> who can't sum a list of 10 numbers without using Google.

I did a hiring round once for my old team when I worked at Yahoo years ago, where out of 10 or so CV's the recruiters passed me, about 5 people who were given a short list of screening questions by e-mail for them to answer at their own leisure produced answers that were obviously plagiarized from various online sources.

I got suspicious when one guy presented me with two pages of nicely formatted answer complete with section headings that didn't make any sense to a question that should have required a one line answer. He'd copied and pasted the whole thing from an on-line copy of an Oracle manual, so I started doing searches of random sentences from their answers.

Another one managed to cut and paste an answer that was flat out wrong from a forum post answering pretty much the same question, and hadn't bother to read further in the comment thread, where several people corrected the answer he'd used...

It's when they can't even answer simple stuff WITH using Google it gets really depressing, and it happened regularly enough that using screening questions by e-mail was a real time saver in weeding out undesirables, despite the ease with which people could "cheat".

This sounds like what happens when you give students problem sets...

Absolutely. At my last C++ job, I did a lot of interviews and a lot of people claimed to know the STL. I left the low-level compiler questions to my colleagues for the most part, but I would ask a few key questions.

My favourite STL-related question to ask is simple and can be done on a whiteboard in minutes: if you didn't have std::set<class T>, but you had everything else in the STL, how might you implement a set<class T> class?

There's an ideal answer[1], and then there's people who start thinking about low-level implementations. The ones who I don't want are the ones who don't have any answer at all. I'll continue to be interested in anyone who starts talking about the way to solve the problem. What matters to me is not any code that you write (although that's interesting) but your thought process. After the candidate has worked through this, if they haven't gotten the ideal answer I always give it to them: this isn't some national secret, and seeing them process the ideal answer and realize that yes, it really would work is a pleasure.

Sadly, there are cases where it doesn't work. I had one candidate who did very well on the phone screen, but didn't know the STL. I looked up what he claimed to know (Qt) and reframed the question in terms of Qt's classes—and he still couldn't come up with any answer whatsoever.

-a

[1] The ideal answer is precisely such because you're writing much less code and using stuff that people smarter than you have written. It's to use a std::map<class T, bool>.

I've never met a programmer who can't sum 10 numbers without using Google. I have however met plenty who go blank when asked a question they weren't prepared for in front of an intimidating person in a small quiet room during an interview.

Not everyone has performance anxiety, but I've met plenty of programmers who do. Unless you spend the first 15 minutes of your interview calming them down and warming them up with simple technical questions that put them in the mode of programming, you might as well be asking them to sing the star spangled banner naked in front of an audience.

On the other hand, I wouldn't mind coding a program that solved problems of the river-crossing kind, expressing it as a search through a state space with constraints.