Loading TutorKit...
How LLMs actually work
The machinery of a large language model at inference — how a prompt becomes a response, step by step: tokenize, embed, attention, transformer layers, next-token probabilities, sample, then repeat. Goes deeper than the LLM overview in What is AI?. Written for people who want to understand AI properly — accurate, with the real mechanism and a few key formulas, kept concrete and explained.
Questions in this set
- What happens when you send a prompt?
- Tokens and tokenization
- Embeddings: turning tokens into meaning
- Attention: how the model uses context
- Inside the transformer
- Word order: positional information
- Producing the next token
- Generating a response (autoregression)
- Why does the answer pause, then stream?
- How does a conversation become one stream of tokens?
- The context window
- What an LLM is not doing
TutorWant me to explain it differently?
AI concepts can be dense. Tell me what's confusing and I'll find a new analogy.