AI Glossary: LLMs, Tokens, Agents, RAG and Hallucinations Explained Simply

The basics

• Large language model (LLM): an AI system trained on large amounts of text that predicts and generates language. Chatbots such as ChatGPT, Claude and Gemini are built on LLMs.
• Prompt: the instruction or question you give a model.
• Token: a small chunk of text, often a word or part of a word. Models read, write and are priced in tokens.
• Context window: how much text a model can consider at once, including your prompt and its answer.
• Inference: running a trained model to produce an answer, as opposed to training it.

Quality and safety

• Hallucination: when a model states something false in a confident tone, such as an invented citation. Always verify important facts.
• Bias: systematic skew in outputs that reflects patterns or gaps in the training data.
• Guardrails: rules and filters that limit what a model will produce.
• Benchmark: a standard test used to compare models. Scores are useful signals but rarely predict how a model performs on your own task.

Making models more useful

• Retrieval-augmented generation (RAG): the system looks up relevant documents and gives them to the model so answers are based on your content.
• Fine-tuning: further training a model on specific examples to change its style or specialise it.
• Embedding: a numeric representation of text that lets software find similar meanings, often used for search.
• Multimodal: able to work with more than text, such as images, audio or video.

Agents and automation

• AI agent: a system that uses a model to plan steps and use tools, such as searching the web or editing files, to complete a goal with limited supervision.
• Tool use (function calling): letting a model trigger external software, like a calendar or a database.
• Workflow automation: connecting AI steps to ordinary software steps so a process runs without manual copying and pasting.

Open and closed models

• Open-weight model: a model whose trained parameters are published so you can run it yourself, subject to its licence.
• Closed (proprietary) model: available only through the provider's app or API.

Why the words matter

Vendors use these terms loosely. When you see "agent" or "AI-powered" in a product page, ask what the system actually does, what data it uses and where a human checks the work.