Module 1 · Lesson 1 of 18How Modern AI Actually WorksFree preview
What AI Really Is (and Isn't)
Cut through the hype: what today's AI assistants actually are, what they do well, where they fall short, and the one mental model that makes everything else click.
On this page
Maybe you've asked ChatGPT to draft an email and been surprised by how good it was. Maybe you've also asked it a simple factual question and received a confident, polished, completely wrong answer. If you've had both experiences, you've already met the central puzzle of modern AI: the same tool can be remarkably capable and surprisingly unreliable, sometimes in the same conversation.
Both behaviors have the same explanation. Once you understand it, you stop being surprised. You start knowing when to lean on AI, when to double-check it, and how to ask for what you actually need. That's what this first lesson gives you: a clear, hype-free picture of what these tools are, and a practical way to work with them starting today.
Learning objectives
By the end of this lesson, you will be able to:
- Tell the difference between AI, machine learning, generative AI, and large language models
- Explain in one sentence what an AI assistant is doing when it answers you
- Sort everyday tasks into "AI is strong here" and "handle with care"
- Get a noticeably better result by giving an assistant the context it needs
AI is a family of technologies, not one thing
"AI" gets used for everything from spam filters to self-driving cars, which makes conversations about it confusing. Here's how the main terms fit together, from broadest to most specific.
| Term | What it means in plain language | Everyday example |
|---|---|---|
| Artificial intelligence (AI) | The broad field of building computer systems that do things we associate with human intelligence | Spam filters, navigation apps choosing a route, chess programs |
| Machine learning | A kind of AI that learns patterns from examples instead of following hand-written rules | Your bank flagging an unusual purchase, streaming services recommending shows |
| Generative AI | Machine learning that creates new content: text, images, audio, video, or code | Drafting a cover letter, creating a mockup of a product image |
| Large language model (LLM) | A generative AI model trained on enormous amounts of text to understand and produce language | The engine behind ChatGPT, Claude, Gemini, and Microsoft Copilot |
When people talk about "AI" at work today, they almost always mean the last two rows: generative AI, and specifically the large language models inside chat assistants. That's the focus of this course.
One more distinction will save you confusion later. The model is the engine: the trained system that reads and writes language. The assistant is the car built around it: the app with a chat window, file uploads, web search, memory, and other features. When an assistant searches the web or reads your spreadsheet, that's the app giving the model extra tools. When it writes a clear paragraph, that's the model itself. This is why the same underlying model can behave differently in different apps.
The one-sentence explanation
Here's the mental model that unlocks everything else in this course:
A large language model is a prediction engine for language. Given some text, it predicts what text should come next, one small piece at a time.
That's it. When you type a question, the model isn't looking up an answer in a database. It's generating the response it predicts is most fitting, based on patterns it absorbed from an enormous amount of human writing during training.
Your phone's autocomplete is a tiny cousin of this idea. It suggests the next word based on what you've typed. Now imagine an autocomplete that has absorbed patterns from a vast library of books, articles, websites, manuals, and conversations, and that has been further trained to continue text in the form of a helpful answer. To predict well across all that material, it had to pick up grammar, facts, writing styles, how arguments are structured, how code works, and how people explain things to each other. That's why it can draft a solid project plan or explain a tax concept in plain English.
It's also why it sometimes gets things wrong. The model produces text that is plausible. For common topics, plausible and true usually line up. But when the model hasn't learned something well, it still produces fluent, confident text, and fluent isn't the same as accurate. You'll learn exactly why this happens in Lesson 1.3.
Three myths to drop today
Myth 1: "It looks things up." By default, the model answers from patterns it learned during training, not by consulting a source. Some assistants can search the web or read files you give them, and that helps a lot. But it's a feature you should confirm is being used, not something to assume.
Myth 2: "If it sounds confident, it's probably right." Confidence is a writing style, not a signal of accuracy. An assistant will often use the same assured tone for a well-established fact and for a detail it has effectively made up.
Myth 3: "You need to be technical to use it well." The most important skill is clear communication: explaining what you need, giving the right background, and reviewing the result with a critical eye. If you can brief a colleague, you can learn to brief an AI.
