Written by Jordan Sale, Rora
Getting an Interview at OpenAI
OpenAI is exceptionally selective. They have a bias toward researchers from top institutions or those with numerous papers in top journals. Historically, they’ve also liked researchers from Google Brain / Deepmind.
That said, one of the things we liked about OpenAI is that when an internal team member refers you, the recruiting team reviews your resume. That sounds like it should always be the case but, at many large tech companies (like Microsoft), many referrals are unread – unless it comes directly from the hiring manager.
If you can’t get a referral we’ve also had clients successfully reach out to recruiters and hiring managers through cold emails and LinkedIn.
Another approach is to join the OpenAI Residency Program. It’s a paid 6-month program and (if it’s a mutual fit) you’ll be given the chance to convert to full-time.
Interviewing at OpenAI
For engineering roles – the interviews are generally very similar to FAANG where you are given medium/hard Leetcode questions and system design questions.
Additionally, you can expect statistics questions and machine learning questions, like to approach fine-tuning a model. If you are a traditional, backend, or systems SWE it’s unlikely that you will get more than a few high-level machine learning questions.
For research roles, the interview process includes:
A Hiring Manager chat: A brief chat before the actual virtual onsite. Intros, informal chat about research interests, behavior questions, and future directions.
Coding: A SWE interview, not questions you can find on Leetcode. The recruiter will send over some topics to get familiarized with ahead of time. But even if you read the docs, it’s still very hard to prepare for. You basically have to be an expert in the topics they list.
ML Coding: A coding question about common topics in ML. It has multiple parts, from simple to hard, including coding and open-ended questions. Having an understanding of numpy and pytorch is helpful.
ML Debugging: This is similar to the DeepMind debugging interview. In the first part, you’ll see a piece of code on some neural network model, and you need to fix all bugs in it. In the second part, you might also be asked to implement some new feature in this model. Familiarity with pytorch is helpful.
A Research Discussion: A two-part interview, first discuss a paper and then your past research experience. The interviewer will send a paper a few days in advance for you to read, and you need to talk about it in terms of the overall idea, method, findings, advantages and limitations, etc. Then you’ll discuss your research, the team’s research, and potential interest overlaps.
A Team Lead Interview: A behavioral interview with the team lead. Mostly “tell me about a time when …” type of questions. Would also be helpful to get familiarized with the company’s vision and mission. Be prepared for “big-picture” questions.
To sum it up – the coding questions (SWE coding, not ML coding) tend to be very, very hard. If you want to work at OpenAI – you really need to be a coding machine. Unlike other companies, the interview process is much more coding-focused than research-focused.
Additionally – the recruiter will give you a two-sentence description of the specific technical interviews in your interview slate – take those seriously! They are telling you exactly what to study.
Their interviews are demanding so we’d recommend scheduling them near the end of your interview tour when you feel most prepared.
Rora also has a 62-page technical interview guide you can download here that overviews the Amazon, Meta, DeepMind, etc. interview processes.
_____________________________________________________________________________________________________________________________________________________________________________
Remember that the Mignone Center for Career Success is here to help you with your interview preparation. Visit Prepare for an Interview – Harvard FAS | Mignone Center for Career Success to learn more.
