From a PhD to OpenAI: 57 Interview Lessons and Process Insights

57 Interviews Reveal Process Management Over Single Skills

A University of Washington NLP PhD applied for research scientist positions, ultimately joining OpenAI, and documented a comprehensive job‑search project involving 11 companies, 57 interviews, 46 recruiter calls, 16 post‑offer discussions, and numerous informal networking conversations.

The author emphasizes that industrial research hiring is not a linear pipeline but a parallel, multi‑company effort where candidates must simultaneously manage timelines, preparation material, headcount changes, team hiring signals, and offer deadlines.

Technical Skills Matter More Than Research Record

Interview categories included ML coding, general coding, technical discussion, research discussion, behavioral, math, and job talk. The author notes that technical skills and knowledge are evaluated far more heavily than research experience , which mainly helps secure interviews rather than guarantee success.

Typical technical questions probe the ability to implement architectures, write decoding strategies, explain positional encoding, compare PPO and GRPO, and produce reliable code without AI assistance. Some ML coding rounds restrict tools to numpy and may require writing a backward pass from scratch, highlighting a gap between academic research training and interview‑ready coding proficiency.

Preparing for Interviews Is Like Re‑taking Core Undergraduate Courses

The preparation process is described as akin to revisiting undergraduate coursework: taking notes, drawing diagrams, solving exercises, and immersing in coffee‑shop study sessions to master fundamental ML concepts.

The author first completed Stanford’s CS336: Language Modeling from Scratch , building an LLM knowledge graph, then deepened each concept through blog posts, papers, ChatGPT, Claude, and hands‑on implementations. Emphasis is placed on mastering CS336 Homework 1 to achieve muscle memory for transformer implementation and debugging, and on disabling AI assistance during coding practice because real interviews do not allow such tools.

Each interview is likened to a short‑term exam with roughly three days of focused preparation, turning the entire job search into a full‑time job that requires continual, customized study based on company, role, recruiter hints, and interview descriptions.

Negotiation Extends the Process Beyond the Offer

After receiving an offer, the process continues with conversations with future teammates, managers, lunch visits, recruiter calls, information alignment, and salary negotiation. The author stresses that PhD training rarely covers negotiation; candidates are at a disadvantage compared to recruiters in market knowledge and bargaining tactics.

Even if compensation is not a primary concern, failing to negotiate can cost significant value; initial offers typically contain room for improvement, and some recruiters explicitly state they do not expect candidates to accept the first offer. Investing weeks in negotiation can be literally equivalent to years of salary at the initial offer .

Practical negotiation tactics include leveraging friends for communication know‑how and market data, scripting what to say and what not to say before each recruiter call, anticipating recruiter questions, and preparing responses that protect personal interests while remaining comfortable.

This stage reframes the job search from a pure ability test to a high‑stakes information‑game where candidates must both be evaluated and actively influence outcomes.

Notes on the Industry Job Search
https://alisawuffles.github.io/blog/job-search/