AI terms in plain English.

The technology behind tools such as ChatGPT, Claude, Gemini and Copilot. It is trained on very large amounts of text and predicts useful responses to what you write. It is very good with language, but it does not know facts the way a database does.

AI that creates new content, such as text, images, audio or code, rather than only classifying or predicting. Large language models are the best known example.

The instruction you give an AI tool. Clear prompts that explain the task, the context and the format you want produce far better results.

When an AI tool states something that sounds convincing but is wrong, such as an invented reference or figure. It is the main reason a person should always check AI output before it is used.

The small pieces of text, often parts of words, that a model reads and writes. Usage limits and prices for AI services are usually counted in tokens.

How much text a model can take into account at once, including your prompt, any documents you share and its own answer. A larger window lets you work with longer documents.

An AI system that does more than answer a question. It can plan steps, use other tools and complete a task, such as gathering documents, checking them and drafting a summary, usually with a person approving the important decisions.

Retrieval augmented generation. The system first looks up relevant documents, such as your own policies or files, and uses them to answer. It makes answers more accurate and lets them cite their sources.

A way of turning text into numbers that capture its meaning, so a computer can find passages that mean similar things even when they use different words. It is what makes RAG search work.

Further training of an existing model on your own examples, so it follows a particular style or task more reliably. Most businesses get further, faster, with good prompts and RAG before they need fine-tuning.

The Model Context Protocol, an open standard that lets AI assistants connect to other systems, such as files, calendars or business applications, in a consistent way.

A model whose trained weights are published, so an organisation can run it on its own infrastructure rather than through a provider’s service.

Rules and checks around an AI system that keep it within safe limits, such as blocking certain topics, checking outputs or requiring human approval for some actions.

Structured tests that measure how well an AI system performs on the tasks you care about, so you can compare tools and spot problems before your users do.

An attack where hidden instructions in a document, web page or email try to make an AI tool do something it should not, such as revealing information. It matters most for agents that read external content and take actions.

A design where a person reviews or approves the AI’s work at important points. For most business uses it is the right default.