This is an excellent thread of common-sense steps for reducing harms from AI. It’d maintain the US’s AI lead (limiting regulatory burden) while mitigating many of the near-term and long-term safety risks. Esp. if it targets frontier systems & exempts the rest Let’s just do it.