Lesson 1.1 — How AI Actually Works

AI Isn't Thinking — It's Predicting

Here's something that might surprise you: AI doesn't understand anything. Not a single word. Not a single idea. It's not thinking about your question. It's not "figuring out" an answer the way you do when you work through a math problem or decide what to say to a friend.

So what is it doing? Predicting.

When you type something into an AI tool, it looks at your words and calculates: "Based on all the text I was trained on, what word is most likely to come next?" Then it picks that word, adds it to the sentence, and asks the same question again. And again. And again. That's it. The entire process is one long chain of next-word predictions.

Pattern Matching: The Engine Behind It All

The AI has seen so much text — billions of pages of books, articles, websites, and conversations — that it's gotten incredibly good at spotting patterns in language. It knows that after "The capital of France is," the next word is almost certainly "Paris." Not because it's been to Paris. Not because it understands what a capital city is. But because that pattern appeared thousands of times in its training data.

Answer: False. AI produces correct answers through pattern matching on training data, not through genuine understanding. It can give a perfect answer about gravity without understanding what gravity actually is — it's just predicting what words typically follow that question.

How AI Learns: Training Day

So how does an AI get so good at predicting? Training.

Think of it like this: imagine you gave someone a million jigsaw puzzles to solve, and every time they got a piece wrong, you told them. Over time, they'd get incredibly fast at recognizing which pieces fit together — not because they understand the picture, but because they've seen so many patterns of shapes and colors that they can predict the right fit almost instantly.

AI training works similarly. Developers feed the model enormous amounts of text. The model tries to predict the next word in a sentence. When it's wrong, the system adjusts its internal settings slightly so it does better next time. After billions of these adjustments, the model gets remarkably good at prediction.

The Training Loop

Training Data (billions of text examples) → Model predicts next words, gets feedback on errors → Billions of parameter adjustments = a trained model

Quick Check: What is training in AI models?

A) Teaching the AI to think like humans

B) Feeding it text data, having it predict, and adjusting settings based on errors

C) Writing rules for what answers it should give

D) Hiring people to teach it correct answers

Answer: B. Training is an automated process where the model sees massive amounts of text, tries to predict the next word, gets feedback on whether it was right, and makes tiny adjustments to its internal settings. This happens billions of times.

Parameters: The Dials Behind the Magic

You'll hear people talk about how many parameters a model has — "This model has 70 billion parameters!" — like it's a scoreboard. More parameters generally means the model can capture more complex patterns. But more doesn't always mean better for every task.

Here's a useful analogy: parameters are like the settings on a massive mixing board in a recording studio. Each slider controls something small — how much bass, how much reverb on one track. No single slider makes the song. But together, thousands of sliders set to just the right positions produce something that sounds amazing.

What Parameters Really Are

Billions of numerical settings, carefully adjusted during training, combine to produce responses that can sound remarkably human. The key thing to remember: there's no actual intelligence behind those dials. It's math — very, very impressive math — but math.

Quick Check: What are parameters in an AI model?

A) The questions users type in

B) Numerical settings adjusted during training that encode learned patterns

C) The websites the AI can access

D) Rules that programmers write by hand

Answer: B. Parameters are the internal numerical settings that get adjusted billions of times during training. They're what encode all the language patterns the model has learned. Nobody writes them by hand — they emerge from the training process.

Why This Matters For You as a Builder

"Okay, cool — AI predicts words. So what?"

This matters because when you understand what AI is actually doing, you become way better at using it. You stop expecting it to be an all-knowing oracle, and you start treating it like what it is: a powerful pattern-matching tool that's great at some things and unreliable at others.

Key Advantages for Builders

AI is great at drafting, brainstorming, and generating first versions — because those tasks play to its pattern-matching strengths.
AI can be confidently wrong — because it predicts what sounds right, not what is right. Your job is to check.
The quality of AI output depends on how you talk to it — because your words are the input it uses to start its prediction chain. Better input = better predictions.
AI doesn't replace your thinking — it accelerates it. You're still the one with the ideas, the judgment, and the final say.

Building with This Knowledge

Throughout this course, you're going to build a real project using AI as your tool. Understanding how that tool works is step one. You wouldn't use a power saw without knowing what it does — same idea here.

The patterns you see in how AI works will help you make better decisions about what tasks to give it, how to describe what you want, and when to verify its output before trusting it.

Try It: Guess Like a Model

Time to think like an AI. In this activity, you'll see a series of text patterns. Your job is to predict the next word, just like an AI model would. Read sentences with missing words, choose from options, and see how your prediction compares.

Which AI Tools Should I Use?

Throughout this course, you'll need access to an AI assistant. Here are free options that work well:

ChatGPT (free tier) — Available at chat.openai.com. Create a free account. The free version is sufficient for everything in this course. No credit card needed.
Claude (free tier) — Available at claude.ai. Create a free account. Generous free usage for conversations. No credit card needed.
Google Gemini (free) — Available at gemini.google.com. Free with a Google account. Good for students who already use Google tools.
Microsoft Copilot (free) — Available at copilot.microsoft.com. Free with a Microsoft account. Good for students who use Microsoft tools.

Which one should I pick? Any of them. They all work for this course. The prompting techniques you'll learn apply to all AI tools — that's the whole point. Pick one, stick with it, and switch if you want to compare later.

Note: If you're under 13, some AI tools require parental consent to create an account. Check the tool's age requirements and ask a parent or guardian to help you set up your account if needed.

Reflect & Write

Think about the last time you used an AI tool (or if you haven't yet, think about a time you used search or autocomplete). Write 2–3 sentences: Did you treat the AI's output as a final answer or as a starting point? Knowing what you now know about how AI works, would you do anything differently next time?

Project Checkpoint

No project action required for this lesson. You'll choose your project track and define your idea in Lesson 1.4. For now, start thinking about what you might want to build. Keep a mental (or actual) list of things you'd love to have — an app, a tool, a website, a dashboard — that you could build with AI as your assistant.

Level Up: Coming Next

Lesson 1.2 — Where AI Falls Apart. Now that you know how AI works, it's time to learn where it breaks. Hallucinations, bias, math failures — AI has some spectacular weak spots. You need to know them before you build with it.

Continue to Lesson 1.2 →