The point
If you understand the one thing AI is doing, you will stop being surprised by it — both by how good it is and by how confidently it gets things wrong.
Walk through it
Here is the whole trick.
An AI chat tool has read an enormous amount of writing. Books, manuals, websites, forum posts, service records, recipes, arguments. From all of that it learned one skill very, very well: given some words, guess what words come next.
That is it. That is the engine.
It sounds like a party trick until you notice what falls out of it. To guess the next word in a plumbing invoice, it had to learn what plumbing invoices look like. To guess the next word in a polite refusal, it had to learn how people say no without burning a bridge. To finish the sentence “The customer said the unit was short-cycling, so I checked the…” it had to pick up a working sense of how HVAC troubleshooting goes.
So when you type a question, it is not looking anything up in a filing cabinet. It is writing the most likely answer, one word at a time, based on everything it has read.
Two things follow from that, and they matter more than anything else in this course.
One: it is spectacular at shaping words. Rewriting, summarizing, drafting, translating, tightening, making a rough thing polished, turning a mess of notes into a paragraph. This is its home ground. This is where you will get value in the first ten minutes.
Two: it will state a wrong fact with a perfectly straight face. Because it is guessing what sounds right, not checking what is right. Ask it for the phone number of a supplier and it may invent one that looks exactly like a phone number. Ask it for the code requirement on a deck railing height and it may give you a number that is close, or a number from another state, and it will sound equally sure either way.
People call this “hallucinating.” That word makes it sound rare and dramatic. It is neither. It is just what happens when a very good guesser hits something it does not know. Assume every specific fact, number, name, date, price, law or citation is unverified until you check it. Assume every piece of writing it does for you is a solid first draft.
The one rule that keeps you out of trouble
Use it for words. Verify the facts.
Write that on a sticky note. It covers almost everything.
What it is not
- It is not a search engine. It does not “know” today’s price of anything unless it goes and looks, and the free tools do not always look.
- It is not a person. It has no opinion about you, does not remember your last chat unless you turn that on, and does not get tired of dumb questions. Ask the dumb questions.
- It is not going to replace you. It cannot crawl a attic, read a customer’s face, or decide whether a job is worth taking. It can do the paperwork you hate at 9pm so you get an hour back.
- It is not one company. ChatGPT, Claude, Gemini and Copilot are different brands of the same kind of tool, like Ford and Chevy. Next lesson you pick one.
Do this now
Nothing to type yet. Just answer this out loud: what is the one task in your week you hate most and that is mostly words? Writing quotes. Answering the same five questions. Posting something. Chasing an invoice. Hold onto that answer — you are going to do that exact task in Lesson 1.4.
You’ll know it worked when
You can say, in one sentence, what AI is doing under the hood. “It’s guessing the next word, really well.” That is enough. That is genuinely enough.
If it goes sideways
- “So it’s basically useless if it makes things up?” No — it is useless for facts you cannot check and outstanding for words you can read. You are going to read everything it writes for you anyway.
- “Which one is smartest?” They trade places every few months and the difference matters far less than knowing how to ask. Next lesson.