Prep
This is the first night, so half the prep is social and half is logistics. Everyone creates a chatbot account today, which means the room will be full of people who have never made an account on anything. Plan for that instead of being surprised by it.
Materials
- Projector, and a browser window you are willing to show.
- Sticky notes, three per person, plus whiteboard space to cluster them.
- Your own laptop with a small model already installed and working offline.
- A printed card of six Hugging Face Spaces you checked this morning.
- A webcam for Teachable Machine, on your machine, not theirs.
Teacher note
Accounts to pre-stage
- Nothing is pre-staged tonight, because account creation is the activity. Fifteen minutes are budgeted for it and they will get used.
- Name a tech buddy or two early: a confident student who handles individual troubleshooting while you keep the room moving. It is a real job and it is a good one to hand out on night one.
- Some students will have a phone and no email password. Have a plan that does not involve them sitting out: pair them with a neighbour for the block and get the account sorted during the wander.
Fifteen minutes before
- Open Quick, Draw! and play one round to clear the first-load prompts.
- Warm every Space on your card. Cold starts take minutes and they will happen at the worst moment. Re-curate the card if any of them are down.
- Turn the wifi off on your laptop and ask the local model a question. If that does not work, the reveal does not happen.
- Test the webcam in Teachable Machine, with the projector lighting you will actually have.
- Write the three sticky-note headings on the board: hope, worry, heard.
Wifi fallback
The local model needs no internet and becomes the whole show. Run the Intelligent Piece of Paper on the board (tic-tac-toe against a sheet of if-then rules that wins without thinking), then do the tool wall as a narrated tour of your own machine. The session survives intact, it just loses the scavenger hunt.
Depends on earlier modules
- Nothing. This is the front door.
- It sets up Module 2 (the first prompts sent tonight are the ones you iterate on next time), Module 5 (predict-the-next-word is the whole basis of the hallucination lab), and Module 6 (the sabotaged classifier is the seed of every bias conversation).
What they should walk out with
Big idea
A model predicts the next word. Fluent and true are different things, and telling them apart is a skill you can learn tonight.
Maps to job skills: AI literacy and vocabulary · evaluating and choosing AI tools · recognizing bias in trained systems · verifying AI output before trusting it.
Teacher note
The real target for tonight is smaller than the objectives above: by the end, "AI" has stopped being a cloud and become a thing they poked. Everyone has an account, everyone has sent a prompt, and everyone has watched a model fail. That is enough for one evening.
Say the agency line once and move on: the first question about any AI tool is whether to use it, not just how.
Session timing
The three hours0:00 – 3:10
| Time | Share | Block |
|---|---|---|
| 0:00 – 0:15 | Welcome and course tour, plus the hopes, worries and heard sticky notes. | |
| 0:15 – 1:20 | Lecture: what AI is, five stops of history, and predict-the-next-word live. | |
| 1:20 – 1:30 | Break. | |
| 1:30 – 3:00 | Lab: Meet the machine. Broken out below. | |
| 3:00 – 3:10 | Wrap-up, back to the sticky-note board, and the homework handoff. |
Teacher note
The predict-the-next-word demo belongs in the lecture block and it is the most load-bearing five minutes in the course. Type a message on the projector, stop mid-sentence, let the room shout the next word, then finish with "the mitochondria is the ______" and let them say it in unison. Module 5 stands on this.
The lab block, minute by minute90 min, from 1:30
| Lab clock | Share | Activity |
|---|---|---|
| 0:00 – 0:10 | First win: Quick, Draw!, two rounds, no account and no typing. | |
| 0:10 – 0:25 | Get a key to the building: one free chatbot account each, then playful first prompts. | |
| 0:25 – 0:50 | The tool wall (big rock): three Spaces together, a scavenger hunt, then the offline reveal. | |
| 0:50 – 1:00 | Sabotage the classifier on the projector. | |
| 1:00 – 1:20 | Wander block. It absorbs the account chaos, which is why it is this long. | |
| 1:20 – 1:30 | Spotlight sweep: three or four students show the coolest thing they found. |
The lab, activity by activity
Warm-uplab 0:00 – 0:10
Quick, Draw!
