In both my experience taking a Google interview and conducting a lot of interviews at my own company, I agree that a good question is open ended. For example "w

In both my experience taking a Google interview and conducting a lot of interviews at my own company, I agree that a good question is open ended. For example "write memcpy" is trivial to get sometime working but can lead down many paths when discussing performance and computer architecture. What gave me trouble, besides being rusty with algorithms, were some trade offs I would never make in the domain I'm familiar with. In my day job my microcontroller has 192KB of RAM, so limiting the amount of data collected to fit is important. I never get to the point where I have to worry about my simple algorithm not scaling to GB of data. Another odd idea was doing lots of speculative computation to reduce the latency of a system. All of these algorithm things can be studied and after failing a couple of interviews you can start to understand what the interviewer is expecting and how to clarify assumptions. I do question what sort of company you build if you screen for people who can write code on whiteboards. There is no giant mono-whiteboard where all the software engineers check in their code, so the interview process is not related to what people actually do for the job. On the other hand I have interviewed hundreds of people and I don’t feel like my questions do much better with the same one hour slot given to me by HR. The whole situation is inefficient for everyone.