Understand what LLMs do well, and where they fail
Decide, task by task, what to delegate to an LLM, what to verify and what to keep, based on an accurate picture of how it works.
Know, for each task in your job, what you can hand to an AI assistant, what you must check and what you keep for yourself. You learn how a large language model (LLM) produces an answer, why it sometimes makes things up with total confidence, what it knows and up to when, what it sees of a conversation, and what web search, reasoning and image reading change. Every idea can be tested in five minutes in the tool you already use, on a free plan. You finish with a "delegate / verify / keep" grid applied to ten real tasks from your work.
What you will be able to do
- Explain how an LLM produces an answer (tokens, next-token prediction, pretraining then post-training) and why a fluent answer is not a verified answer.
- Anticipate hallucinations, sycophancy, poorly calibrated confidence and run-to-run variability, and choose the right check for each case.
- Locate what the model knows (training data, knowledge cutoff, uneven coverage) and what web search, provided documents and connectors change.
- Explain the context window as working memory (limit, degradation over long contexts, no memory across conversations without a dedicated feature) and derive working rules from it.
- Assess what reasoning (thinking) and reading images or documents bring, and what they do not guarantee.
- Sort the tasks of your job into "delegate / verify / keep" based on the cost of an error, how easy it is to verify, the model property involved and the value of doing it yourself.
Prerequisites
- Having asked a conversational assistant a few questions (Claude, ChatGPT, Gemini, Mistral's Vibe or Copilot); a free plan is enough for every exercise.
- No technical or mathematical skills are required.
- This course explains how models work and where they fail. To write good requests, continue with "Prompting for Product Managers: delegate, describe, verify"; to choose your tools, "Choosing and using AI tools at work"; to know which data you can share, "Use AI at work without exposing your data or your company".
Syllabus
What will I be able to do at the end of this course, and in what order?
- Course overview3 steps
Objective · By the end of this overview, you will know what you are going to produce (a "delegate / verify / keep" grid applied to ten tasks from your job), which running case the course follows and which four modules take you there.
Where does an AI answer come from, and why can it be wrong while sounding right?
Objective · By the end of this lesson, you will be able to explain how an LLM produces an answer (tokens, next-token prediction, pretraining then post-training) and to infer, for a given task, whether the main risk is an invention or a mere clumsiness of form.
Objective · By the end of this lesson, you will be able to spot the passages of an AI answer that are exposed to hallucinations, recognise sycophantic behaviour, take run-to-run variability into account, and choose for each case a check that does not rely on the model's own opinion.
What does the model know on its own, what do tools add, and what does it retain from our conversation?
Objective · By the end of this lesson, you will be able to say, for a given question, whether the model can answer from memory, whether you need to turn on web search or whether you must provide a document, and to choose the matching check.
Objective · By the end of this lesson, you will be able to explain what the model sees of a conversation, why it sometimes overlooks an instruction or a passage in a long document, and to organise your work (conversations, documents, summaries) to keep answers reliable.
What changes when a model "thinks" before answering or reads images, and what does not change?
Objective · By the end of this lesson, you will be able to decide when to turn on an assistant's reasoning mode, and to spot in the reading of an image, screenshot or PDF the elements that need checking.
For each of my tasks, what do I hand to AI, what do I verify and what do I keep?
Objective · By the end of this lesson, you will be able to sort a task into delegate, verify or keep using four questions (cost of an error, verifiability, model property involved, value of doing it yourself), and to name the planned check for each task delegated with control.
Objective · By the end of this lesson, you will have produced your "delegate / verify / keep" grid on ten real tasks from your job, tested at least three of them in your tool, and planned how to apply it over 7 and 30 days.