Totally agree, I'm somewhat in sympathy with Shalizi's argument that g is just the result of iterated factor analysis. That being said, done well, these kinds of tests can provide useful data, especially if you're willing to randomly hire to calibrate performance (which no one ever is, sadly).