What is an LLM
In interviews, how do you explain what a large language model is and how it generates text?
Answers use simple, clear English.
Audio N/AQuick interview answer
An LLM is a neural net (usually a Transformer) trained to predict the next token on massive text. Given a prompt, it samples a probability distribution over the vocabulary repeatedly (decoding: greedy/temperature/top-p). It doesn’t ‘look up’ a database by default — knowledge is compressed in weights and can be wrong (hallucination).
Detailed answer
An LLM is a neural net (usually a Transformer) trained to predict the next token on massive text. Given a prompt, it samples a probability distribution over the vocabulary repeatedly (decoding: greedy/temperature/top-p). It doesn’t ‘look up’ a database by default — knowledge is compressed in weights and can be wrong (hallucination). Products add tools/RAG for grounding.
Real example & use case
Chatbot streams tokens; temperature 0 for deterministic extraction.
Pros & cons
Pros: flexible language interface. Cons: not a source of truth alone.