LLM Decoding Stopping Time / Probability Distributions

Problem Overview

You are given an oracle LLM whose inference/problem-solving time T follows some probability distribution with known expected value E[T] = 1. The interview is structured as a multi-cell Jupyter notebook (or whiteboard) with escalating sub-questions: 1. Use the Markov inequality to upper-bound P(T > 5). 2. If you are allowed…

• 5 candidate-reported follow-ups — the exact probes interviewers asked, with the trigger for each
• The rest of the problem statement — full requirements, constraints, and edge cases
• Approach and trade-offs — what passing candidates did, and the mistakes that sink people