Ask HN: Evidence which settle the C/C++… vs Python/Lisp/Haskell… Deathmatch?

I am most interested in whether programs written in "high-level" languages have less errors. (Note that I am talking about errors that can go undetected, and are actually shipped). That's a disagreement between me and a colleague:

To me, a high-level language will yield simpler and shorter programs, and tend to remove some classes of errors. That let your mind concentrate more on the problem, and less on unrelated details. Therefore, your program tend to have less errors.

To my colleague, using a high level language makes you lazy. By not exercising your mind regularly with trivial problems (such as indices in `for` loops), you tend to be less awake, if not a bit dumber. And that effect at least nullifies the benefits of a shorter and simpler program (when it comes to count the errors). In the end, your program does not have less errors. It even tend to have more.

So empirical evidence would be nice.