I think it's a little odd that AI safety people put so much emphasis on "existential risk." Just seems to set a needlessly high burden of proof. It is very hard for *anything* to kill literally all humans. Nothing's done it yet! By contrast, lots of stuff has killed a huge