I’m starting out

AI in 12 terms

Twelve words are enough to understand what everyone else is talking about. They are arranged in the order in which they explain one another: each one builds on the previous one, and together they climb the four floors.

Beginners, no prerequisites · 12 terms · 25 minutes · final quiz

The sequence

0 / 12 seen

Generative AISystems that produce text, images or sound, instead of only classifying or predicting a number.
ModelThe engine: a machine trained once and for all, which predicts what comes next in a text.
Large language model (LLM)A model trained on enormous quantities of text. It handles language remarkably well, without knowing anything about the world.
TokenThe unit the model manipulates: a piece of a word. Length and cost are both measured in tokens.
PromptWhat you write to the system. How you word it changes the answer, because it changes what the model has in front of it.
ContextEverything the model has in front of it at the moment it answers: your question, the history, the documents added.
HallucinationA false answer stated with confidence. It is not a breakdown: it is the normal behaviour of a machine that predicts what is plausible.
AssistantA product that waits for your message and answers. The model is inside it, but the model is not what you are using.
ToolWhat the system is allowed to do outside of text: read a file, call a service, send a message.
AgentA system that pursues a goal across several steps and chooses the next one itself.
Automation or augmentationReplacing a task, or equipping the person who does it. That choice decides everything else.
Shadow AIAI used inside an organisation without anyone having decided on it, or set any bounds around it.

Your progress stays in this browser. No account, no data sent.

What to remember

The summary

There is an engine, and there is everything built around it

The model predicts text, with no memory and no access to the world. Everything else (the instructions, the documents, the tools, the limits) is ordinary software installed around it. Most arguments about “AI” are in fact about what has been built around the engine, not about the engine itself.

Hallucination is a property, not a defect to be fixed

A machine that predicts the most plausible continuation will sometimes produce something plausible and false. You do not remove that behaviour: you set bounds around it, by giving the model the right documents to read and by checking what matters.

An assistant answers, an agent pursues

As long as you prompt again at every step, you are using an assistant. The day the system chains several steps on its own towards a goal, the nature of the problem has changed: cost, monitoring and stopping points become subjects in their own right.

The question that decides is not a technical one

Automating a task or equipping the person who does it: that choice commits skills, responsibilities and trust. And if nobody asks it, it gets settled anyway, silently, by undeclared use.

The final quiz

3 questions · click your answer

Level 1 · Recognise

Why does a language model sometimes invent a reference that does not exist?

Level 1 · Recognise

You paste a long document into a conversation, then close the tab. What is left to the system?

Level 2 · Distinguish

In your team, several people use a consumer assistant to process client documents, without this having been planned. What is this called?

v2026-08 · pathCite this path