AI terms, without the jargon

Every AI word you keep hearing, explained the way we would explain it to a friend. You do not need any of this to start, but here it is when a term trips you up.

Artificial intelligence (AI)

Software that can do tasks we used to think needed a human, like writing, summarising or recognising a picture. It is not alive and not thinking; it is a very capable tool.

Generative AI

AI that makes new content, words, images, audio or code, rather than just sorting or finding things. The chat tools you have heard of are generative AI.

Large language model (LLM)

The engine inside chat tools like ChatGPT and Claude. It is a system that has read an enormous amount of writing and is very good at predicting what words should come next. That prediction is how it talks with you.

Prompt

What you type to an AI. Your instruction or question. A clear, specific prompt gets a far more useful answer than a vague one.

Token

The small chunks of text an LLM reads and writes, roughly a word or part of a word. Tools count usage and length in tokens. You almost never need to think about them.

Hallucination

When AI states something false in the same confident, fluent tone it uses when it is right. It is not lying; it predicted what sounded right. This is why you verify any fact before relying on it.

Training data

The huge collection of text and other content an AI learned from. Most tools learned up to a certain date, which is why they can be wrong about very recent events.

Model

A single AI system, like a specific version of ChatGPT or Claude. New models are released regularly, usually a bit more capable than the last.

Context window

How much an AI can hold in mind at once in a conversation, including what you have said and what it has replied. Very long chats can push earlier details out of view.

Machine learning

The broader field of teaching software to find patterns from examples rather than being given fixed rules. LLMs are one kind of machine learning.

Multimodal

An AI that can work with more than just text, for example reading an image you upload, or listening to your voice. Most leading tools are now multimodal.

AI agent

An AI set up to take steps on its own toward a goal, such as booking or filling something in, rather than only answering. Useful, but it deserves closer supervision.

Inference

The moment an AI produces an answer from your prompt. When people say a tool is "thinking", this is what is happening.

Fine-tuning

Taking a general model and training it a little more on specific material so it is better at a particular job or style.

Bias

When an AI reflects unfair patterns from its training data, for example treating some groups or topics differently. A reason to keep a human checking important decisions.

Deepfake

A fake image, video or voice made by AI to look or sound like a real person. The reason a family code word and a second check on any money request now matter.

Guardrails

The limits a company builds into an AI to stop it doing harmful things. Helpful, but not perfect, so your own judgement still matters.

API

A way for one piece of software to talk to another. It is how apps connect to an AI behind the scenes. A term you will see but rarely need.

Open weight (open source) model

A model whose inner workings are released publicly so anyone can run or build on it, as opposed to one only used through a company’s app.

Prompt engineering

A grand name for the simple skill of writing clear instructions to an AI: giving it context, a role, the format you want, and refining from there.