2.2.2 Alternative interview formats
Given how expensive, time-consuming, and inaccurate the traditional interview process is, companies are experimenting with new interviewing formats.
- Multiple-choice online quiz. In the last few years, there have been several startups providing online multiple-choice quizzes as a service to companies who want to use this format as part of their hiring process. The quizzes are timed so that you won’t have time to google the answers. Some companies are using it as the first screen in replacement of the coding challenge.
- Quiz. One or more interviewers might rapidly throw questions at you to test your knowledge and experiences. DeepMind was the company best known for this. Their quiz lasted two to three hours. At some point, their list of questions was leaked and passed among candidates, which might be why DeepMind abolished this format.
- Take-home assignment. Much of ML is still empirical -- knowing the concepts doesn’t necessarily translate to getting things done -- and companies want to see if you can get things done in your comfortable environment. They might give you a paper to reproduce or a problem related to their products to solve and ask you to present your solution. You can look up the solution online, but be ready to defend every architectural choice you make.
- Code debugging. The hard part of coding isn’t writing code, but building upon other people’s code. In this interview, candidates are asked to fix a buggy program, which might be a buggy ML model. This interview allows companies to evaluate your code comprehension, your reactions to other people’s code, and your ability to debug. Stanford uses code debugging when recruiting section leaders41. OpenAI and Snorkel AI are two companies that also use this format.
- Pair programming. This is very similar to the traditional coding interview, except that when you get stuck, the interviewer is willing to write code with you, allowing you to continue to the next part.
- Good cop, bad cop. Most companies have one-on-one interviews, but some companies, including Apple and Element AI42, do two-on-one interviews. Candidates have described this format similar to “good cop, bad cop”, with one interviewer asking difficult questions and another helping you out.
41: A role similar to teaching assistants.
42: Yoshua Bengio’s Montreal-based AI startup that suffered a rather bleak outcome. See Element AI sold for $230-million as founders saw value mostly wiped out, document reveals (Globe and Mail, 2020).