Who is it for?
Non-technical teams and professionals who use — or are about to use — generative AI in their daily work. No programming background required.
Objectives
By the end of the day, participants can:
- Explain what an LLM is and is not, and why it hallucinates
- Choose the right tool for a given task (general assistants, research tools)
- Write structured prompts and verify model output methodically
- Discuss AI’s societal stakes with informed, balanced arguments
Program
Morning — Understand and demystify
Opening demo — naive prompt vs methodical prompt. Same question, two ways of asking, radically different answers. You feel the core message before any theory: AI amplifies its user.
What AI is — and what it is not. Not Google, not a brain, not a conscience. A short history of AI, and where LLMs sit in the machine learning landscape.
How it works, really. Next-word prediction, tokens, context window, embeddings, training vs inference — the handful of intuitions that explain prices, hallucinations, and why the model “doesn’t know yesterday”. RAG, agents, multimodality, reasoning models. Where biases come from.
Live demo — three models, same question, one hallucination. ChatGPT, Claude, and Mistral side by side on a trap question, web search off. We watch a hallucination appear and learn to spot the signs.
Afternoon — Use, verify, criticize
The actors and the tools. The six generalist players and how to compare them (quality, price, open vs closed, ecosystem). Research and document tools: NotebookLM, Perplexity, document Q&A.
Hands-on session — prompting with method. Participants work on real (anonymized) cases from their own jobs: role, constraints, output format, level of proof. Audit and improve each other’s prompts.
Hands-on session — verification reflexes. Make a model read a document and extract facts; cross-check claims; spot-the-hallucination exercise on prepared outputs.
Society and the road ahead. Jobs, education, copyright, environment, privacy, concentration of power: the contradictory arguments, presented without preaching. What to expect in 5 years. Closing: a method to keep learning after the training.
Practical details
- One full day, on site at your offices (anywhere in France) or remote
- Live demos throughout; interactive hands-on sessions
- Includes post-training follow-up calls to review your AI projects