Generative AI: 10 key words to know
Model, token, hallucination, prompt… The vocabulary of generative AI explained simply, with what each term changes for you.
Updated on
Conversations about AI are riddled with jargon. To follow the news without dropping out, about ten words are enough. Here they are, with no jargon to explain the jargon.
1. Model
A model is the trained “brain” that produces the answers. ChatGPT, Claude and Gemini each rely on a model. When a provider announces a new version, it usually means a new model, more capable than the last. To understand how this “brain” produces its answers, see what an LLM is, explained simply.
2. Generative AI
This is the family of AI that creates content (text, image, code) rather than merely classifying or predicting. An assistant that drafts an email is doing generation. It is what sets it apart from “classic” AI, which classifies and predicts.
3. Prompt
The prompt is your instruction: the question or command you give. The quality of the answer depends directly on the quality of the prompt. It is the most underrated lever.
4. Token
Models do not read words but “tokens,” fragments of text. A common word is often one token; a long word, several. It is the unit used to measure length and, often, price.
5. Context window
This is the amount of text a model can keep “in mind” at once. The larger it is, the more documents you can submit at once without it forgetting the start.
6. Hallucination
When a model states something false with confidence, we call it a hallucination. It is not a rare bug: it is a structural limit. Hence the rule: check what matters. The mechanism is detailed in why AI “hallucinates”.
7. Fine-tuning (specialisation)
Adapting an existing model to a specific use by re-training it on targeted data. Useful for a particular tone or domain, but rarely necessary for everyday use.
8. RAG (retrieval-augmented generation)
A technique that connects a model to a base of documents so it answers from your content rather than from its memory alone. It is what lets an assistant cite your own files.
9. Agent
An assistant that does not merely answer but chains actions together to reach a goal. You give it an objective, it handles the steps.
10. Multimodal
A multimodal model understands several formats: text, image, sometimes sound or video. You can show it a photo and ask a question about it.
Key takeaway
You do not need to be an engineer to follow AI. With these ten words, you can decode the essentials of the announcements. The rest are mostly commercial variations on these same ideas. To dig deeper into these notions, browse our deep dives.
Sources
Frequently asked questions
What is a token in AI?
Models do not read words but tokens, fragments of text: a common word is often one token, a long word several. It is the unit used to measure length and, often, price.
What is the difference between a model and an agent?
A model is the trained brain that produces the answers (ChatGPT, Claude and Gemini each rely on a model). An agent is an assistant that does not merely answer but chains actions together to reach a goal you give it.
What does context window mean?
It is the amount of text a model can keep in mind at once. The larger it is, the more documents you can submit at once without it forgetting the start.
What is RAG?
Retrieval-augmented generation connects a model to a base of documents so it answers from your content rather than from its memory alone. It is what lets an assistant cite your own files.