• Captain Aggravated@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    88
    ·
    6 days ago

    He’s not just a regular moron. He’s the product of the greatest minds of a generation working together with the express purpose of building the dumbest moron who ever lived. And you just put him in charge of the entire facility.

    –GLaDOS

  • frezik@midwest.social
    link
    fedilink
    arrow-up
    29
    arrow-down
    2
    ·
    6 days ago

    It’s worse than that. It’s right just often enough to make people think it’s right all the time.

    • sp3ctr4l@lemmy.zip
      link
      fedilink
      English
      arrow-up
      8
      ·
      5 days ago

      No way, the overconfident, incompetent, very often factually wrong narcissist sociopaths that run big corporations think that a chat bot that acts just like them is the best business investment in the history of mankind?

      I can’t believe it!

      The second best idea ever is using that chatbot to read all our emails for us, and then just have meetings for us, and send us the transcripts of all our AI doppleganger avatars making business plans based on hallucinating our actual personalities and knowledge sets, that’ll save so much time!

      (Not kidding about that, that is actually being worked on.)

      https://people.com/zoom-ceo-wants-to-use-ai-to-create-digital-clones-that-can-attend-meetings-for-you-8659456

  • SGforce@lemmy.ca
    link
    fedilink
    arrow-up
    22
    arrow-down
    3
    ·
    6 days ago

    For real. Asked chatgpt to help with a batch script. Literally the easiest thing it could possibly program. And guess what? Almost every example it gave was wrong. It seemed to know a lot about the command line program I was trying to run but it didn’t really understand what it was repeating to me.

    • Optional@lemmy.worldOP
      link
      fedilink
      arrow-up
      17
      ·
      6 days ago
      "Oh I'm sorry!  You're right mxzlptx.jar was deprecated in 2008.  Here's the actual script that will definitely work 100%."
      

      *INCORRECT BUZZER*

      • SGforce@lemmy.ca
        link
        fedilink
        arrow-up
        6
        ·
        6 days ago

        It kept fucking up double quotes and escape characters. Like, if it can’t get that right in a batch file it’s hopeless.

    • wise_pancake@lemmy.ca
      link
      fedilink
      arrow-up
      12
      ·
      6 days ago

      I like when you ask it to help, and it tells you to import a library that would solve your problem exactly, then you google it and it doesn’t exist.

      • ch00f@lemmy.world
        link
        fedilink
        arrow-up
        6
        ·
        6 days ago

        I once tried to use it to find the title of a book my wife enjoyed in her childhood based on her description of the plot. It described a possible book that could be written with that content and even suggested a title.

        Cool, thanks.

      • Optional@lemmy.worldOP
        link
        fedilink
        arrow-up
        13
        ·
        5 days ago

        Because it doesn’t understand any words at all. Everything is a web of probabilities that it can attach words to that are believable. That’s the magic.

        Facts, reason, intelligence - Nope.

        • They definitely have it working well enough that it can create coherent sentences that are entirely readable and understandable. Now if they could just make it so that the text it generates is actually factually accurate they might have a useful tool. I still wouldn’t call it intelligence though. Spicy auto-summarize still would not think and reason for itself.

      • ShadowRam@fedia.io
        link
        fedilink
        arrow-up
        9
        ·
        6 days ago

        It’s so scary that the top line is an AI response now in google.

        Lazy unaware people will just assume its correct,

    • zout@fedia.io
      link
      fedilink
      arrow-up
      2
      ·
      6 days ago

      Could be, but my (limited) experience with ChatGPT and PHP was quite different. This was in 2023, so it may have deteriorated since.

    • zout@fedia.io
      link
      fedilink
      arrow-up
      8
      ·
      6 days ago

      That would answer a lot of questions. But if Elon turned out to be Grok himself, it would answer even more.

    • snooggums@lemmy.world
      link
      fedilink
      English
      arrow-up
      7
      ·
      6 days ago

      If Musk claimed to be doing all the Grok answers I would assume it is really just the guy he paid to play Path of Exhile.

    • dual_sport_dork 🐧🗡️@lemmy.world
      link
      fedilink
      English
      arrow-up
      8
      arrow-down
      2
      ·
      edit-2
      6 days ago

      This is so, but LLM’s are the only most visible public facing ones and they’re the dumbshit technology that every company on Earth is now trying to shoehorn into every product. They’re what people see, interact with, and get pissed off by.

      Learning neural nets are good for various specific types of pattern recognition tasks, but trying to be an electronic butler who answers your questions ain’t one of them and it never will be.

      • kitnaht@lemmy.world
        link
        fedilink
        arrow-up
        5
        ·
        edit-2
        6 days ago

        LLM’s are the only public facing ones

        They aren’t at all. Mixamo uses AI to generate animated skeleton models that follow dance videos. Photoshop has “generative fill”, plenty of AI is public facing and used constantly besides LLMs. Many of these things have sped up workflows to a tremendous degree. One of my favorite music programs can use AI in order to replicate certain harmonics for instruments which would have taken me hours by hand before.

        • Fingolfinz@lemmy.world
          link
          fedilink
          arrow-up
          5
          ·
          6 days ago

          The audio side is incredible. I can basically sample the nonlinearities of audio gear that would usually cost fuck loads to own a rack of. It’s limited still but it really injects that grit you want from certain gear that algorithms can’t touch

        • dual_sport_dork 🐧🗡️@lemmy.world
          link
          fedilink
          arrow-up
          6
          arrow-down
          1
          ·
          6 days ago

          Well, okay, they’re the mainly public facing ones. Gemini, GPT, Copilot, whatever the hell Musk’s one is, Grok? All of that bullshit is just LLM’s, the ones nobody asked for and nobody wants. That’s what everyone is complaining about.

    • heavydust@sh.itjust.works
      link
      fedilink
      arrow-up
      4
      ·
      6 days ago

      You’re right that some AIs are useful, but we are literally flooded with “scam as a service” web sites that sell LLMs, and they are hyped by people who can’t code or do anything useful in life who tell us that we are gatekeeping knowledge, and those people would be very happy if we were all unemployed.

  • snooggums@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    6 days ago

    I love how when I try to search for a setting I can’t find the stupid AI answer is always “go to settings, find the setting and turn it off” when that isn’t a setting and never was on the software. It just spouts out to do the thing I can’t find.

    Generally the first search response below the AI slop is a five year old reddit post pointing out that the setting doesn’t exist or the real directions from the official support page that are completely different.

    • Optional@lemmy.worldOP
      link
      fedilink
      arrow-up
      6
      ·
      6 days ago

      Samesies.

      It’s comically tragic how much venture capital they destroyed to make something that sucks so bad no one wants it.

  • IninewCrow@lemmy.ca
    link
    fedilink
    English
    arrow-up
    9
    arrow-down
    1
    ·
    6 days ago

    But we already have people that can do that

    Tech Bro: … yeah, but ours can do it billions of times faster and more often!