Sure, I consider the "(for humans)" parenthetical to be doing a lot of work. If we interpret "good" as meaning what (some agent-like system) would want if it knew more, thought faster, &c., it's going to be true that some humans are quantitatively less good with respect to the "good" (extrapolated volition) of everyone else. But I expect most people to wildly overestimate the quantitative extent to which this matters; selfishness is a much more powerful force in the world than outright evil. So, you should read me as claiming that smart is improvable by a much higher percentage.