I examined point 2 in this section of my cost-benefit analysis. I collect estimates of revival probability here (I subjectively judge thesetwo metaculus estimates to be most trustworthy on the forecast, due to the track-record of performance).
As for point 3: Functional fixedness in assuming dependencies might make estimates too pessimistic. Think about the Manhattan or Apollo project: doing a linked conditional probabilities estimate would have put the probabilities of these two succeeding at far far lower than 1%, yet they still happened (this is a very high-compression summary of the linked text). Here is EY talking about that kind of argument, and why it might sometimes fail.
I examined point 2 in this section of my cost-benefit analysis. I collect estimates of revival probability here (I subjectively judge these two metaculus estimates to be most trustworthy on the forecast, due to the track-record of performance).
As for point 3: Functional fixedness in assuming dependencies might make estimates too pessimistic. Think about the Manhattan or Apollo project: doing a linked conditional probabilities estimate would have put the probabilities of these two succeeding at far far lower than 1%, yet they still happened (this is a very high-compression summary of the linked text). Here is EY talking about that kind of argument, and why it might sometimes fail.