Two rounds of Quick, Draw! on the projector, then everyone plays a round themselves. No account, no typing, no way to be bad at it. The room laughs inside ninety seconds, which is exactly what night one needs.
"That thing just recognised your terrible drawing of a kangaroo. Nobody programmed it with kangaroo rules. It was shown a lot of drawings. Hold on to that, it is most of tonight."
1. Get a key to the building
lab 0:10 – 0:25 · solo, with buddies circulatingActivity15 min
One account each, then a silly first prompt
Everyone creates one free chatbot account, their choice of the big three, and sends a low-stakes first prompt: a fusion recipe, a fantastical vacation plan, a poem about the class. Nobody's first prompt should matter. The account setup is the activity, not a delay before it.
Launch script
"Fifteen minutes, one account, and I do not care which one you pick. If you get stuck, put a hand up and one of us comes to you. Nobody sits here quietly stuck, that is the one rule."
"Now ask it for something ridiculous. A recipe that combines two cuisines that should never meet. This is the lowest-stakes thing you will ever type into one of these."
Use the prompt list on the board, word for word. An account made and a prompt sent is a full pass on this block.
Run the same prompt through two assistants and iterate twice on each with "make it better". Write down what actually changed between rounds, and which tool refused what.
Anticipated wrong turns
- Phone number verification. Some sign-ups want one and some students will not want to give it. Know in advance which of the three is least demanding tonight and say so out loud.
- An email account nobody can get into. Common, and embarrassing for the student. Handle it privately, pair them up, fix it in the wander.
- The confident student finishes in ninety seconds and starts poking at things loudly. Give them the tech-buddy job immediately.
- Somebody asks it something personal on the first try. Good moment to name the information-governance idea early: what you paste in does not stay in the room. Module 5 goes further.
Discussion, with the answers you are steering toward
"How did it feel to type into that box?"
Some people say "silly", some say "watched". Both are worth hearing out loud on night one, because both are why people avoid these tools.
"Did anyone get a refusal? What did it refuse?"
Different products draw different lines. That is a product decision made by a company, not a law of nature, and it is the first crack in the idea that "AI" is one thing.
2. The tool wall
lab 0:25 – 0:50 · together, then a hunt in pairsActivity · big rock25 min
Walk the tool aisle, then see one on a laptop with the wifi off
Frame Hugging Face as the tool aisle at the hardware store. Walk three Spaces together, each doing one obvious job: speech to text, background remover, image describer. Glance at the label on the box each time (the model card, read in plain terms: what it does, what it is good at, what to watch for). Then a short scavenger hunt from your curated card. Finish with the reveal: the same kind of tool running on your laptop with the wifi switched off.
Launch script
"This is a hammer. This is a screwdriver. Thousands of free tools, each made for one job, most with a try-it button. You do not need to build any of them, you need to recognise one and read its label."
"Watch this." (turn the wifi off) "Same job, no internet, no company. This one lives in my toolbox. That is the difference between being a customer and being an owner, and it is most of why this course exists."
Press the try-it button on one demo, watch it work, and tell your neighbour what job it is for. That is the whole objective.
The deep end, with a guide standing in it: install LM Studio or Ollama with help, read a model card's fine print (sizes, quantization, what fits on your hardware), and watch a network learn in TensorFlow Playground. This stays open for the rest of the course, not just tonight.
Anticipated wrong turns
- A Space is cold and takes four minutes to wake. This is why you warmed them. Have a second tab already running the same job and switch to it rather than watching a spinner with thirty people.
- The hunt turns into browsing. Give a concrete job ("translate this menu photo", "clean up this voice memo") rather than "find something interesting".
- Model cards read as impenetrable. They are, at first. Read one out loud in plain words and move on. The point is that a label exists, not that they parse it tonight.
- The offline reveal lands flat if the room did not see the wifi go off. Make it physical: show the toggle, show the disconnected icon, then run it.
