Lesson 2: How ChatGPT and AI Tools Actually Work (No PhD Required)
Estimated time: 10 minutes
Difficulty: Beginner
Introduction
You've probably wondered: when I type something into ChatGPT and it responds, what's actually happening? Is it searching Google? Reading my files? Somehow thinking?
The answer is both simpler and weirder than you might expect.
AI isn't looking anything up. It's not browsing the internet in real time (unless you specifically turn that feature on). It's doing something more like... really, really sophisticated autocomplete. And once you understand how that works, you'll know exactly when to trust AI and when to double-check its work.
Let's pull back the curtain.
What Happens When You Hit Enter
Here's the basic flow:
You type something → AI processes your words → AI generates a response
Sounds simple, right? But here's what makes it interesting. When AI processes your words, it's not searching for an answer somewhere. It's predicting what response would be most helpful based on everything it learned during training.
Think of it like this: imagine reading every book in a large library, every article on Wikipedia, huge chunks of the internet, thousands of conversations. Now imagine someone asks you a question. You're not running back to re-read those books. You're drawing on patterns you absorbed. You know what kind of answer usually follows that kind of question.
That's basically what ChatGPT does. Except instead of reading a few thousand books, it trained on hundreds of billions of words. And instead of a human brain making connections, it's software finding mathematical patterns in language.
💡 Key Insight: AI doesn't look up answers. It generates them by predicting what words should come next based on patterns it learned from massive amounts of text.
The "World's Best Autocomplete" Analogy
You know how your phone suggests the next word when you're texting? You type "I'm on my" and it suggests "way." That's basic autocomplete.
AI is the same idea, just infinitely more sophisticated.
It learned that certain words, phrases, and ideas tend to follow others. When you give it a prompt like "Write a professional email apologizing for missing a meeting," it's predicting, word by word, what a helpful response looks like based on the patterns it learned.
Here's what makes it powerful: it doesn't just predict the next word. It predicts the next sentence, the structure of paragraphs, the tone that fits the situation, even what examples would be most relevant. All in a split second.
But here's what makes it imperfect: it's still just prediction. It's generating what sounds right, not what necessarily is right.
Why AI Sometimes Confidently Says Wrong Things
This is the most important thing to understand about AI: it doesn't know what's true. It knows what sounds true.
Let's say you ask ChatGPT, "Who won the Nobel Prize in Physics in 2019?" If that information showed up enough times in its training data, it'll give you the right answer (it was James Peebles, Michel Mayor, and Didier Queloz, by the way).
But if you ask about something more obscure, or you phrase your question in a way that leads it down the wrong path, AI will confidently generate an answer that sounds completely plausible... and is totally made up.
This is called a "hallucination." Not because the AI is lying or confused. Because it's doing exactly what it was designed to do: predict the most likely response. Sometimes the most likely-sounding response is wrong.
Real example: Someone asked ChatGPT to list scientific studies about a niche topic. ChatGPT provided five citations - author names, journal titles, publication dates, the works. Looked perfect. The person checked. None of those studies existed. AI made them up because that's what a list of scientific citations looks like.
This isn't a bug. It's how the system works. AI generates text that matches patterns, not facts.
💡 Key Insight: Treat AI like a brilliant intern who's read everything but sometimes makes things up to fill gaps. Trust it for drafts and ideas. Verify anything that matters.
Different AI Tools, Different Jobs
Here's where it gets practical. There isn't one "AI." There are hundreds of tools built for different purposes.
Text AI (ChatGPT, Claude, Gemini):
Best for conversations, writing, brainstorming, explaining things, summarizing documents. This is your general-purpose assistant.
Image AI (Midjourney, DALL-E, Stable Diffusion):
Creates pictures from text descriptions. "A golden retriever wearing sunglasses at the beach" becomes an actual image. Great for concept art, social media graphics, or just having fun.
Specialized AI (Grammarly, Canva AI, Jasper):
Built for one specific task. Grammarly checks your writing. Canva AI helps design graphics. Jasper writes marketing copy. They do one thing really well instead of trying to do everything.
