• tristynalxander@mander.xyz
    link
    fedilink
    English
    arrow-up
    0
    ·
    1 day ago

    AI-designed makes it sound like an LLM came up with an idea. If you follow the links you’ll get to the paper where you can see the list of authors who are doing protein design.

    I suppose protein design does use generative AI algorithms, but it’s not thinking or coming up with anything. It’s mostly just getting the atom positions more correct. In fact this one was super heavily restricted and still filtered a ton of garbage out. Frankly, it’s not even impressive as protein design goes as it’s barely a new protein. Maybe the predicting future variants is sorta impressive, but we’ve been doing that for years now.

    Honestly, I wonder how the first author feels about this. I can’t speak for them, but as someone who does protein design, if some journalist attributed my work to AI, I’d have a conniption fit.

    • darkangelazuarl@lemmy.world
      link
      fedilink
      English
      arrow-up
      0
      ·
      4 hours ago

      Yeah, Protein design is absurdly complicated and time consuming. Collaborative computing programs like folding at home have helped a lot but its still like trying to find a grain of sand in a desert. Having a well tuned AI help narrow down what needs additional attention via other computer models or even moving past computer models is a reasonable use of a well tuned specific AI. Not these general purpose LLMs that just tell you what you want to hear.

    • PotatoPie@lemmy.zip
      link
      fedilink
      English
      arrow-up
      0
      ·
      1 day ago

      I’m anti AI but i have to give it to the technology, for once, it’s used for something good and for a task no one wanted to manually do, with incredible results

      Until the sector is regulated i’ll just say this never happened

      • gedfromgont@piefed.ca
        link
        fedilink
        English
        arrow-up
        0
        ·
        24 hours ago

        Technically this probably isn’t AI but rather ML (machine learning) which is something that made great strides in science already in the past decade or so. E.g. alphafold, a program to predict folding of protein sequences predates all LLM crap and has become a cornerstone of anything protein expression.

        • tristynalxander@mander.xyz
          link
          fedilink
          English
          arrow-up
          0
          ·
          16 hours ago

          Nah, Alphafold came out at about the same time (actually a little after) AI image generation. They all came out together because the same hand full of people were working on them. They’re also all very similar algorithms; though, I don’t know where you want to draw the line between AI and machine learning – those are just two terms for the same thing to me.

          • gedfromgont@piefed.ca
            link
            fedilink
            English
            arrow-up
            0
            ·
            11 hours ago

            So when I talk about LLM, I mean in first line ChatGPT since it was the big event bringing LLMs to the public.

            Alphafolds first version came out in 2018, ChatGPT was only released in 2022. I don’t know where AI image generation comes into this, but that wasn’t my point anyway.

            • tristynalxander@mander.xyz
              link
              fedilink
              English
              arrow-up
              0
              ·
              8 hours ago

              Idk, there was a lot of hype about image generation at about the same time there was hype about Alphafold, and I bought a new computer immediately for Alphafold. My Amazon history says 2023, so 2022 might be about right. I can tell you the hype for Alphafold wasn’t until after COVID regardless of when the first version might have technically come out.