Curated. I am broadly skeptical of existing “coordination”-flavored efforts, but this post prompted several thoughts:
I have very incomplete models of what those efforts are (much more so than my models of technical alignment efforts, which are also substantially incomplete)
It seems likely to me that some people have better models of both existing efforts, and potential future efforts, which have not been propagated very well, and it’d be good for that to change
I think this post does a good job of highlighting representative objections to various proposed strategies and then demonstrating why those objections should not be considered decisive (or even relevant). It is true that we will not solve the problem of AI killing everyone by slowing it down, but that does not mean we should give up on trying to find +EV strategies for slowing it down, since a successful slowdown, all else equal, is good.
Curated. I am broadly skeptical of existing “coordination”-flavored efforts, but this post prompted several thoughts:
I have very incomplete models of what those efforts are (much more so than my models of technical alignment efforts, which are also substantially incomplete)
It seems likely to me that some people have better models of both existing efforts, and potential future efforts, which have not been propagated very well, and it’d be good for that to change
I think this post does a good job of highlighting representative objections to various proposed strategies and then demonstrating why those objections should not be considered decisive (or even relevant). It is true that we will not solve the problem of AI killing everyone by slowing it down, but that does not mean we should give up on trying to find +EV strategies for slowing it down, since a successful slowdown, all else equal, is good.