A large language model (LLM) is a deep neural network trained to continue text. Given the words so far, it predicts the next token. Scaled up on a huge corpus, that skill becomes writing, summarizing, translating, and answering questions.
Tokens, context, and parameters
| Term | Meaning |
|---|---|
| Token | A chunk of text the model reads or writes, often a word or part of a word |
| Context window | How much text the model can see at once |
| Parameters | The learned weights. More parameters can mean more capacity, not automatically more accuracy |
| Temperature | How much variety to allow. Lower is more stable; higher is more varied |
What training does
- Pre-training. The model reads a very large public corpus and learns language, facts, and patterns by predicting the next token.
- Instruction tuning. Further examples teach it to follow questions and requests, not only to continue a paragraph.
- Preference tuning. Human rankings teach it which answers people prefer.
What an LLM does not do by itself
It does not look up your private files unless you connect them. It does not guarantee that a number, a citation, or a policy quote is correct. It produces the most plausible continuation of the prompt. For work that must be sourced, pair it with retrieval or with a human check.
A practical example
A relationship manager pastes meeting notes and asks for a follow-up email with three action items. The model drafts the email in seconds. The manager corrects the client name, the amount, and the date before sending.
What to remember
- An LLM predicts the next token. Useful behavior comes from scale and later tuning.
- The context window is the model’s working memory for that request.
- Treat unsupported facts as drafts until they are checked.
