> Since phase 2 prices vary at most 10x and the big successes generate returns of at least 100x, investors should pick startups entirely based on their estimate of the probability that the company will be a big success and hardly at all on price.

To give concrete numbers to pgs statement:

Pick 2 hypothetical startups: A and B. A will go on to be a 10 billion dollar company and B will be a 100 million dollar company. Now, valuations at round B series vary from say $40 million to $400 million (as pg said 10x). Note: they arent yet worth what they will be worth later. Now, say you take a 20% equity cut for the round and there is no dilution between this and when they go public (just a simplifying assumption). At the end, the 20% equity is worth either 200 million dollars for A or 20 million for B. The difference in profit is 180 million dollars; much more than any additional amount you would have paid to get in on a higher valuation. Therefore, if you believe the company to be of the A type, you will pay that 20% of a higher valuation.

> pay whatever the price happens to be.

This is what pg means: the difference in profit between A and B was 180 million dollars which far exceeds the difference in cost in investing in the two. This makes the investors rather price insensitive IF they think that you are in the A category. The reason that investors have the mental model of assigning to categories rather than guessing the percent chance of success is that they know often they guess wrong. There are simply too many variables to create any sort of accurate chance of success.