What AI assistants do well, and where to be careful
| Task | How AI typically does | Why |
|---|---|---|
| Drafting and rewriting emails, posts, and documents | Strong | Producing fluent language is its core skill |
| Summarizing material you provide | Strong, but check details | It works from your text, but can drop or blur key points |
| Brainstorming ideas and angles | Strong | It generates volume and variety quickly |
| Explaining well-known concepts | Strong | Common topics are well represented in what it learned |
| Specific facts, figures, names, and dates | Handle with care | It may produce a plausible guess instead of a true answer |
| Citations, quotes, and sources | Handle with care | It can invent realistic-looking references, especially without web search |
| Recent events | Depends on the tool | What it learned stops at a cutoff date; web search can help |
| Calculations on real data | Handle with care | More reliable when the assistant runs an actual calculation tool; still verify |
| Legal, medical, financial, or HR decisions | Input only | A qualified person must review and decide |
Notice the pattern: AI is strongest when it's working with material and goals you provide, and weakest when it's asked to recall precise facts on its own. Keep that one sentence in mind and you'll avoid most of the trouble people get into.
A practical way to think about it: the brilliant new hire
Here's a mental model that works well in practice. Treat an AI assistant like a brilliant, fast, extremely well-read new hire on their first day.
This new hire knows a lot about the world in general. They write quickly and well. But they know nothing about your company, your customers, your manager's preferences, or what you mean by "make it pop." They're also eager to please, so if your request is vague, they'll fill the gaps with reasonable-sounding guesses rather than stopping to ask. And you would never send their first draft to a client without reading it.
That framing tells you almost everything about how to get good results:
- Give context. Who's the audience? What's the situation? What's the goal?
- Be specific about the deliverable. Length, format, tone, and what must be included.
- Review before you use. You're responsible for what goes out.
Example: Compare two requests for the same task.
Vague: "Write an email about the delay."
Specific: "I manage customer support for a small software company. A bug has delayed our mobile app update by one week; the new release date is next Friday. Write a short email to customers explaining the delay. Be honest and apologetic without groveling, keep it under 150 words, and end with a placeholder link to our status page."
The first request forces the AI to guess about everything: who's writing, what's delayed, for how long, who's reading, and what tone fits. The second gives it what any good writer would need. The difference in quality is usually obvious, and the extra detail costs you about 30 seconds.
What's ahead in this course
This lesson is the foundation. Here's where the rest of the course takes you:
- Module 1 finishes the "how it works" picture: tokens, training, why hallucinations happen, and what a context window is.
- Module 2 tours ChatGPT, Claude, Gemini, and Microsoft Copilot, the features worth knowing, and how to pick the right tool for a task.
- Module 3 teaches a simple, repeatable prompt framework and how to iterate from a so-so answer to a great one.
- Module 4 applies it all to everyday work: writing, summarizing, research, brainstorming, spreadsheets, and images.
- Module 5 covers privacy, verification, bias, and intellectual property, so you can use AI without putting data or credibility at risk.
- Module 6 turns it into a habit with your own personal 30-day plan.
Along the way you'll build a personal prompt library, and you'll get a downloadable prompt starter pack and an AI safety checklist to keep next to your desk.
Try it
Time: about 10 minutes. Use any AI assistant you have access to; a free account is fine. For now, use made-up or public details only. You'll learn exactly what's safe to share in Module 5.
- Send a vague request. Type:
Write an email about a delay.Read the result and count how many details the AI had to invent. - Send a specific request. Pick a real but non-confidential situation from your work or life and fill in this template:
I am a [your role] at a [type of organization]. [Describe the situation in one or two sentences.]
Write a [type of message] to [audience] that [goal of the message].
Keep it under [number] words and use a [tone] tone.
Make sure to include: [any must-have details].
If you need information I haven't given you, use a placeholder in brackets instead of making it up.
- Compare. Which version would you actually send? How much editing would each one need?
- Test its limits. Ask a specific question about something you know very well but that isn't widely known, such as the history of a small local business, a niche process in your job, or a detail about your hometown. Look for anything that sounds confident but is wrong. Keep a note of what you find; you'll use it in Lesson 1.3.
Key takeaways
- "AI" is a family of technologies. The assistants you use at work run on large language models, a type of generative AI.
- An LLM is a prediction engine for language: it generates plausible text one piece at a time, based on patterns learned from huge amounts of writing.
- Plausible usually means correct for common topics, but not always. A confident tone tells you nothing about accuracy.
- AI is strongest when working with material and goals you provide, and weakest when recalling precise facts on its own.
- Treat it like a brilliant new hire: give context, be specific, and review everything before you use it.
Enjoyed this preview?
Unlock all 18 lessons, 2 templates, and your certificate for $149.