ryan_greenblatt comments on Mitigating extreme AI risks amid rapid progress [Linkpost]