"using machine learning" --- what does that even mean? The test was designed with ML? What?
> Two-thirds of the tested students could not even write code that compiles
Did they get to choose what framework to use when responding to this question? Did they get to test their code? My intro to CS course in college had us handwrite code for tests, so guess what, I bet 95% of student's code wouldn't compile due to an occasional missed character or incorrect capital letter. Is this a similar situation?
This whole article needs a lot more detail around how this test was administered and what a borderline 'fail' looks like.
Assuming this research can be trusted, 5% of them means 75 000 new "good engineers" per year.
That is still a big number.