The AI words you keep hearing, explained
Tokens, hallucinations, agents, AGI: a friendly guide to the AI words that keep cropping up, what they mean and why they matter to you.
Video transcript
Tokens, hallucinations, agents, AGI. AI comes with a lot of jargon. Here are the words you keep hearing, in plain English.
Start with the basics. Machine learning means software learning from examples. What it learns is stored in a model. Large language models are the ones behind chat assistants: a kind of generative AI, because they create new text.
You type a prompt into an AI assistant. It reads and writes in tokens, small chunks of text, and its context window is how much it can keep in mind at once.
Many assistants are now multimodal. That means they understand more than text: you can show them a photo, talk out loud, or upload a document.
Then there are the quirks. A hallucination is a confident mistake, like a made-up fact. Sycophancy is when it tells you what you want to hear, rather than an honest view.
And three words you’ll hear more and more. AI agents carry out tasks for you, step by step. Open-weight models can be downloaded and run by anyone. And AGI, an AI as capable as people at almost anything, doesn’t exist yet.
Now you can follow the conversation. Read the full guide below for every word, with everyday examples.
In 30 seconds
- Chat assistants run on large language models, which read and write text in small chunks called tokens.
- AI’s two big quirks: hallucinations, or confident mistakes, and sycophancy, telling you what you want to hear.
- Agents act for you, open-weight models can be run by anyone, and AGI doesn’t exist yet.
A colleague mentions “hallucinations”, the news is full of “AGI”, and an app invites you to try its new “agent”. If AI talk sometimes sounds like a foreign language, you’re not alone. Here are the words worth knowing, in plain English.
The basics
AI, short for artificial intelligence, is software that learns from examples instead of following rules written by hand. Every word below is a part of it, a way of using it or one of its quirks. For the bigger picture, see what AI is and how it works.
- Machine learningLearning from examples rather than rules. Jargon busterMachine learning: The branch of AI where software learns from examples instead of being programmed rule by rule. is how your bank spots a card payment that doesn’t look like you.
- ModelThe app is the car; the Jargon busterModel: What training produces: a very large set of numbers that captures patterns, used to make predictions. is the engine. A “new model” means a new engine under the same bonnet.
- Generative AIJargon busterGenerative AI: AI that creates new text, images, sound or video, rather than only sorting or labelling things. creates something new, such as an email, a picture or a song, rather than just sorting what’s already there.
- Large language modelThe engine inside chat assistants. A Jargon busterLarge language model: An AI model trained on vast amounts of text to predict the next word. It powers assistants like ChatGPT, Claude and Gemini. has learnt from so much text that it can write fluently, word by word.
Using it
- Jargon busterAI assistant: A chat app such as ChatGPT, Claude or Gemini that answers questions and does tasks you describe in plain language.: the chat app you actually use, such as ChatGPT, Claude or Gemini. Each has a large language model inside.
- Jargon busterPrompt: What you type or say to an AI to tell it what you want.: your request. The more you explain what you want and who it’s for, the better the reply. Try our simple prompt recipe.
- Jargon busterToken: A chunk of text, often part of a word, that a language model reads and writes one at a time.: a chunk of text, often part of a word. Models read and write in tokens, and their limits are counted in them too.
- Jargon busterContext window: How much text an AI can take into account at once, including your conversation so far.: how much it can keep in mind at once. In a very long chat, early details can drop out, so start a fresh chat for a new topic.
- Jargon busterMultimodal: An AI that understands more than one kind of input, such as text, pictures, sound and video.: able to handle more than text. Show it a photo of a boiler error code, or simply talk to it.
Its quirks
A Jargon busterHallucination: When AI states something false as if it were true, because it predicts plausible words rather than checking facts. is a confident mistake: a made-up quote, statistic or book, presented as fact. It happens because AI predicts likely words rather than checking them. Here’s how to catch one.
Jargon busterSycophancy: When an AI tells you what you want to hear, agreeing with you instead of giving an honest view. is AI’s habit of telling you what you want to hear. For an honest view, ask “What’s wrong with this plan?” rather than “Isn’t this a good plan?”
What’s next
- AI agentAn Jargon busterAI agent: An AI that carries out a task in several steps, such as searching, clicking and filling in forms, rather than just replying. doesn’t just answer: it acts. Ask it to book a table and it searches, clicks and fills in the form.
- Open-weight modelAnyone can download an Jargon busterOpen-weight model: A model whose numbers are published, so anyone can download and run it themselves. and run it on their own computers, so their data never has to leave the building.
- AGIJargon busterAGI: Artificial general intelligence: a hypothetical AI that could match people at almost any thinking task. It doesn’t exist yet. would match people at almost any thinking task. Some companies say they’re building it; experts disagree on when, or whether, it’ll arrive.
Agents are the ones to watch, as AI starts doing jobs for you: here’s what AI agents can do, and how to stay safe as you use AI.
Explain what [the AI word you’ve heard] means in plain English, with an everyday example. Then tell me why it might matter to someone like me: [who you are, such as a nurse, a student or a retired teacher].
Check yourself
3 quick questions nothing is savedTools in this guide
Spotted a mistake? Tell us and an editor will check it.
