@lennart apologies if this is a silly question, but either there's an error in footnote 4, or I misunderstand something fundamental:

A petaflop/s is ${10}^{15}$ floating point operations per second for one day. A day has $86,400seconds\approx {10}^{5}seconds$. Therefore, ${10}^{20}$ floating point operations

Shouldn't this read something like (in verbatim spoken words) 

"A petaflop per second is ten to the power of five floating point opeations per second. A day has [...] 10 to the power of five seconds. Therefore, a 'petaflop-per-second' DAY is 10 to the power of twenty floating point operations."

You've said a petaflop/s is x flop/s for one day, which seems like a typo maybe?

Would you say "petaflop-per-second" days if reading out loud?

Thanks for the kind words! 

Thanks also for the suggestion re: narrating the full document, rather than the summary. This does seem like it would be valuable in many cases and feedback like this is likely to push us in that direction :)

 

>So on technological grounds, at least, there seem to be no strong reasons to think that the distribution of war outcomes continues all the way to the level of human extinction.

Is this possibly missing a 'not' in there? It seems like you just concluded that there was no reason to think there was a cap on how many deaths could come from advanced military AI systems, then conclude there's no strong reason to think the distribution continues all the way to extinction?