We’re developing a blueprint for evaluating the risk that a large language model (LLM) could aid someone in creating a biological threat. In an evaluation involving both biology experts and students, we found that GPT-4 provides at most a mild uplift in biological threat creation accuracy. While this uplift is not large enough to be conclusive, our finding is a starting point for continued research and community deliberation.
  • rook@awful.systemsEnglish
    0·
    11 months ago

    This reads to me more like assuming all terrorists are fundamentally incapable of anything remotely intelligent

    The first paper you linked there lists 9 deaths and 806 injuries across 50 years. Conversely, you can look at a single example like the Manchester Arena bombing in 2017 and see deaths and more injuries from a single event using simple techniques where materials and instructions are readily available. It isn’t unreasonable to look at the lack of success of amateur biological and chemical attacks and assume that plausible future attackers will be intelligent enough to simply take the tried and tested approach.

    On the other hand, there might be some mileage in hyping up the threat of diy countertop plagues in the hopes that would-be terrorists are as credulous as so many politicians and media figures are, and will take the pointlessly inconvenient and inefficient option which will likely fail and make life a little safer for the rest of us.

    • Umbrias@beehaw.orgEnglish
      0·
      11 months ago

      tried and tested

      Nobody is saying terrorists won’t keep using conventional bombs. Terror attacks aren’t just about maximum kills nor casualties per dollar, however, and as the barrier to entry lowers and lowers it’s important to consider ramifications from many technologies.

      hype them up to fail

      This does not seem a reasonable countermeasure when the risk of failure is potential pandemics.