Generative AI for Beginners is Microsoft’s free AI course: 21 lessons, MIT licensed, just under 120,000 stars on GitHub. Reading it costs nothing and needs no account — no sign-up, no card, no API key.
The course: github.com/microsoft/generative-ai-for-beginners — the same lessons, easier to read on a phone: microsoft.github.io/generative-ai-for-beginners
Where to start
| Lesson | What you walk away with |
|---|---|
| 1. Introduction to Generative AI and LLMs | What these models actually do, in plain terms |
| 2. Exploring and comparing different LLMs | How to pick a model instead of defaulting to one |
| 3. Using Generative AI Responsibly | Where it makes things up, and what to do about it |
| 4. Understanding Prompt Engineering Fundamentals | The best-practice list — including the tip below |
| 5. Creating Advanced Prompts | Examples in the prompt, and when they earn their place |
That’s the useful half for a business owner. Lessons 6 to 21 are about building apps — go there only if you want to.
đź“‹ Paste this into your AI
Copy this into ChatGPT or Claude and fill in the two brackets. It takes a minute.
I run a [type of business — e.g. a plumbing company] and the first job I’d hand to AI is [that job — e.g. answering new leads]. I’m working through Microsoft’s “Generative AI for Beginners”, lessons 1–5: what these models do, picking a model, using them responsibly, prompt engineering fundamentals, and advanced prompts.
Turn those five lessons into a five-day plan, 20 minutes a day, where each day ends with something I apply to my business — not a practice exercise.
For day four, write the exact line I should add to my prompts telling you what to say when you can’t answer, worded for my business.
Then ask me the three things about my business you’d need to know to tailor the plan.
What comes back is a week of learning pointed at your business, plus one line you can paste into every prompt from now on.
The tip from lesson 4
Microsoft calls it giving the model an “out” — a fallback answer it can give you when it can’t complete the task. In their words, it “can reduce chances of models generating false or fabricated responses.”
Source: lesson 4, Prompt Engineering Fundamentals, in the best-practices table.
So it isn’t a fix for accuracy. It’s giving the model somewhere to go other than inventing something.
The three traps
- “Short video” isn’t short. Lessons 1 to 18 have one, and they run 10 to 20 minutes; lessons 19 to 21 have none yet. Read the lesson text first and watch only where you get stuck.
- Free to read, not always free to run. Every lesson is free. The optional code exercises need your own model provider, and the GitHub Models free tier the course leaned on retired at the end of July 2026 — Microsoft now points to Foundry Models or a model running on your own machine.
- The fallback line reduces, it doesn’t fix. It cuts invented answers. It does not make the model right, so anything that matters still gets checked by you.
Start this week
- Read lesson 1. Twenty minutes, nothing to install.
- Paste the prompt. Get your five-day plan and your fallback line.
- Use the line. Add it to the prompts you already run, this week.