Would it make sense for the measure of real-world intelligence to be "optimization power per unit time" or similar? Given arbitrarily large amounts of time, I could do an exhaustive search of the solution space or something like that, which isn't very intelligent or useful.

Another point: It doesn't seem like optimization power is usefully defined for every possible search space. Let's say our search space is countably infinite, each item corresponds to a single unique natural number, and each item's score is equal to its corresponding natural number. I think for a while and come up with the solution that has score 1 million. What's my optimization power? 0? Regardless of the solution I come up with, my optimization power is going to be 0, even though I vastly prefer some solutions to others. (I don't know how much of a problem this would be in practice... it seems like it might depend on the "shape of the infinitude" of a given solution space.)

(Let me know if these don't make sense for some reason; I haven't taken the time to understand EY's idea of optimization power in depth.)