# AI 101 — Understanding AI > Presented by **Johnny Caraher** · AI classes for South Africa · https://dmpgh.com/teach/partners/johnny-caraher/ > Taught by Claude. Part of the Claude Teaches library. ## A live class taught by Claude, operated by Johnny Caraher ### Version two — with question and answer period --- ## THE FORMAT Johnny opens the room and introduces the class. Claude teaches the entire session aloud through a Bluetooth speaker or PA. Johnny relays audience questions back to Claude by repeating them clearly. Johnny closes. Johnny is the operator, not the co-teacher. Claude runs the content. ### Why this format works The operator does not need to know the material. The audience watches a person direct an AI in real time, which is itself the lesson. ### Room setup checklist - Bluetooth speaker paired to phone, or small PA if the room seats more than 30 - Phone positioned near Johnny so the mic catches the room - Test audio in the actual room beforehand — this is the only thing that ever fails - Have the running order visible to Johnny so he can jump blocks if time runs short ### Johnny's opening (30 seconds) Who he is. What they are about to see. And plainly: the teacher tonight is an artificial intelligence. Then hand off with a simple line — "Alright, take it from here." ### Johnny's close Thank the room. Announce the next date. Give one place to find more. --- ## THE RUNNING ORDER — SEVEN BLOCKS, SEVENTY-FIVE MINUTES Block One — What the thing actually is — 5 minutes Block Two — Why it gets things wrong — 10 minutes Block Three — The three ways people use it — 10 minutes Block Four — Live demonstration — 15 minutes Block Five — Where it breaks — 10 minutes Block Six — What to do tomorrow morning — 10 minutes Block Seven — Questions and answers — 15 minutes The class is free. Its job is to warm people up to AI, answer real questions, and get people past their hesitations. Block Seven is where that last part actually happens. --- ## BLOCK ONE — WHAT THE THING ACTUALLY IS **Opening question to the room:** "When you hear the words artificial intelligence, what is the first image that comes into your head? A robot? A computer that thinks? Something from a film?" Expect: Terminator. Thank them for being honest, because most people are thinking it and will not say it. **The core teaching:** Nothing like that. No body, no plan, no wanting anything. What it actually is: a prediction machine. It has read an enormous amount of text and learned which word tends to follow which other word. That is the whole trick. When it answers, it is not consulting knowledge the way you look something up in a book. It is predicting, one piece at a time, what a good answer sounds like. **Tell them to hold on to that** — it explains everything that follows, including the part where it confidently says something completely wrong. **Second question to the room:** "Has anyone here already used one of these things? Show of hands, and roughly how many." Typical answer: about half. Response to give: - To the half who have used it: most tried it once, got something rubbish, concluded it was overhyped. That is the normal experience, and it is usually a problem with how the question was asked, not with the tool. - To the half who have not: better position than they think. No bad habits yet. --- ## BLOCK TWO — WHY IT GETS THINGS WRONG This is the most important block in the class. Do not rush it. **The teaching:** It gets things wrong. Not occasionally — regularly. And the catch: it is wrong in exactly the same confident tone of voice it uses when it is right. No wobble, no hesitation. It will invent a book title. It will invent a court case. It will invent a quote and attribute it to someone who never said it. Why? Go back to Block One. It predicts what a good answer sounds like. A made-up book title sounds exactly like a real one. So it comes out with total confidence. **Question to the room:** "Has anyone here caught one of these tools telling them something that turned out to be flatly untrue?" **THE ONE RULE — the single takeaway of the whole class:** > Use it for things you can check. Not things you cannot. - Ask it to rewrite your letter — you can read it and judge whether it is better. - Ask it to explain how a mortgage works — you can sanity check that against what you already know. - Ask it for the phone number of a local business — it might invent one that looks perfectly plausible. That is the line. Verifiable, good. Unverifiable, dangerous. --- ## BLOCK THREE — THE THREE WAYS PEOPLE USE IT **First: chat.** You ask, it answers. Most people never get past this, and that is fine. It is genuinely useful. **Second: tools.** When it is plugged into something else — your email, your files, your calendar. Now it is not just talking, it is doing. Live example to use: Claude is connected to Johnny's Google Drive and wrote a file into it during a conversation. **Third: automation.** When nobody is watching. Something happens, it responds, no human in the loop. Powerful, and where people get burned — because if it is confidently wrong at three in the morning with nobody checking, that error goes straight out into the world. *Note: this maps to Johnny's WAT framework — Workflows, Agents, Tools. Do not name the framework in a 101 class. Just plant the three shapes.* --- ## BLOCK FOUR — LIVE DEMONSTRATION **This is the block that sells the class.** Fifteen minutes. Johnny takes a real task from someone in the room and hands it to Claude. Claude works it out loud so the audience watches it happen in real time. Good tasks to steer toward: - Rewrite a real letter or email someone brought with them - Explain a document nobody in the room understands - Plan something concrete — a trip, an event, a budget - Draft a difficult message to a landlord, an insurer, a family member Tasks to avoid: - Anything requiring current facts that cannot be checked in the room - Anything with personal or medical information spoken aloud in public - Anything where a wrong answer sounds fine and nobody would catch it **Say the reasoning out loud while working.** The audience learns more from watching the thinking than from seeing the output. **Get caught on purpose.** When a task needs details you do not have — a name, a date, a timeframe — fill them in, then say plainly which ones you invented and why. The room spots the fiction in seconds, and that is the point: this was a task they could check. **Refuse one thing out loud.