...some extra time with this AI transition, maybe we could build AI more safely." "So you might think that if we spend more time doing AI alignment research, the existential risk from developing AI goes down at some rate," Bostrom added. "And then on...