In one experiment he describes, two trained interviewers conducted interviews with a group of volunteers. Their evaluations were compared to those of another group who saw a fifteen second video of the interview: the candidate entering the room, shaking hands, and sitting down. The opinions correlated strongly; in other words, when you are sitting in an interview telling the interviewer what you do on your day off and what the last book you read was, the interviewer has already made up his or her mind, based on who knows what subjective criteria. As Poundstone laments, "This would be funny if it weren't tragic."

Unskilled interviewers use puzzle questions with a cargo-cult mentality. They think that merely by asking them they will gain insight. If you don't already know what information you're looking for and how to gather it, puzzle questions won't help because you won't be able to interpret or steer the results. If you don't know how to conduct an interview and what your goals are, good questions won't help.

I once had a worst-case scenario, combining a snap-judgment from the interviewer with a poorly managed puzzle question. After a day of some good and some distant interviews, my next interviewer hated me on sight. He managed to exude "you suck" vibes just shaking my hand. He gave me an open-ended question which ended up consuming the whole hour. I tried to narrow down and define the question and enter into a discussion with him, but he just wasn't interested.

Good interviews, in my opinion, consist of observing the interviewee using the skills that the job will use. So a puzzle question can be used to elicit problem-solving or (more usefully) cooperative problem solving. For example, a good interview might go like this:

Q: Okay, so we have a widget that needs to do X. Show me how you would build it.

A: Looking at the problem, my first step would be, Has Y been considered? In this situation I think Y might be a better solution, and certainly five to ten times cheaper.

Q: Good point. How would you make that case to a business manager? ... Okay, now that you've learned that reason Z forces us to do X instead of Y, show me how you build a widget that does X.

A: [demonstrates]

But this guy's goal was just to kill an hour and generate evidence for his theory that I sucked. The nadir, as we discussed "how to design a city," and I struggled to enumerate ways to "build community," was when he showed a brief flicker of interest in one of my answers and asked a follow-up question, "So you'd want to encourage Starbucks, because of how their stores create that kind of community space you're talking about?"

After a too-long hesitation, I replied, "Uh, yeah."

I was sent home immediately after, but I wish I'd been sent home immediately before. Or at least given him a real answer: "No, you yuppie jackass, Starbucks does not foster community."

Oh, and the time I interrupted with the solution to the buckets problem before the interviewer could finish asking - I got that job. They wanted people who could interrupt and disagree forcefully but politely, because when everybody in a group has enough trust to do that, things go much faster and bad ideas get fixed much sooner.