Although the article's title is off (no "best question" is ever given), I'll offer up the question with which I've had the best luck. I ask the candidate to de

Although the article's title is off (no "best question" is ever given), I'll offer up the question with which I've had the best luck. I ask the candidate to describe what they've currently been working on (sadly, this alone filters out a lot of people). Once I understand, I make up a feature that I'd like to add, and then I ask them to task it out and estimate how long the tasks would take. Depending on how bright the candidate is, you can poke at their suggested implementation to see if there's any good alternatives, if they understand the associated scaling or performance issues, etc. etc. Interviews are highly imperfect at best, but this has worked well for me. Your mileage may vary...