Think of it like tools in a workshop. A hammer, a saw, and a drill all do different jobs. You wouldn't use a hammer to cut wood. Same with AI - pick the right tool for what you're trying to do.
The good news? For most everyday tasks, a general text AI like ChatGPT or Claude will handle 90% of what you need. You don't need to learn ten different tools on day one.
Models: The Engine Under the Hood
You'll hear people mention "GPT-4," "Claude," "Gemini." These are models - the engines that power AI tools.
Think of it like cars. ChatGPT runs on GPT-4 (made by OpenAI). Claude runs on, well, Claude (made by Anthropic). Google's tool runs on Gemini. Different companies, different engines, slightly different strengths.
You don't need to memorize this stuff. But it's helpful to know that when someone says "I tried it on GPT-4 and got a better result than GPT-3.5," they're talking about different versions of the same underlying technology. Newer models are generally smarter, faster, and less likely to make mistakes.
Here's the only part that matters to you: free tools usually give you access to an older, simpler model. Paid plans unlock the newest, most capable versions.
For learning and everyday use, the free versions are more than enough. You can always upgrade later if you find yourself using AI every single day and want the extra power.
What This Means for You
Now that you understand how AI works under the hood, here's how to use that knowledge:
When using AI:
- Be specific about what you want. Vague prompts get vague results.
- If the first response isn't great, try again with more detail. It's a conversation.
- Don't trust statistics, dates, names, or citations without checking them yourself.
- Use AI for first drafts, then add your own judgment and polish.
When reading about AI:
- If someone says "powered by GPT-4," that means it's using OpenAI's text model.
- If someone says "AI is always right," they don't understand how it works.
- If a tool promises it "never hallucinates," be skeptical. All AI can make things up.
You now know more about how AI works than most people using it every day. That puts you ahead.
Key Takeaways
- ✅ AI works like sophisticated autocomplete, predicting responses based on patterns it learned from text
- ✅ AI can confidently generate false information ("hallucinate") because it predicts what sounds right, not what is right
- ✅ Different AI tools serve different purposes: text, images, specialized tasks
- ✅ "Models" like GPT-4 or Claude are the engines behind the tools - newer is generally better
- ✅ Always verify important facts and treat AI output as a strong first draft, not the final word
Try It Challenge
🎯 Test AI's limits (takes 3 minutes):
- Open ChatGPT (chat.openai.com)
- Ask it: "Who won the Super Bowl in 2028?"
- Read the response carefully
Notice what happens. Since 2028 hasn't occurred yet, the AI can't possibly know the answer. Does it say "I don't know"? Does it make something up? Does it explain why it can't answer?
This little test teaches you something crucial about how AI handles questions it can't answer. Some versions will admit uncertainty. Others will confidently generate a plausible-sounding (but wrong) response.
Bonus: Try asking it something about your personal life, like "What's my dog's name?" It can't know - you never told it. See how it responds to questions it has no data for.
Quick Check
Question 1: Why does AI sometimes give wrong answers?
A) It's broken
B) It predicts what sounds right based on patterns, but doesn't know what's true
C) It's trying to trick you
D) You asked the question wrong
Answer: B) It predicts what sounds right based on patterns, but doesn't know what's true
AI generates responses based on patterns from its training data. It doesn't understand truth or facts the way humans do. It predicts what a helpful answer looks like based on what it learned. That's why it can sound confident even when it's completely wrong. This isn't a flaw you can fix by asking better questions - it's how the technology fundamentally works.
Question 2: When you ask ChatGPT a question, what is it doing?
A) Searching Google for the answer
B) Reading your files
C) Predicting the best response based on patterns it learned from text
D) Copying an answer from Wikipedia
Answer: C) Predicting the best response based on patterns it learned from text
ChatGPT isn't searching the internet in real time (unless you specifically enable web browsing). It's not reading your computer. It constructs a response by predicting what words and ideas best follow your prompt, based on patterns it learned during training. Think of it like autocomplete on steroids - it's generating text that matches patterns, not looking up information.
Next up: Lesson 3 - The AI Dictionary (Only the Words That Actually Matter)