If looking for programmers: write a hard, challenging (but not unnecessarily complicated) coding test - you don't want to waste their time, but to excite them (and, at the same time, filter them). Like Matasano does.
Instead, I'd just avoid "real-world" problems that your company has actually worked on unless it's clearly stated as such and you're compensating the candidate for his time.
I'd also avoid challenges that are actually "solve this math problem, and express your solution in the form of a program." You see these all over in places like codility. Perhaps it's just because I have extensive education in research and mathematics already, but I consider these things to be "numerical parlor tricks" more than they are a guage of whether a person can reason out a programmatic solution to an actual problem. They almost always seem to put the candidate in the mindset of "I have to solve a math problem" rather than, "I have to write a program." Contrary to what some mistakenly believe, the two aren't the same, and for all but the most research oriented CS work the latter skill can be entirely independent of mathematical reasoning ability.