Your goal: explain how a language model works so that a PM could repeat it, go deeper on
request without losing the thread, and say where it fails and what you would do about it.
How this works (about 30 minutes)
- Felix is an interviewer at an AI company. He starts by asking which role you are interviewing
for, and pitches his follow-ups to it.
- One question, taken deep: first for a PM, then the engineer's follow-ups, a curveball, and a
trade-off built from your own explanation. Ask for any numbers you want; the details are made
up for the exercise.
- If you want to sketch or jot, say "can I use the notepad?" and Felix puts it up.
- No hints and no scores during the call. Your report comes after.
What scores well
- A plain line a PM could repeat, then one example, then the catch
- Tokens, embeddings, attention and the next-piece pick in a sensible order, on an example
- Fluent is not true, said without being asked
- Changing depth when the listener changes, and checking who you are talking to
- A call when your own advice backfires, with a reason and a line that would reverse it
Tips
- Take Gen AI Coach · How LLMs Work first.
- Use one sentence as your example and stay with it.