My basic filter question was always this: in any language you know, write a function that reverses a string. No tricks, no gotchas. Any reasonable solution will be discussed briefly and then we'll move on to discuss other things.
But a shocking fraction of people -- even ones with CS degrees from fancy schools -- would spend 30 minutes stuck trying to write that function.
That's what I think of when I hear "can't code their way out of a wet paper bag".
In the early days of computing IBM hired candidates with the right aptitude but zero programming skills and then trained them. It Worked out fine.
In the year 2014 there's no excuse for having "zero programming skills" if you're looking for a job programming computers. We all have computers, it's not like you can't practice on your own time before expecting someone to pay you for it.
And we do have existence proofs that businesses can succeed by hiring candidates with the right aptitude but no programming skills and then training them. For example see IBM a few decades ago or Zoho today. If you don't want to do that then you have no right to complain about the market supply of skilled developers.
I'm not arguing that there are no people trying to get developer jobs who have no business doing it, or that it's always inappropriate to ask algorithms or data structures questions. I'm arguing that there is an overemphasis on trivialities, minutiae, and irrelevant details.
That being said, I could probably still come up with it, but I encounter this problem when I interview sometimes.
People will ask me to recall something we covered for ten minutes back in college that I've never used since and could easily Google in five minutes if I ever did need to use it, and then they go... "ooooh, he must not be very good..." when I can't instantly spout out the solution from memory in thirty seconds, and might not be able to picture it clearly in my head without writing it out or spending a few minutes thinking about it, possibly even writing some quick code to see what happens, like I get to do whenever I'm actually working and on the job.
Most recently I got that with a combinatorial function (write a function to come up with all possible combinations for blah). I was having a hard time picturing how it would show every combination in my head, including in a jumbled order. I could figure it out if I could hack away at the code for an hour, and I remembered going over it briefly in college, but I've never needed to do it in my personal or professional projects since, and so it wasn't fresh in my head. The feedback I got was they were passing on me because I got that one question wrong. I'm perfectly competent and could easily handle the job, but because I didn't study every possible algorithmic and non-algorithmic question I could possibly be asked in the three days notice I had before the interview, tough.
I'm more a fan of seeing the results of people's work than asking them trivia over a field whose sphere of knowledge keeps doubling every five years that they can't possibly store every single thing in their heads at all times.
I'm not sure how many developers Hal Salzman has tried to hire, but I find a lot more of them are weeded out at the "seeing if they can code their way out of a wet paper bag" stage than at the "salary negotiation" stage.