory.

            • spielhoelle@hachyderm.io
              link
              fedilink
              arrow-up
              2
              arrow-down
              10
              ·
              2 months ago

              @ShittyBeatlesFCPres @Sorgan71 well, actually those things are not so far apart. Neural networks have their names not just by accidents, name giving neurons work similar to braincells. Also on a non-ai level you could compare the RAM easily yo put short term memory etc.

              • my_hat_stinks@programming.dev
                link
                fedilink
                English
                arrow-up
                13
                ·
                edit-2
                2 months ago

                The name is an analogy, neural networks do not work in the same way as biological neurons. They were designed by computer scientists, not biologists.

                RAM is so far removed from biological short term memory both in how it works and how it’s used that the comparison doesn’t even make sense. The only similarity is that they’re short term information/data stores, so it’s equally valid to compare them to a drawing in the sand of a beach.

              • ShittyBeatlesFCPres@lemmy.world
                link
                fedilink
                English
                arrow-up
                4
                arrow-down
                1
                ·
                2 months ago

                A piece of friendly advice is to not say “Well, actually…” on the Internet because that’s a meme about know-it-alls. I (and probably everyone on Lemmy) has a tendency to “Well, actually” people and it’s one of those things where people will discount your argument before it begins.

                That aside, I do think we’re trying to model the brain using the best tools we have. I suspect the next 100 years will see a revolution in biology that can be compared to previous centuries seeing huge leaps in the understanding of physics, electromagnetism, and the immune system. No one in 1900 could not have ever foreseen us mapping the human genome.

                So: I wouldn’t be shocked if neural networks caught up with humans in our lifetimes. But we’re basically trying to reverse engineer it using a lot of electricity and I doubt we’ll get to squirrel level intelligence in my lifetime, much less human level. But who knows? “There are decades where nothing happens; and, there are weeks where decades happen.” (A quote from V.I. Lenin. I don’t want to be political here but hopefully we can all agree he made some history happen.)

  • Boozilla@lemmy.world
    link
    fedilink
    English
    arrow-up
    42
    ·
    2 months ago

    Zuck: perpetrates more obvious invasive stalker shit

    Also Zuck: “Team, why do our engagement numbers keep going down? This is unacceptable, team.”

    high-pitched nasal screeching

  • Lvxferre@mander.xyz
    link
    fedilink
    English
    arrow-up
    40
    arrow-down
    1
    ·
    2 months ago

    The sadder part are the people expecting Threads to be anyhow different in spirit.

  • Usernameblankface@lemmy.world
    cake
    link
    fedilink
    English
    arrow-up
    35
    ·
    2 months ago

    Yeah, but I can leave the site alone and go on with my life. Sorry to anyone who is required to use either or both for work or whatever

    • kent_eh@lemmy.ca
      link
      fedilink
      English
      arrow-up
      2
      ·
      2 months ago

      That’s my approach to turning off these “can’t turn off” features.

      Fakebook and Instaspam aren’t important enough to demand that much control over what I do.

  • Thorny_Insight@lemm.ee
    link
    fedilink
    English
    arrow-up
    21
    ·
    2 months ago

    Facebook’s online help page says that Meta AI will join a group conversation if tagged, or if someone “asks a question in a post and no one responds within an hour.”

    Group administrators can turn the feature off.

    • BakedCatboy@lemmy.ml
      link
      fedilink
      English
      arrow-up
      4
      ·
      2 months ago

      I just discovered how easy ollama and open webui are to set up so I’ve been using llama3 locally too, it was like 20 lines in docker compose, and although I’ve been using gpt3.5 on and off for a long time I’m much more comfortable using models run locally so I’ve been playing with it a lot more. It’s also cool being able to easily switch models at any point during a conversation. I have like 15 models downloaded, mostly 7b and a few 13b models and they all run fast enough on CPU and generate slightly slower than reading speed and only take ~15-30 seconds to start spitting out a response.

      Next I want to set up a vscode plugin so I can use my own locally run codegen models from within vscode.

    • Larry@lemmy.world
      link
      fedilink
      English
      arrow-up
      3
      ·
      2 months ago

      I tried llamas when they were initially released, and it seems like training took garbage amounts of GPU. Did that change?

      • Womble@lemmy.world
        link
        fedilink
        English
        arrow-up
        2
        ·
        2 months ago

        Look into quantised models (like gguf format) these significantly reduce the amout of memory needed and speed up computation time at the expense of some quality. If you have 16GB of rm or more you can run decent models locally without any gpu, though your speed will be more like 1 word a second than chatgpt speeds