AI words explained: the 15 you'll actually hear

Every new technology comes with a cloud of words, and AI has more than most. The good news: you only need about fifteen of them, and none of them is difficult once someone explains it without showing off.

Read this once. Then, when your grandson says “I just prompted the model,” you’ll know exactly what he means — and you’ll notice he’s using two of these words slightly wrong.

The five you’ll hear every day

AI (artificial intelligence). A broad label for computer programs that can do things we used to think needed a human — writing, answering questions, recognising a face in a photo. When people say “AI” today, they almost always mean the next word.

Chatbot. A program you talk to by typing (or speaking), and it types back. ChatGPT, Gemini and Copilot are all chatbots. The “chat” part is the point: it’s a conversation, not a search box.

Prompt. Whatever you type to the chatbot. “Write a birthday card for my sister” is a prompt. So is “What’s the capital of Peru?” Nothing more mysterious than that. People who say “prompting” just mean “asking well.”

Model. The engine behind the chatbot. ChatGPT is the shop front; the model is the machinery in the back. You’ll see names like GPT-5, Gemini, Claude — those are models. You don’t need to remember which is which.

Generate. To make something new — a sentence, a picture, a recipe. “Generative AI” just means AI that produces things rather than only sorting or finding them.

The five that explain how it works

Training. How a model learned. It was shown an enormous amount of text — books, websites, articles — and learned the patterns of how words follow other words. It did not memorise the internet, and it cannot look things up in its own head like a librarian.

Hallucination. The most important word on this list. It’s when the AI states something false with complete confidence — a made-up quote, a book that doesn’t exist, a wrong date. It isn’t lying; it has no idea it’s wrong. This is why you always check anything that matters.

Context. What the AI is currently “holding in mind” — your conversation so far. Inside one chat, it remembers what you said. Start a new chat and, usually, it’s a blank slate.

Token. A piece of a word. The AI reads and writes in tokens, not letters. You’ll only meet this word when something mentions a limit (“2,000 tokens”) — it’s just a measure of length.

Search. Some chatbots can now look things up on the internet while answering, and some can’t. If it can, it will usually show you links. If it doesn’t show links, assume it’s working from memory — and memory can be wrong.

The five you’ll see in the news

Large language model (LLM). The technical name for the kind of model behind chatbots. “Large” because it learned from a vast amount of text; “language” because words are what it deals in. If you hear “LLM,” think “the engine in ChatGPT.”

Deepfake. A fake photo, video or voice recording made with AI, realistic enough to fool people. Usually of a real person. This is the technology behind several of the scams aimed at people over 50.

Algorithm. An older word that predates AI: a set of rules a computer follows. “The algorithm” on social media decides what you see. Not the same as AI, though the two now overlap.

AGI (artificial general intelligence). A hypothetical future AI as capable as a person at everything. Much discussed, not yet real. When you see this word, you’re reading speculation, not news.

Open source. A model whose inner workings are published so anyone can inspect or run it. The opposite is a “closed” model, kept private by the company that built it. Neither is automatically better or safer.

Three words that are used wrong all the time

“The AI knows…” — It doesn’t know anything the way you do. It predicts what words are likely to come next. That’s why it’s brilliant at sounding right and only usually right.

“The AI is learning from me.” — Mostly not, in the moment. Your conversation doesn’t change the model while you’re talking to it. (What companies do with your words later is a separate question — we cover it here.)

“Ask Google.” — Search and chatbots are different tools. Search shows you pages other people wrote. A chatbot writes you a new answer. Search is better for facts you need to be sure of; a chatbot is better for help with a task.

If you remember only three

Prompt is what you type. Hallucination is when it’s confidently wrong. Model is the engine. With those three, you can follow almost any conversation about AI — and you’ll be ahead of most people having it.