Artificial intelligence comes with its own vocabulary. Here are the words you’ll read most often in the news and in apps, explained in a sentence or two. Bookmark this page: we’ll keep adding to it.
The basics
| Term |
What it means |
| Artificial intelligence (AI) |
Systems that do tasks we associate with human intelligence: understanding, reasoning, creating. |
| Machine learning |
AI that learns from examples instead of following hand-written rules. |
| Deep learning |
Machine learning with many-layered neural networks. It’s the foundation of today’s chatbots. |
| Neural network |
A program made of layers of mathematical “neurons” that pass numbers to each other. Very loosely inspired by the brain. |
| Model |
The trained “brain” that makes the predictions. GPT, Gemini and Claude are families of models. |
| Generative AI |
AI that creates new content: text, images, music or video. |
Talking to an AI
| Term |
What it means |
| Chatbot |
A program you converse with by text or voice, like ChatGPT or Gemini. |
| Prompt |
What you type to the AI: your question or instruction. Tips in how to write good prompts. |
| Token |
The chunk of text the model works with. It can be a word or part of one. Prices and limits are usually measured in tokens. |
| Context window |
How much text the model can consider at once: your conversation, the documents you give it… Exceed it and it “forgets” the earliest parts. |
| System prompt |
Rules given to the model before the conversation, such as the tone or what it shouldn’t do. |
| Multimodal |
Able to understand or create several types of content: text, images, audio, video. |
How a model is made
| Term |
What it means |
| Training |
The process where the model learns from huge amounts of data by adjusting its numbers. |
| Parameters |
The billions of numbers a model adjusts as it learns. More parameters doesn’t always mean better. |
| Training data |
The text, images or sounds the model learns from. |
| Fine-tuning |
Training an existing model a bit more for a specific task. |
| Transformer |
The type of neural network almost all language models have been based on since 2017. |
| Attention |
The technique that lets the model decide which words matter for understanding each other word. |
| Knowledge cutoff |
The date the model’s training data runs up to. |
What you’ll read most in the news
| Term |
What it means |
| Agent |
An AI that doesn’t just answer but does things for you in several steps: searching, filling in forms, coding… |
| Hallucination |
An answer that sounds right but is false. We explain why in AI hallucinations. |
| Frontier model |
Each company’s most advanced current models, like Claude Fable 5.2 or Gemini 4 Argon. |
| Benchmark |
A standard test for comparing models, like an exam graded the same way for everyone. |
| Open source / open weights |
Models whose files are published for anyone to download and use, under the terms of their license. |
| Bias |
When AI treats a group unfairly or repeats prejudices present in its data. |
Missing a term?
If you’ve come across an AI term you don’t understand and it isn’t here, email us via our contact page and we’ll add it. And to see how it all fits together, start with what artificial intelligence is.
