How a PhD Switch Led to an OpenAI Offer: 6 Surprising Interview Lessons

Yong Zheng‑Xin, a fifth‑year PhD student at Brown University under advisor Stephen Bach, focuses on AI safety and scalable oversight. He will join OpenAI as an Astra Fellow in June 2026.

Inspired by Alisa Liu’s viral interview recap, he writes this post to highlight six surprising observations from his own research‑scientist job search.

Surprise 1: Papers matter little

He found that only one or two papers can open interview doors, and sometimes none at all. Despite publishing many multilingual papers and winning the NeurIPS 2023 SoLaR best‑paper award, those works had little influence on his AI‑safety interview outcomes.

Surprise 2: Interview formats are diverse

Beyond the expected LeetCode‑style coding and LLM questions, he was asked system‑design problems, async parallel‑programming tasks, and even to demonstrate the use of AI agents during the interview.

Surprise 3: Work trials are common

He experienced paid trial periods that differ from onsite interviews; some trials lasted up to a week and required full‑time collaboration on an open‑ended task, making it hard to prepare for other interviews simultaneously.

Surprise 4: Timing is crucial

The job market shifted dramatically: AI‑safety positions were scarce last fall, but now many startups (e.g., Lila, Mechanize) are hiring. Seizing the right moment can dramatically affect interview success.

Surprise 5: Retention offers are rare

Unlike many software‑engineer roles that provide return‑offer guarantees, research positions depend on team size and project context. His Meta internship in 2024 offered few full‑time conversion chances, and even with the Astra fellowship he still had to pass every interview round.

Surprise 6: Many interviews are unrelated to your research

Despite focusing on AI safety, many interview rounds covered topics far from his expertise, confirming that interviewers often assess overall research competence rather than niche specialization.

For safety researchers, he co‑authored a LessWrong article on AI‑safety technical interviews (https://www.lesswrong.com/posts/dvsFfGuXXyHYkyifp/tips-for-cracking-the-ai-safety-technical-interview-1) and lists additional reading resources such as Nathan Lambert’s market analysis and Alisa Liu’s job‑search notes.