A buzzy new startup called Radical Numerics claims to have found a novel way to treat bacterial infections with AI, though questions abound.
  • pomegranatefern@sh.itjust.worksEnglish
    29·
    3 days ago
    1. For context, this is a bacteriophage, which is a virus which infects bacteria. It is being used in a compassionate use scenario as a last-ditch effort against an antibiotic resistant bacterial infection.
    2. Bacteriophages tend to be sufficiently specialized to infect viruses that there’s not really a risk of them infecting humans. (There is some possibility of bacteriophages infecting beneficial bacteria such as the human gut biome, but I don’t know how much of a risk that would be in this case)
    3. While I’m not clear on the details of the AI being used here, it is described as a “multimodal genome language model” by the company and appears to be trained very specifically on the genetic data of bacteriophages which infect the target bacteria.

    When it comes to medicine, it’s best to move slowly and keep things intact, and while AI trained like this does have promise for exactly these sorts of medical purposes, it’s still algorithms outputting stuff which cannot be blindly trusted. But this also isn’t ChatGPT spitting out some idea of bacterial genetic code it generated from predictive text based on the collected works of Michael Crichton. Let’s not spread misinformed panic.

  • brem@lemmy.worldEnglish
    411·
    3 days ago

    At least the zombies created with AI will be extra fucking lame and predictable

    “That’s absolutely correct, I’m definitely going to eat your brains with my teeth! Would you like more recommendations on how to eat brains?”

    • shalafi@lemmy.worldEnglish
      6·
      3 days ago

      Colloquially they’re the same, but wouldn’t this sort of thing more fall under machine learning? I’ve never been clear on the difference.

      You’re right in any case. Just like the DC hate, blanket AI hate is going to result in friendly fire.

    • dgdft@lemmy.worldEnglish
      41·
      3 days ago

      I don’t disagree with the general statement… but this was specifically done with an LLM, so it seems like a moot point?

  • peopleproblems@lemmy.worldEnglish
    171·
    3 days ago

    … is it a virus generated by LLMs, or a virus designed by a model explicitly trained on viruses?

    The worst damage LLMs do is when the term AI is automatically rejected by people who don’t understand it.

    And that might be intentional.

    • dgdft@lemmy.worldEnglish
      5·
      3 days ago

      Based on the linked paper, it reads like they took an existing open weight LLM, tacked on a scheme to parse and output genomics data, and finetuned it a bit. It’s the same approach chatgpt and other big-name models use for reading and producing images, but applied to a different data format.

      • peopleproblems@lemmy.worldEnglish
        2·
        2 days ago

        Yeah see that is what I couldn’t figure out, but it lines up.

        It’s not how I would generate anything other than text and certainly not any sort of medical treatment

  • Treczoks@lemmy.worldEnglish
    5·
    3 days ago

    That was one of the headlines I was very afraid of. The point that this is “just” a bacteriophage mitigates it a bit, assuming that they just modified the anchoring and injection RNA code, but even this might get out of hand.

  • shalafi@lemmy.worldEnglish
    5·
    3 days ago

    They can’t be too new or they wouldn’t be anywhere near human trials. Anyway, this is the sort of thing AI (machine learning?) excels at. Not getting excited until we see peer review, but the methodology seems sound (to me, an armchair biologist).

      • shalafi@lemmy.worldEnglish
        6·
        3 days ago

        I’m down for that and think it needs expanded. If I’m terminal, I want the choice to be used as an experiment, assuming the FDA finds the treatment somewhat viable.

        Dad had mesothelioma from serving on a WWII destroyer (asbestos wrapped pipes, all those men died by the 80s). He went to some quack in the Bahamas who mixed up snake oil from all his patient’s plasma, patients who later tested positive for AIDS. AIDS hit the public conscious right about when dad died. One wonders…

        • dgdft@lemmy.worldEnglish
          4·
          3 days ago

          Yeah, totally agreed.

          It’s a tough balancing act to filter the snake oil while taking viable moonshots, but I think that’s a niche that the US regulatory regime has generally done well on, all things considered.

    • Avicenna@programming.devEnglish
      2·
      3 days ago

      Yeah my first question was “how the hell did they get approval for a human trial”. Until covid I don’t think you would even be able to get approval for mRNA flu vaccine human trials.

  • ZephyrXero@lemmy.worldEnglish
    42·
    3 days ago

    Well, that’s the dumbest headline I’ve read all week. What could possibly go wrong?