In the links you referenced Eliezer Yudkowsky argues against applying the outside view for the singularity and AI foom. His argument is that the singularity is like nothing we’ve seen before, and so while there are many applicable outside views, in none of them is the analogy between the particular event being considered and others in the reference class compelling.
In contrast, you are making an argument about startups, which have already been done many times, and are encouraging others to do something in the pre-existing reference class. This is exactly the domain of applicability for outside view arguments and Eliezer Yudkowsky’s counterarguments are inapplicable.
My response will just lead to ‘reference class tennis’, which is why I think that for the purposes of having a conversation, we should take the inside view.
I think that people are getting too attached to my 80-90% success rate claim, and impulsively arguing that it’s wrong. My intentions in this article (that I thought were pretty clear) are just to get a conversation going about the components, work towards agreeing on those, and then maybe try to guess at what the resulting success rate will be.
In the links you referenced Eliezer Yudkowsky argues against applying the outside view for the singularity and AI foom. His argument is that the singularity is like nothing we’ve seen before, and so while there are many applicable outside views, in none of them is the analogy between the particular event being considered and others in the reference class compelling.
In contrast, you are making an argument about startups, which have already been done many times, and are encouraging others to do something in the pre-existing reference class. This is exactly the domain of applicability for outside view arguments and Eliezer Yudkowsky’s counterarguments are inapplicable.
My response will just lead to ‘reference class tennis’, which is why I think that for the purposes of having a conversation, we should take the inside view.
I think that people are getting too attached to my 80-90% success rate claim, and impulsively arguing that it’s wrong. My intentions in this article (that I thought were pretty clear) are just to get a conversation going about the components, work towards agreeing on those, and then maybe try to guess at what the resulting success rate will be.