Discussion, with the answers you are steering toward
"What is the difference between the thing you signed up for and these?"
The chatbot is a product wrapped around tools like these. Useful framing, and it makes the whole field look smaller and more approachable.
"Why would anyone run one of these on their own machine?"
Cost, privacy, no account, it still works when the company changes its mind. Say it once, do not sell it.
"What job would you want a tool for?"
Take real answers and write them on the board. Several of them will come back as capstone projects in Module 8.
3. Sabotage the classifier
lab 0:50 – 1:00 · projector demo, the room supplies the dataActivity · fun peak10 min
Train it, then ruin it, on purpose
Train a two-class thumbs-up and thumbs-down model in Teachable Machine on the projector, with the class shouting examples. Then a volunteer poisons the training data with skewed, sloppy examples and it fails confidently. Debrief in one question: what did it actually learn?
Launch script
"Give me thumbs up. More. Now everybody thumbs down. Right, it works. Somebody come up here and break it for me."
"It is not confused. It is confident and wrong, which is much worse. It learned exactly what we fed it."
Watch, shout suggestions, and answer the debrief question. No device needed.
In the wander block or at home, pairs build their own multi-class model and stress-test its edges. What is the smallest change that flips the answer?
Anticipated wrong turns
- Lab machines have no webcam. This is exactly why it is a projector demo. Do not promise students they will run it themselves tonight.
- The sabotage is too subtle to see. Ask the volunteer for something blatant. Train "thumbs down" on an empty chair. Blatant is legible from the back row.
- It stays accurate anyway. Happens. Add more bad examples, or narrate it honestly: it took real effort to break, which is its own useful finding.
- The room concludes AI is useless. Not the lesson. It learned what it was shown, which is a statement about the data, not about the machine.
Discussion, with the answers you are steering toward
"What did it actually learn?"
Whatever was in the pictures, including the background, the lighting, and the sleeve of whoever was volunteering. Not the concept you had in mind.
"Where would this matter outside a classroom?"
Anywhere a trained system sorts people. Do not chase it tonight, just plant it. Module 6 spends three hours here.
Spotlight sweep and the wander block
Spotlight sweeplab 1:20 – 1:30
Three or four students, thirty to sixty seconds each. On night one this sets the tone for the whole course, so make it easy and make it warm. Volunteers first, then a quiet "that is great, will you show it?" tap. Never a cold call.
Prompts to hand the student showing
- "What did you find, and what job is it for?"
- "Show us the weirdest thing it did."
- "What did you type to get that?"
Prompts for the room
- "Who else tried something and got a completely different answer?"
- "Hands up if you made an account tonight." (Everyone claps. It matters.)
Wander block menulab 1:00 – 1:20
Twenty unstructured minutes. On night one a chunk of it goes to finishing account setup, and that is fine. Offer the menu, do not assign from it.
- Chase whatever Space caught your eye. The most common choice and the best one.
- Install LM Studio or Ollama with help, for anyone who leaned into the offline reveal.
- Games stations: Semantle, ELIZA, Real or Fake Text. All account-free browser play.
- Ask the parking-lot question. Start a visible parking lot on the board tonight for "how does it actually work" questions, and commit to a time you will answer them.
- Finish the account, quietly, with a buddy.
Synthesis, exit ticket, homework
Land it here
It predicts, it does not understand. Everything you build on top of it has to account for that.
Synthesis, five minutes
Go back to the sticky-note board from the start of the session. Read two or three of the "heard" notes out loud and fill in what the room now knows. Leave the rest up: the whole course keeps answering that board, and returning to it in Module 6 is one of the better moments in the term.
Exit ticket
One-minute paper, anonymous: what surprised you tonight? On the first session this doubles as a read on who is nervous and who is bored, and it tells you who to pair with whom next time.
Homework handoff
The comparison notes go home, plus Chapter 1 of Elements of AI (free, from the University of Helsinki and MinnaLearn, and we are not affiliated with them). Say what next module does in one sentence: we stop admiring the tools and start getting good at asking.