Moreover, whatever money you make on any startups that do not make $BIGNUM is rounding error by comparison.

If you'd reword that as "the basic rules of expected value are difficult to apply to startups", there'd be some chance it was true :).

And yet, difficult as it may be to apply, expected value is the framework to rationally make a decision about low probability / high payoff investments. Of course, if YCombinator is already invested at an early stage (note: when the valuation is quite low), I can understand why pg wouldn't care too much about the later stage valuations. If you look at what's best for YC, it's first and foremost that companies get the money they need to succeed (an incentive aligned with the founders and any investors) but probably also that the valuations (after their own investment) be as high as possible so that more of the pie remains for later investments. This latter incentive is clearly not aligned with investors and as an investor I'd take the advice to disregard valuation with a big grain of salt.

Excerpt: "To a first approximation, a VC portfolio will only make money if your best company investment ends up being worth more than your whole fund." This is the big hit, and VC's are trying to optimize their chances of getting one of these. That's different from trying to precisely calculate expected return. As long as you've found it, it won't matter if you paid a bit too much.