[caethan]

I've heard bitching about some of the interview questions our team asks before, but here's the thing: each of those questions is about a problem our team actually had to solve before, reformulated into an interview-style question. Yes, we've had to use Hamming distances, worry about the scaling (N log N vs N^2) of particular solutions, use error-correcting codes, interesting data structures and all of that. Is that most of the job? No, we do a lot of more boring stuff too, but the algorithms and data structures are definitely a part of it. I don't want someone who can only glue pieces together, developing novel tools to solve the problem is important too.

[lhorie]

One needs to be especially careful with adapting experience to interview questions. Unlike the candidate, you probably were already immersed in the problem domain for months or years and have good context in what techniques exist and are relevant to the problem domain.

One mistake some interviewers make is implicitly assuming that candidates can somehow conjure the same level of context from first principles, or that a specific algorithm might be familiar or reusable outside of its original context. Another mistake is "looks-like-me" bias.

For example, I happen to have a lot of context on a very specific algorithm that underlies basically every modern web framework but if I wanted to evaluate a candidate on web performance, I'd look at performance optimization as a open ended problem domain rather than drilling them on the particulars of this specific algorithm. In fact, out in the world of web framework performance, the most novel advancements come not from revisiting the algorithms but from looking at the problem domain from entirely new angles that had not even been considered before.

[xsmasher]

Every time I feel "this would be a good interview question!" it is not. It's usually something juicy that I chewed on for two days. How can I expect a candidate to solve it in 45 minutes?

[riddleronroof]

I bet you have. But how long did you have to solve it? Did you have access to the internet, talking out ideas with your colleagues, coffees? I agree that you can’t simulate real scenario in an interview, but you can also acknowledge that the process is a little cartoonish.

[caethan]

Interviewees are encouraged to use Google, code on a laptop rather than on a whiteboard, etc. Sadly, I only get an hour to interview people rather than a week. But I have had people complain about "completely unrealistic problems unrelated to the real work" when the problem is something I literally had to solve 3 months ago.

[walterclifford]

> each of those questions is about a problem our team actually had to solve before

On my last job search I had one interviewer state (very proudly) the systems design question I was being asked was an actual problem his team had to solve. I don't doubt the veracity of his claim at all, but it probably wasn't solved by a single person under the time constraints and pressure of an interview.

Most likely someone on that team spent hours or days researching and designing potential solutions before drafting a design document that was shared and discussed with others, perhaps informally or perhaps in a meeting (or over the course of multiple meetings) where tradeoffs were considered among people with deep knowledge of the existing system and problem space. Expecting a candidate with only superficial (at best) knowledge of your current system to come up with the same or similar solution on their own in 30 or 40 minutes seems a bit unrealistic to me.

[caethan]

The context is more like this: we regularly have internal brainstorming sessions when we run up an interesting or tricky problem, to come up with ideas on how to solve it.

So in the context of an interview, I'm trying to treat the interviewee like a colleague who I'm coming to with a problem I'm having, so we can come up with a solution together. That often involves drawing things out on a whiteboard: not code, but more diagrams to describe the problem. Then we come up with ideas on how to do it, under various constraints that I share.

Usually I have in my pocket 2-3 different approaches that we tried when we did it ourselves, and I'm looking for: can you understand the tradeoffs between these different approaches, do you understand how they work, and are you capable of implementing them to test and cross-compare them?

[burnished]

Is your team hiring? Sounds like fun.

[ZephyrBlu]

What kind of stuff does your team work on? Seems like you have to solve a lot of diverse problems.

[caethan]

Scientific computing. We write software to do domain-specific scientific analysis on customers' machines. So we have a diverse set of needs of the specific analyses to do: sometimes we just use algorithms out in the open literature, sometimes we have to develop them ourselves. We've also had substantial work in devops, because we have to release software packages that not-very-savvy academic users can deploy and run on their own wildly-varying machines.

It's interesting work.