OpenAI·Data Scientist

OpenAI Data Scientist interview experience

No offer

Easy

Negative

RemotePosted Nov 2024 · Candidate report, summarized by Dataford

I interviewed for an OpenAI Data Scientist role with a strong focus on ML loss behavior, and it felt approachable but very technical.

Recruiter/initial screening — a mostly problem-led discussion anchored on MNIST and cross-entropy loss.
Technical interview: loss/accuracy bounds — I was asked to compute lower/upper bounds on loss for a single training example given different classifier accuracy assumptions.
Technical interview: dataset-level extension — I derived the same type of loss bounds but for an entire dataset rather than one observation.
Technical interview: learning curves & overfitting — I discussed the expected shape of train/validation error curves and explained when/why training moves

Create a free Dataford account to unlock the full round-by-round breakdown, the questions asked and the outcome.

Preparing for this interview?