While I don't necessarily disagree with this, I also wonder how much his 'data' about Nobel Prize winners and Institute of Princeton grads actually holds up vs how much of it is just very expected regression to the mean. He talks about Shannon; at some point Shannon was always going to have his last great idea. Given that the idea that made him famous was his greatest, you wouldn't expect many other ideas like that just from normal variation.

Essentially, if you take scientific ideas, including Nobel Prize ideas, and put them all on a bell curve of how difficult it is to find them, you wouldn't expect the same person to have multiple ideas all the way on the right, even if they are very above average.