6 steps · about 50 minutes
Getting started with AI
No prerequisites. Follow the steps in order: each one builds on the previous one. You can read the suggested entries or simply stick to the essentials summarised here.
0 / 6 steps read
Your progress stays in this browser: nothing is sent.
What exactly is AI?
Words before tools. “AI” covers programs that learn from examples rather than following rules written by hand. The AI everyone has been talking about since 2022 is generative AI: it produces text, images or sound.
Key takeaways
- An AI does not “think” like us: it spots patterns in huge amounts of data.
- Generative AI creates new content from an instruction.
How an assistant like ChatGPT works
Behind ChatGPT, Claude or Le Chat is a large language model (LLM). It splits your text into small pieces called tokens, then predicts the most likely continuation, word after word, using an architecture called the Transformer.
Key takeaways
- The model predicts how a text continues: that is what lets it write, summarise or translate.
- The length and cost of an exchange are measured in tokens.
Talking to it well
The quality of the answer depends a lot on the request. Giving context, a goal, an expected format and examples makes all the difference. This is called the prompt.
Key takeaways
- Say who you are, what you want to achieve and in what form.
- Feel free to correct and clarify: it is a conversation.
Knowing its limits
An assistant can confidently state something false: this is called a hallucination. Important facts should therefore be checked, especially figures, names and quotations.
Key takeaways
- Always check important information against a reliable source.
- Do not share personal or confidential data without knowing the terms of use.
When AI goes further
Companies connect AI to their own documents (RAG), specialise it (fine-tuning) or give it multi-step tasks (agents). These ideas explain most of the products you will come across.
Key takeaways
- RAG lets the AI answer from your documents rather than from the model’s memory alone.
- An agent does more than answer: it acts (searching, clicking, writing code…).
Find your bearings, then get practical
Who are OpenAI, Anthropic, Mistral or DeepSeek? What has happened since ChatGPT? The landscape gives you the map. Then describe your project and we will point you to suitable tools.
Key takeaways
- An application (ChatGPT) is not a model (GPT): most tools rely on a handful of large models.
- The right tool is the one that fits your need, your skill level and your constraints.
Your progress
Clears the progress saved in this browser.