Peer mocks

Practise with a person, too

The AI interviewers grade you on the rubric. A human partner adds what they can't: real interruptions, silence and nerves. Here's a kit to run a fair peer mock with the same rubrics.

How to run a peer mock

  1. 01Pair up. Find a partner preparing for a similar role. Book 90 minutes: two 30-minute sessions, two debriefs, a short break.
  2. 02Pick a round. Choose a role-play lesson below. The interviewer prints its script; the candidate doesn't read the score sheet first.
  3. 03Run it closed book. Cameras on, no notes beyond what a real loop allows. The interviewer keeps time and stays in character.
  4. 04Score and debrief. Score each rubric line, then debrief: the candidate first, then the two lowest lines with one moment each.
  5. 05Swap. Change roles on a different round, so neither of you has seen the scenario.

Weekly loop night

A 90-minute format for a study group of four to eight, same time every week.

  1. 0:00Check-in: each person names the round they're weakest at (from their scorecard).
  2. 0:05Session 1: 30 minutes in pairs, on that weakest round.
  3. 0:35Debrief 1: 10 minutes with the score sheet.
  4. 0:45Session 2: swap roles, 30 minutes.
  5. 1:15Debrief 2, then the whole group shares one fix each and books next week.

Bring your scorecard so the group works on everyone's weakest round, not their favourite one.

Interviewer scripts

Built from each lesson's own text and rubric. The AI interviewer's private brief stays private, so the human version is a little plainer: play it straight and use the score sheet. Free rounds are listed now; the rest appear with full access.

Interviewer script · C5

A Values Interview

About 32 minutes · pass mark 70%

Your part

You play Sam Okoye, Values interviewer (practice) at a frontier AI lab. Stay in character for the whole session.

“Hi, thanks for making the time. I'm Sam. This conversation is less about your technical skills and more about how you think and what you care about. There are no trick questions. To start: what's pulling you toward working at an AI lab right now, as opposed to anywhere else you could go?”
  • Open with the line above, word for word, then let the candidate lead.
  • Ask one follow-up at a time: “Why that?”, “What would you measure?”, “What if that fails?”
  • Push back once on their main decision and see if they defend it with reasons or numbers.
  • Don't coach or hint during the session. If they ask for a fact the brief doesn't give, say “I'd have to check”.
  • Keep time. Give a two-minute warning and stop on time.

What the candidate has read

The lesson's own text. Read it before the session so you know the scenario.

The culture interview is a real filter at mission-driven labs, and lesson 01 sets out what's known about it and how much each claim deserves to be trusted. Nobody can tell you the "right" answers, because there aren't any. What interviewers look for is specific, honest evidence of how you behave. How this practice works On the right, Sam will interview you for about eight turns. Sam is a fictional practice interviewer played by Claude, not a real person at any lab. Answer the way you would in the real room: out loud first if it helps, then type. Use real situations. "I would..." is weaker than "I did...". When you've answered at least two questions, press End and get feedback for a scored debrief against the rubric below. The shape of a strong answer Use STAR loosely (lesson 02 covers it in depth), and spend most of your words on the last two letters: Situation: one or two sentences of context. Task: what was at stake and what was yours to do. Action: what you did. Say "I", not "we". Result: what happened, including what went wrong, and what you'd do differently now. Before you start Have three true stories from your story bank ready: a disagreement on principle, a mistake that affected other people, and a time you slowed something down (or wished you had) because it wasn't ready. If you did lesson 03, you've already written two of them. Most values questions can be answered with one of the three. For the mission question, use the four moves from lesson 04.

Score sheet

Score each line after the session, then add them up. 100 points in all.

Rubric lineScore

Specific, real examples

Answers with real situations: who, what they personally did, what happened. Not hypotheticals or 'I would always...'.

___ / 25

Honest and self-aware

Owns mistakes without spin, admits uncertainty, describes what they learned and what changed.

___ / 25

Judgment under tradeoffs

Weighs speed, safety and people's interests explicitly; can say when they'd push back or say no, and how.

___ / 20

Motivation beyond money and hype

Motivation is concrete and tied to the work or the mission; engages genuinely with the mission question rather than flattering.

___ / 15

Direct communication

Answers the question asked, concisely; responds to follow-ups instead of repeating a script.

___ / 15

Debrief (10 minutes)

  1. The candidate says how they think it went, first.
  2. You give the total and the two lowest lines, each with one moment from the session.
  3. One thing to keep, one thing to fix first. Then swap roles.

Want the AI-graded version too? The candidate can run the same lesson on the site afterwards and compare the scores.