← all articles

The AI cheat sheet
in plain business terms

glossary · 6 min · interactive article ✦

AI jargon too often serves to impress rather than explain. Here are the terms that come up in meetings, decoded in one sentence — click a card to flip it. You'll never have to nod along without understanding again.

LLM
click to flip ↻
A "large language model": trained to predict the next word over huge texts. It knows nothing, it generates plausible output — so we verify.
Prompt
click to flip ↻
Your instruction to the AI. A good prompt is a brief: role, context, bounded task, expected format.
Hallucination
click to flip ↻
When the AI invents a wrong-but-credible answer (a date, a source). Inevitable: hence verifying important facts.
RAG
click to flip ↻
We plug the AI into YOUR documents so it answers from them, with sources — without retraining. Fixes "it doesn't know our docs".
Fine-tuning
click to flip ↻
Retraining a model on your data to change its behavior or style. Heavier than RAG; useful when you want to change how it answers.
Agent
click to flip ↻
An AI that doesn't wait: it chains actions, consults tools and reports back to reach a goal.
MCP
click to flip ↻
The "universal plug" that connects an agent to your tools and data in a standardized, governable way.
Token
click to flip ↻
The unit of text chunking for the AI (often part of a word). Billing and limits are in tokens: hence "context window".
Generative AI
click to flip ↻
An AI that produces new content (text, image, code) rather than only classifying or predicting a value.
AI Act
click to flip ↻
The European regulation that classifies AI uses by risk level and imposes obligations accordingly.
take it with you · free

The AI glossary as a 1-page PDF

All these terms (and a few bonuses) in an A4 cheat sheet, to print and keep at hand in meetings. Leave your email to receive it, or download directly.

or download directly, no email →

Frequently asked questions

What is an LLM?
A Large Language Model is a system trained to predict the next word from huge amounts of text. It doesn't "know" anything: it generates plausible answers. Hence the importance of checking facts.
What's the difference between RAG and fine-tuning?
RAG connects the model to your documents so it answers from them, without retraining. Fine-tuning retrains the model on your data to change its behavior. RAG answers "it doesn't know my documents"; fine-tuning answers "I want it to respond a certain way".
What is an AI agent?
An agent is an AI that doesn't just answer: it chains actions, consults tools and reports back, to achieve a goal. MCP is the "universal plug" that lets it connect to your tools in a governable way.
Vocabulary is the first step.

In IAPLC, each term becomes a reflex: you practice it on YOUR cases, through 30 levels generated for your IT role.

Explore the course →