Research Scientist interview prep.

Behavioural questions to expect

1. Walk me through your CV.
2. Tell me about the paper or project you're proudest of.
3. Tell me about a weakness, a failure, or feedback you've received and worked on.
4. Why research scientist - and why this lab vs academia or eng IC?
5. Which research area would you want to focus on, and why?
6. Why the firm?
7. How would you describe the firm's research program + safety posture in your own words?
8. How does research actually create value at a frontier AI firm?

Technical concepts to master

• ML fundamentals - transformer + scaling + generalization

  Transformer + attention · Scaling laws + compute-optimal · Generalization + emergent capability · Optimization + training stability

• Experiment + evaluation design

  Hypothesis → measurable prediction · Controls + ablations · Confounds + bias · Eval design - capability + safety

• Alignment + safety + interpretability

  Alignment training (RLHF + DPO + Constitutional AI) · Alignment failure modes · Mechanistic interpretability · Responsible scaling + frontier eval

• Research process + paper-writing + community

  Paper structure + writing · Peer review + venues · Reproducibility + open-source · Research collaboration + advisorship

Practical drills

• Walk me through your strongest research paper or project end-to-end - I'll probe deeply on method, results, ablations, limitations.
• Pick a paper from the last 6 months that's relevant to the firm's research areas. Tell me about it + critique it.
• Design an experiment to test whether 'Constitutional AI training reduces sycophancy in a way that's not captured by standard helpfulness evals'.

Smart-question anchors

• Research areas + investments - the firm's research priorities, recent papers, future directions
• Safety + responsible scaling - the firm's safety posture, frontier evals, release policy
• Researcher autonomy - directed vs bottom-up research, publication freedom, conference attendance
• Compute + infrastructure - resources, scale, partnership with eng / infra
• Collaboration + community - academic collaborations, internal cross-team work, publication openness

Sourced from

Interview Query — Anthropic Research Scientist Interview Guide · Interview Query — OpenAI Research Scientist Interview Guide · Interview Query — Meta Research Scientist Interview Guide · Scaling laws + transformer architecture literature (canonical ML) · Alignment + safety literature (AI safety community) · ML research process + paper-writing community (NeurIPS / ICML reviewer guidelines)