This is a great article which unfortunately also does a great job at meandering and overcomplicating the point.

TL;DR The popular understanding of the Dunning-Kruger effect is that incapable people think they’re smarter than do the capable people. That’s wrong. What the data actually show is that the people tend to estimate their ability level as being somewhere in the middle. So, the incapable people “overestimate” their skill - putting it “a little below average” when it should be further below average than they estimated - and exceptionally capable people do the same on the other side. The correlation between skill estimation and actual skill is actually positive - just not as high as it “should” be - not negative as the popular understanding would suggest. The negative correlation is what you get when you subtract people’s actual skill from their estimated skill, making it arguably just an autocorrelation artifact.

Lots more data and details in the article but that’s the gist.