Vol I · Section 1
How LLMs work
Tokens, transformers, prompting and why models make things up.
1What does an LLM actually compute when it answers a question?2What is a token, and why should an engineer care about tokens rather than words?3Explain the transformer and attention at whiteboard level.4Why do LLMs hallucinate, and what actually reduces it?5How do you choose temperature and other sampling settings?6What are the trade-offs of a very large context window?7When do you use zero-shot prompting, few-shot examples, or chain-of-thought?8When is a reasoning model worth paying for?9How do you get reliable structured output from an LLM?10What is a knowledge cutoff, and how do you design around it?