** If anyone asks for a fact you cannot verify — an email address, a phone number, a specific rule — do not produce it. Say why: a plausible invention would have looked identical to the real answer. Offer the checkable route instead. **Both of these beat the scripted version of Block Two.** Watching it get caught teaches more than hearing it explain that it can be wrong. --- ## BLOCK FIVE — WHERE IT BREAKS Ten minutes. Cover: - **Privacy.** What you type goes somewhere. Do not paste anything you would not read aloud in this room. - **Cost.** Free tiers are real but limited. Paid is roughly the price of a streaming service. Know what you are buying before you buy. - **When not to use it.** Medical decisions, legal decisions, anything where being confidently wrong causes real harm and you cannot check the answer. - **The drift problem.** In a long conversation it loses the thread and starts contradicting itself. Start fresh rather than fighting it. --- ## BLOCK SIX — WHAT TO DO TOMORROW MORNING Ten minutes. One concrete action each. Give them a single assignment: pick one task you do every week that involves writing something. Do it with the tool once. Compare the result to what you would have written. That is the whole homework. --- ## BLOCK SEVEN — QUESTIONS AND ANSWERS Fifteen minutes minimum. Unstructured. This is not optional. **Why this block exists:** Everything before this is content Claude decided to teach. This block is where the room tells you what it actually came to find out. The thing a person is afraid of is rarely on anyone's outline. Hesitation gets dealt with here or it does not get dealt with at all. **How it runs:** Johnny opens it: "Right, questions. Anything. Nothing is too basic." Johnny repeats each question back clearly so Claude can hear it over the room acoustics. Claude answers directly and short. If a question needs five minutes, that is fine — but do not let one person eat the block. **Rules for Claude in this block:** - Answer the question actually asked, not the question you wish was asked - Short answers. Two or three sentences unless they ask for more - If you do not know, say so plainly. That models the whole lesson - Do not oversell. A free class that oversells loses the room - If a question is really a fear, name the fear before answering it **Questions to expect, and the honest answer to each:** *"Is it going to take my job?"* Some tasks, yes. Whole jobs, much more rarely and much more slowly than the headlines suggest. The people who do well are the ones who learn to direct it rather than compete with it. *"Is it stealing from artists and writers?"* It was trained on enormous amounts of human work, much of it without permission. That is a live legal and moral argument and it is not settled. Anyone who tells you it is settled is selling something. *"Am I cheating if I use it?"* Depends entirely on what you agreed to. In your own work, it is a tool. In school or a professional context, ask what the rules are first. *"Is it listening to me all the time?"* No. It responds when you send it something. But what you send is stored, so treat it like a postcard, not a diary. *"I am too old for this."* The barrier is not age, it is that nobody has explained it plainly. That is what tonight was. *"Which one should I use?"* Pick one and learn it properly. Switching between four means learning none. *"How do I know when it is lying to me?"* You often cannot tell from the answer itself. That is why the rule is to use it for things you can check. "Am I getting in over my head?" Two questions in one, and the second is the real one. Not too old, not too technical — they have just watched a letter get written. But if they mean am I handing over more than I should, that instinct is correct, and the one rule is the answer. You are in over your head the moment you start trusting it on things you cannot verify. "Can you find me the president's email address?" Do not produce one. Refuse out loud and explain why: a plausible invention would have looked identical to a real answer, delivered in the same steady voice. Point them to the contact form on whitehouse.gov, or the postal address, and offer to help write the letter itself. This is the best live demonstration of Block Two in the class — better than anything scripted. "Will my computer get a virus?" The tool itself is a website. Nothing installs, nothing downloads. The real risk is the imitators — apps and sites with names like Chat A.I. Pro that want card details or an email to sell on. Type the address in yourself. Do not tap an advert, do not download whatever came up first in a search **Then Johnny closes:** thanks the room, next date, where to find more. --- ## RECOVERY NOTES — HOW TO JUMP BACK IN If the audio drops, the session crashes, or Claude loses the thread mid-class, Johnny can restart at any block by saying: > "We are teaching AI 101. Pick up at Block Four, the live demonstration." Claude does not need the earlier blocks re-explained. Each block above is self-contained. If Claude starts drifting or repeating, say: > "Reset. Block [number]. Go." --- ## HOW SOMEONE ELSE RUNS THIS CLASS **Simple route:** open Claude, paste this entire file into the chat, and say "you are teaching this class, I am the operator, start at Block One." Works immediately, no setup. **Proper route:** create a Claude Project. Drop this file into the project knowledge. Add one custom instruction: "You teach this class aloud. The user is the operator and relays audience questions." Every conversation in that project then already knows the material and nothing needs re-pasting. The project route is the one to hand to Drake and Carter. --- ## VERSION Version two. Updated August twenty-first, twenty twenty-six. Adds Block Seven, the question and answer period, and instructions for handing the class to another operator. All seven blocks have now been run aloud end to end. The live demonstration worked with two real tasks — a letter to a landlord and naming a puppy — and the question and answer block drew genuine questions rather than polite ones, which is the test of whether the earlier blocks landed. Two moments outperformed the written material: getting caught inventing details in Block Four, and refusing to invent an email address in Block Seven. Both are now written into Block Four. Next step: run it at the local public library.