• NoneOfUrBusiness@fedia.io
    link
    fedilink
    arrow-up
    35
    ·
    2 days ago

    According to this non-paywalled article, it created thousands of potential genomes from which the scientists running the experiment chose three hundred promising candidates. They then had bacteria make those and used them for their bacteriophagy experiment. So essentially the AI made the blueprints that the scientists then turned into real stuff.

    • danekrae@lemmy.world
      link
      fedilink
      arrow-up
      13
      arrow-down
      2
      ·
      2 days ago

      Let’s get real here. Could the bacteria make my Octophant if I clean up the pencil lines?!

    • barnacul@lemmy.world
      link
      fedilink
      arrow-up
      4
      arrow-down
      3
      ·
      2 days ago

      So the AI’s role in all this, given that creating new viruses in a lab is already understood and well-practiced, was just to generate a list of virus configuration options (presumably working from existing examples), from which scientists had to discard ~90% of what the AI generated before moving on to the next step.

      Sounds like an attempt at a headline more than a genuine scientific effort.

      • Bluescluestoothpaste@sh.itjust.works
        link
        fedilink
        arrow-up
        1
        ·
        6 hours ago

        Scientists had no clue at all how to design bacteriophages is my understanding. Only with the help of LLM have they been able to generate the first man-made bacteriophages ever.

      • NoneOfUrBusiness@fedia.io
        link
        fedilink
        arrow-up
        9
        arrow-down
        1
        ·
        2 days ago

        So the AI’s role in this process, which is already known and well-practiced,

        Uh… no? Creating custom-made genomes for curing specific diseases isn’t “known and well-practiced;” it’s an area of active research.

        from which scientists had to discard ~90% of what the AI generated before moving on to the next step.

        That’s basically in line with how AI is used in similar applications; even if most of what you get is busts AI can make so much stuff that it can compete with (or even outdo) other alternatives by sheer quantity. Sure someone has to filter out the useful stuff from the trash, but the very fact that it produces useful stuff makes it worth it. This method is well-established in other areas, so there’s no reason to dismiss it here.