Behavioral questions
Meta
In your own words, explain what a large language model is to someone without a technical background.
BehavioralLast seen Nov 20, 2026 · Software engineer
TeamworkTechnical impact
What they're looking for
This is a live explanation, not a story about the past. Treat the interviewer as someone with no technical background. They listen for whether you can turn a technical idea into plain words without saying anything false. Good answers use one simple picture, cover what the model does, how it learned to do it and where it goes wrong, and check that the listener is following. Clarity and accuracy both count: a simple answer that is wrong is as weak as a correct one full of jargon.
What works
- Start with one picture the listener already knows, such as the word suggestions a phone makes while you type, and build the rest of the answer on it.
- Say how it learned: it practiced guessing the next word across a huge amount of written text, and that practice is what lets it write and answer questions.
- Name the limits in everyday terms: it can sound sure and still be wrong, and it only knows recent or private facts if it is connected to them.
- Keep it to about two minutes, leave out words like parameters or transformers, and end by asking which part the listener wants to hear more about.
What hurts
- Using words such as tokens, neural networks or training data without saying what each one means.
- Saying the model thinks, understands or looks facts up like a person does, which leaves the listener with a wrong picture.
- Going so deep into how it works that the main idea gets lost.
Sample answer
Think of your own story first, then compare.