• AmbiguousProps@lemmy.today
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    1
    ·
    edit-2
    5 months ago

    Have you seen GPT4o (which was just a way for OpenAI to get away with destroying performance and accuracy to save money)? Or even the kneecapped GPT4 these days? It’s actually terrible compared to other LLMs.

    • Eheran@lemmy.world
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      1
      ·
      edit-2
      5 months ago

      I have used it, yes. It is no god, but holy cow it knows how to program an ESP32 to throw a web surface at people (so both backend and frontend) which I would have never learned to do. It also knows how to code a 2D heat transfer simulation in Python. Or help me find errors and much faster ways in my rotary kiln simulation. It also helps me understand this or that concept without googling and reading 10 pages to understand the one specific detail I need to know. It is like using Excel vs. a simple calculator.

      And yes, I think GPT4 from like a year ago was even more powerful. But 4o now also seems pretty good again. Which LLM is even better?

      • floquant@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        5
        arrow-down
        2
        ·
        edit-2
        5 months ago

        which I would have never learned to do.

        it knows how to code

        Please take a step back and reconsider what you think your mind can do vs what LLMs can. The ability to understand is what separates us from machines, at least for now. Not saying AI is bad, but it’s important to keep in mind what it is, and what we are. Also, fuck OpenAI, run local models if you can.

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

          What is your point? I do not have time to learn everything about every topic and do it all by hand too. This tool saves me time. I do not have to learn the intricate details of how to code a frontend with the fucking retarded functions and what not. Let alone the matching backend. I just need to know what I want and then get it, even if it takes a few tries and debugging.

          • shiftymccool
            link
            fedilink
            English
            arrow-up
            3
            ·
            edit-2
            5 months ago

            I do not have to learn the intricate details of how to code a frontend with the fucking retarded functions

            I hope this isn’t your job…

            If you don’t know what you’re doing, you have no idea if chatpt is doing it right. Spoiler alert, it isn’t. Chatgpt RARELY gets things even remotely right, it’s worse than a junior dev in that you have to keep an eye on it to make sure it’s not doing something super-stupid. This requires that you know at least more than an llm, which requires that you learn. Llms should be used as a convenience, not a replacement for knowledge.

            Source: full-stack developer for 20+ years

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

              Both frontend and backend work. The same result one could expect if I had spend the time to learn it and then debug for this one specific issue for ages. What does not work is the “set” button as well as the mDNS. The first I am sure a good program could fix, the other is due to a lack of updates to the old phone to get mDNS working.

              We are only taking about an ESP32 project here, not an actual website. Just for convenience for data input and output this is far(!) better than a display and some buttons… Which is what I used to do in the past decade.