12 questions to validate the fundamentals: what an LLM is, the tool landscape, first exchanges and limits to verify.
Answer everything, then click Check: your score and explanations will appear.
0.1 · Recall
1. What does an LLM fundamentally do?
The LLM predicts plausible text: brilliant on form, no guarantee on facts.
0.1 · Application
2. Amine asks his assistant “847 × 293?”. What should he do with the result?
LLMs calculate approximately: every figure is verified at the source before use.
0.1 · Arbitration
3. Two drafts: one fluent but unverified, one rough but checked. Which one to share?
Fluency is never proof: share the verified, not the plausible.
0.2 · Recall
4. What is criterion No. 1 for choosing an assistant AT WORK?
Company data → approved tool; integration with the ecosystem does the rest.
0.2 · Application
5. Amine must write with internal figures. Which tool should he choose?
Internal data → company-approved business tool only; the rest to the favourite assistant.
0.2 · Application
6. What do Copilot on the Microsoft side and Gemini on the Google side mean for the choice?
Copilot lives in Office, Gemini in Google Workspace: integration makes the difference.
0.3 · Application
7. Amine mixes everything into one thread and answers become confused. What advice?
One clean thread per topic prevents old contexts from polluting answers.
0.3 · Application
8. Amine changes topic after a long conversation. What should he do?
Memory holds within the thread, not between threads: new topic = new thread + context.
0.3 · Arbitration
9. Before his first professional use, which setting should Amine check first?
Privacy settings are checked from sign-up, before any real use.
0.4 · Recall
10. What is a hallucination?
Undetectable by tone: figures, references and quotes are always verified.
0.4 · Application
11. The assistant cites three books with links. What should he do before sharing them?
Memory-generated links may be broken: only the real content of pages counts.
0.4 · Arbitration
12. Document for a client with AI-assisted figures. What verification?
The higher the stakes, the heavier the verification: external use requires systematic checking.
What's next?
Score ≥ 80%: move to Level 1 — the art of prompting (context, iteration, examples, mistakes to avoid).
Score between 50% and 80%: revisit the failed section(s) (0.1 LLM principle, 0.2 tools, 0.3 first exchanges, 0.4 limits), reread the explanations above, then retake the test.
Score < 50%: reread all of level 0 while applying each 48h challenge (3 tests, sorting rule, first two threads, provoked hallucination) before retrying.
And next? Keep the golden rule: AI proposes, you verify — level 1 makes your requests formidable.