After a technical screen of random basic ML questions, you are invited to an actual onsite in their SFO office. All interviews are done on their laptop. They don't trust any interview candidate that they will not cheat. They provide their own AI assistant for syntax checking during the interview and encourage its use. - Recruiter before the interview said interviewing at this position has been a revolving door of candidates (why? They are not Deepmind/a AI research company). There are enough qualified candidates for a pure applied position with no need for model training. - Many interview questions are 'gotchas!' - Some interviewers on research side are very junior (no research experience, 2 years in a startup with an MS) - For a general research question, your answer is expected to be a 'framework' name that they have used. Experienced researcher will come up 5 different ways of doing the same thing or invent their way in the answer.