Applied AI at work / Module 1
Module 1. How an LLM works and where it fails
Your role: General (any role). The exercises open on its variant. Change role
Time: 2 hours (19 minutes of video lessons, 70 minutes of exercise, 15 minutes of quiz). Your task in this module: an AI assistant has drafted a document from a source you hold, and some of its sentences are wrong. In the variant for your role it is a policy summary, a welcome note, a proposal email, a shift briefing, a contract summary, or (finance) a planning memo about a fictional fund, Pinecrest Meridian SICAV. You will find them, correct them and name the source that proves the correction. Everything you learn below serves that task, and the same routine works on any AI draft you receive at work.
All data in this module is fictional, so you may paste it into any tool, including a personal one on a personal device.
What you will be able to do
- Explain in plain words how an LLM generates text and why fluent output can be wrong
- Tell apart a plain chat, a tool grounded in your firm's files (such as Microsoft 365 Copilot) and an internal chatbot on a document store
- Recognise six failure patterns in AI drafts: invented facts, wrong regulatory references, out-of-date law, missed clauses in long documents, unchecked arithmetic and over-confident tone
- Run the five-step verification routine on an AI draft and name the source for every correction
- Write prompts that restrict the source, ask for quotes and allow NOT FOUND
5 video lessons, 19 minutes
- 1.1 What happens when you press Enter 3:48Free
- 1.2 The three kinds of tool at work, and what you may paste 4:01Free
- 1.3 Where it fails in finance work, and the check for each 4:02
- 1.4 The five-step verification routine, and prompts that make fewer errors 3:31
- 1.5 Why your employer asks for this, and where invented facts hide in your role 3:47
Exercise: Find what the AI invented
A variant for each role, opening on yours: General (any role). Fictional data, scored automatically against the answer key, with the model answer after you submit.
Module quiz
8 questions, pass mark 6. Passing it counts towards your course certificate.
Sources
- AI Act, original Article 4 and Article 3(4): https://web.archive.org/web/20250306021629id_/https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng (capture of 6 March 2025)
- AI Act, Article 4 as amended by Regulation (EU) 2026/1744: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4
- European Commission, AI literacy questions and answers: https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers
- CSSF thematic review on AI, May 2025, pages 5 and 23: https://www.cssf.lu/wp-content/uploads/Thematic_report_AI_2024.pdf
- CSSF annual report 2025 (French), PDF pages 18 and 25: https://www.cssf.lu/wp-content/uploads/CSSF_RA_2025_FR.pdf
- CSSF site search, "artificial intelligence": https://www.cssf.lu/en/search/artificial+intelligence
- Circular CSSF 24/856, point 35, Table 2: https://www.cssf.lu/wp-content/uploads/CSSF24_856eng.pdf
- Microsoft Learn, Data, privacy, and security for Microsoft 365 Copilot: https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy
- Microsoft 365 Copilot application card, Limitations: https://learn.microsoft.com/en-us/copilot/plan/microsoft-365-copilot-application-card
- OpenAI enterprise privacy (updated 8 January 2026), Wayback capture of 2 June 2026: https://web.archive.org/web/20260602132611id_/https://openai.com/enterprise-privacy/
- Kalai et al., Why Language Models Hallucinate (2025): https://arxiv.org/abs/2509.04664
- Liu et al., Lost in the Middle (2023): https://arxiv.org/abs/2307.03172
- PwC Luxembourg annual review 2025: https://www.pwc.lu/en/about-us/pwc-luxembourg-annual-review-2025.html
- Spuerkeess interview, The Banking Scene, 29 April 2024: https://thebankingscene.com/opinions/generative-ai-in-benelux-banking-spuerkeess-augmenting-employees