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.
  • @saucerwizardEnglish
    5•8 months ago

    Have you read Barriers to Bioweapons?

    • @Umbrias@beehaw.orgEnglish
      2•8 months ago

      I have not. Is it good?

      Keep in mind that it was written in 2014, the Field of bioengineering in the past ten years has advanced considerably.

      • @saucerwizardEnglish
        3•8 months ago

        Yeah. It addresses your points I think.

        • @Umbrias@beehaw.orgEnglish
          1•8 months ago

          I’ll have to check it out.

          The general point seems to be yours, that intellectual availability is the largest restriction on bioterrorism. I don’t disagree, but a big part of my argument is that access to this information has never been higher (which is better than not for a variety of reasons) and access to resources usable for this has never been higher. We have plenty of garage scale bio labs as it is. So yes, the biggest limit is availability of people with knowledge to do it, that’s not a hard roadblock, at least not anymore.

          And the prediction horizon on biotech is tiny. Give it another ten years? Twenty? It’s not a zero threat because nobody has done it right now yet.

          • @saucerwizardEnglish
            5•8 months ago

            Not just intellectual availability, but the complexity of the job itself. iirc it goes into the Russian experience.