This only adds (Bayesian) evidence to an abstract frame of knowledge that doesn't matter to anyone in real decision making processes: a frame in which the observer uniformly samples all startups.
People who actually care about these odds (in the sense of betting on them) are founders and investors. Neither, in their decision making process, gets to (or wants to!) uniformly sample all companies.
PG's numbers are thus much more useful to anyone actually trying to make a decision about a pool of investments: assuming the distribution of YC startups is fixed over time, and you are someone like Start Fund who will bet on the pool (i.e. equivalent to a repeated uniform sampling in expected value), the 0.5% is actionable information and the 0.00006% is not.
I don't disagree with you re:investors (or pg's numbers) who seek to obtain as much evidence as possible so as to maximize their ROI. But priors do matter.
For example, if the prior on "making a successful company" (defined however you want) were a vastly higher 40%, then I'd imagine a lot more laypeople would take the plunge. Reading sites like TechCrunch makes it seem to the layperson that building a successful company is much easier than it really is. So yes, knowing that "mega success" is a massive outlier (to the tune of 1:1,000,000) is indeed actionable information to a layperson thinking about starting a company without any additional evidence.
As the source article notes: The goal of the entrepreneur is to learn as much as they can, thereby increasing their own odds of success (or minimizing their odds of failure). Obviously, getting into YC massively improves your odds and would probably be a good decision! As a YC alum, my advise would jive with this observation. ;)
People who actually care about these odds (in the sense of betting on them) are founders and investors. Neither, in their decision making process, gets to (or wants to!) uniformly sample all companies.
PG's numbers are thus much more useful to anyone actually trying to make a decision about a pool of investments: assuming the distribution of YC startups is fixed over time, and you are someone like Start Fund who will bet on the pool (i.e. equivalent to a repeated uniform sampling in expected value), the 0.5% is actionable information and the 0.00006% is not.