Week 24. Communication and negotiation
Learn in the app: tutor, coding problems →
Core: research pitch, job talk, 15 STAR stories, negotiation scripts, a paper review and the last mock · Depth: a second paper review, a position on AI safety · ≈ 10 h core / 15 h total
Technical rounds test what you can do; the job talk, behavioral interviews and negotiation decide whether you get hired and on what terms. Strong candidates fail these formats too, not from lack of knowledge but because they did not rehearse. This is the last week because you can only talk about work that is already done.
- Job talk: 30 minutes, one piece of work at the center plus related work in the background. Run it through 3 times on recording
- Research pitch at three lengths: 1 minute / 5 minutes / 20 minutes. Adapt it to the role: hitting the keywords helps a tired interviewer believe your profile is relevant
- Behavioral: write out 15 STAR stories (situation → task → action → result) and map them to the typical questions in advance. Strong candidates fail behavioral interviews precisely because they decided no preparation was needed
- Questions about AI safety and societal impact: prepare a position
- Reviewing someone else's paper: 2 times, 30 minutes each, using the template from the rubrics, on papers from the "Sources" of weeks 19–22. A research round can be about someone else's work too
- Negotiation: before every call with a recruiter, write down in advance what you are and are not willing to share, and prepare exact phrases. The first offer is designed to leave room for bargaining
In plain terms. A behavioral question: "Tell me about a failure." Bad: "Once we missed a deadline, but the team pulled through and I learned a lot." There is no situation, no action of yours, and nothing that changed afterward. Good (STAR: situation, task, action, result): "We were choosing between two models before a release, with a two-week deadline. I was responsible for evaluation. I did not check the overlap between the validation set and the training data, and for two weeks we compared the models on a contaminated set. I found the error when the result did not reproduce on a fresh set. I told the team and rebuilt the evaluation. We lost a week. Since then the overlap check has been a CI test, and on the next project it caught a leak before training." The failure is named directly, the action is yours, and the takeaway is verifiable.
Negotiation follows the same rule. Bad: "Well, I am expecting roughly…" You named the first number, and with nothing to back it. Good: "It is important for me to see the full offer first. Then I can compare it with my other processes and share my expectations."
Score each format using the rubrics for spoken formats: the job talk, research pitch and behavioral answers are scored against them after every recording.
Interview question of the week: "Tell me about a time you disagreed with a technical decision by a colleague or a manager." This is behavioral, answered with STAR: 2 minutes of story and a minute for follow-ups. Structure: (1) Situation, two sentences: project, scale, deadline, no backstory; (2) Task: your role and what was at stake, in one sentence; (3) Action: half the time, and "I", not "we": how you replaced the argument with a test (for example, an experiment with a criterion written down in advance, week 22) and how you talked to the person who disagreed; (4) Result: a number, and what changed in the decision or the process; be honest if you were not the one who turned out right; (5) the lesson in one sentence: what you have done differently since; (6) expect "what would you do differently?", because "nothing" reads as a lack of reflection. Take the story from your 15; do not make it up on the spot.
Week outcomes
- I can give a 30-minute job talk on recording and finish on time with room left for questions.
- I can present my work at three resolutions (1, 5 and 20 minutes) without preparation.
- For any typical behavioral question, I can pick a story from my 15 and tell it with STAR in 2 minutes.
- Before a call with a recruiter, I can state what I will and will not share, and say my prepared phrases.
Self-check
- A 1-minute research pitch, right now, out loud.
- Tell me about your failure using STAR: what exactly did you do differently afterward?
- Your position on AI safety in 2 minutes: what risk do you consider the main one, and what should be done about it?
Mock interviews of the week (12 of 12). Experiment design following session C of week 23, 45 minutes, right after reviewing someone else's paper: "How would you test this paper's main claim more cheaply than the authors did?" A research round often goes exactly like this.