Best policy idea for AI safety? Best one I won’t have heard of? Best 10? (Any policy ideas floating around in AI safety that are bad/doomed?) If we live in a world where people can accidentally kill everyone by making powerful AI, what policy levers should we pull?
Takes on the track hardware, mandatory licensing for large training runs, monitor large training runs with capability evals & red-teaming & audits, pause training runs with concerning eval results plan? Takes on other plans, like training compute cap that gradually grows over time or the underspecified-but-evocative IAEA for AI?
Best policy idea for AI safety? Best one I won’t have heard of? Best 10? (Any policy ideas floating around in AI safety that are bad/doomed?) If we live in a world where people can accidentally kill everyone by making powerful AI, what policy levers should we pull?
Takes on the track hardware, mandatory licensing for large training runs, monitor large training runs with capability evals & red-teaming & audits, pause training runs with concerning eval results plan? Takes on other plans, like training compute cap that gradually grows over time or the underspecified-but-evocative IAEA for AI?