A plain-English AI glossary

A quick reference for the words that get thrown around. Each is defined plainly, with a pointer to the fuller explanation where there is one.

The basics

• Large language model (LLM) — a program that predicts text one chunk at a time; the engine behind Claude, ChatGPT, and Copilot. See the full explainer.
• Token — a short chunk of text the model reads and writes in; roughly ¾ of a word. See tokens and context.
• Context window — how much text the model can consider at once. Older parts of a long chat fall out of view.
• Prompt — what you type in. Better prompts get better answers.

How it learns

• Training — the process of tuning the model on lots of text. See how a model is trained.
• Training cut-off — the date after which the model knows nothing first-hand.
• Hallucination — a confident answer that simply isn’t true. The reason you verify.
• Fine-tuning — extra training that specialises a model for a task or tone.

Going further

• Embedding — text turned into numbers that capture meaning. See embeddings and RAG.
• RAG — answering from your own documents by retrieving relevant chunks first.
• Agent — a model given tools and a goal so it can take actions, not just answer. See AI agents explained.
• Prompt injection — hidden instructions that try to hijack a model. See the security explainer.

If a term here is new, follow its link before reading on — the rest of the guide assumes these.