06 — Behavioral Questions
AI orgs (Anthropic, OpenAI, DeepMind, Meta AI, xAI, Mistral, Cohere) have specific behavioral signals they probe for. Use the STAR structure (Situation → Task → Action → Result), keep stories to ~2 minutes, end with measurable impact.
Stories You Should Have Ready
Prepare 4-5 stories that you can flex to multiple questions:
- Ambiguous-problem story: open-ended problem, you scoped it, picked an approach, delivered.
- Cross-team / collaboration story: you depended on or unblocked another team.
- Failure / mistake story: real failure, with what you learned.
- Impact story: measurable business / research outcome you drove.
- Speed/scrappy story: short timeline, you cut scope intelligently and shipped.
Common Questions, Mapped
Anthropic-style (mission alignment, safety mindset)
Q. Tell me about a time you raised a safety / ethical concern about a project.
Q. Why Anthropic specifically? What about our research direction excites you? Have a specific paper or post in mind. Cite the technical substance, not just "I care about safety."
Q. How do you handle disagreement with a senior researcher? Probe: do you defer too much, or argue without evidence? Best answer: structured experiment that resolves the disagreement empirically.
OpenAI-style (impact, scale, ownership)
Q. Describe the most ambitious technical project you've shipped.
Q. Tell me about a time you had to make a decision with incomplete information.
Q. When have you pushed back on a product or research direction?
DeepMind-style (rigor, depth)
Q. Walk me through a paper you've read recently and what you'd do differently.
Q. Tell me about a result you initially believed but later disproved.
Q. How do you decide an experiment is "done"?
Meta / xAI-style (velocity, ownership)
Q. Describe a project where you owned the whole stack end-to-end.
Q. Tell me about a time you cut scope to ship.
Q. When did you last do something for a teammate that wasn't your job?
Anti-Patterns to Avoid
- "We"-itis: every sentence "we" — interviewer can't tell what you did. Use "I" for your contributions.
- Vague impact: "performance improved." Replace with: "p99 latency dropped 38%, from 1.4s to 870ms, in production within 3 weeks."
- Tech-stack tourism: listing tools without saying why you chose them or what tradeoffs.
- Hero narrative without humility: leave room for "and what I'd do differently."
- Unprepared "why us": shows lack of interest. Have one specific reason per company.
Compensation & Negotiation Talking Points
- Know your numbers: research target ranges on Levels.fyi for the company + level + location.
- Mention competing offers honestly (don't fabricate).
- Negotiate the equity refresh and starting bonus, not just base.
- Anthropic / OpenAI / DeepMind: a lot of comp is in equity / units; understand the vesting cliff.
Questions To Ask Them
Always have 5-7 ready. Best ones probe their actual day-to-day:
- "What's the most recent technical disagreement in this team and how was it resolved?"
- "Where do you think this team's research direction will be wrong in 2 years?"
- "What does the first 90 days look like for this role? What does success look like at month 6?"
- "How do priorities shift week-to-week? Walk me through last week."
- "What's a piece of internal infrastructure that you wish was 10× better?"
- "How does this team interact with safety / alignment / policy teams?"
- "What kind of person is not a fit here?"
Pre-Interview Routine
- The night before: re-read your 4-5 stories. Don't memorize, just refresh.
- Morning of: review the master cheatsheet (file 01) once. Don't cram.
- 30 min before: walk, water, no caffeine spike.
- During: take a beat before answering. "Let me think for 10 seconds" is a strong signal, not a weak one.