That's not what that paper says. Work sample tests are more valid than GMAs; the paper just presumes they're too expensive. Meanwhile, we don't have to axiomatically derive any of this: we know that relying on general cognitive assessments to prospective software developers wouldn't work. That's why almost no firms use them.

If you exclude work sample testing from your analysis, all this paper is really saying is that active examination of candidates beats subjective interviews and resume scans. Well, obviously.