Real programming projects are best. I think one of the best methods would be to give a person a reasonable assignment that should only take a couple hours to complete (if that long) and ask for the results within a few days. (This will give them plenty of time to clean it up, add surprising features, and create a nice design and front-end after they are finished with it. A day can be used for the promise of completion, and the remaining time can be used to overdeliver.)
My point was that nobody else is talking about that. You just brought that up completely out of the blue, but in response to someone. When you reply to a post, those of us reading the thread tend to assume your post will in fact be in reply to the parent post, not a completely unrelated post just stating your opinion on something else entirely.
I hate that, to most folks who are into CS, CS is a yardstick for how much value a hacker will develop. "Can't optimize this from O(n^2) to O(n)? I have no idea how you keep your job." There is far more to the value produced by a hacker than the typical interview questions, and this elitist attitude out of most compsci folks is obnoxious.
This is why, even though I understand CS, grilling me on theoretical CS in your interview is automatic points off and I'll accept the position that actually quizzes on practical. I'm definitely of the mind now, being a self-taught programmer with a working knowledge of compsci, that I will learn things when they're needed to execute (not so they'll sit around in my brain waiting for the manager that Googled hard interview questions to ask).
Every programmer should be able to fizzbuzz. 95% can't. This is a problem. Every CS grad should be able to build a tree. 95% can't. These are serious problems.
What I find is: almost nobody can program, and almost nobody knows basic CS 101 data structures.
It does not matter if you are a self-taught programmer with ten years of experience in the best companies, odds are that you cannot solve a trivial coding exercise.
It does not matter if you have a masters degree in Computer Science, odds are that you cannot successfully build a tree structure.
I have no idea how these people keep their jobs or how they graduated, but this is the norm. People who can do CS at all, and people who can code at all, are both rare. Or at least, are rare in the pool of people applying for work.