> The whole point of the article is there is zero correlation between people who've graduated from college and people who haven't in term of how well they do on the job.

... among the population who have been hired to do the job. This is conditioning on the collider. The people without a degree who got in are a highly selected sample compared to all people without degrees. Anyone who used this to argue that a degree wasn’t a strong useful signal Wouk be making the same mistake as those trying to get rid of the GRE in graduate admissions because in the population admitted the GRE doesn’t predict anything. If it did that would show under or overweighting ofbthe signal it sends. Zero correlation shows it has been given appropriate weight.