Just because there are exceptions, and we have names for both the positive and negative forms of the exceptions, doesn't mean there's not a strong correlation overall.

In the Dunning-Kruger paper, estimated ability is positively correlated with ability.