Module 2 of 5Lesson 1 of 2~12 min
How an AI assistant produces an answer
Knowing in a few sentences how an AI assistant builds its answer changes the way you use it. This lesson explains the principle without jargon and draws practical consequences: the tasks where the tool is reliable, and those where you need to feed it information or check what it says.
Lesson objective
By the end of this lesson, you will be able to explain in a few sentences how a generative AI assistant produces an answer, and deduce which tasks it handles reliably and which ones require you to provide information or check.
Topics covered
- generative AI
- how LLMs work
- AI assistant
- reliability
Where it fits
Understanding what these tools do
What happens when I ask an AI assistant a question, and what becomes of what I paste into it?
Lessons in this module
- How an AI assistant produces an answer (this lesson)
- Fabrications, sources and confidential data
What you will learn in the course
This lesson is part of the course Choosing and using AI tools at work
- Explain how a generative AI assistant works and what it does well or badly, so you know what to ask of it.
- Identify a tool's limits (fabrications, sources, freshness, data privacy) and the precautions to take before using it.
- Write a clear request (task, context, data, format, criteria) that produces a usable first result.
- Diagnose why a response falls short and fix it with a targeted clarification.
- Select the right tool for a deliverable using a grid of criteria (deliverable, data, sources, cost, language, integration) and justify the choice.
- Choose the right way of working for a task (conversation, project, deep research, agent) and the level of control that goes with it.
- Build your toolkit by comparing at least two tools on a real task with criteria written in advance.
Related courses
- Build reliable Claude Skills for your product workAdvanced · ~2 hr 30 min
- Prompting for product managers: delegate, describe, verifyJunior · ~2 hr 45 min
- Understand what LLMs do well, and where they failAll levels · ~2 hr