Let’s start with you 01 / 05
Which AI tools have you actually tried?
Choose any that fit.
Hand raise or chat
Hand raises or chat both work. Students can choose several. Invite a quick example from someone who has used a coding tool. These buttons only highlight choices on this screen.
Let’s start with you 02 / 05
When a course concept doesn’t click, how do you use AI?
Choose any that fit.
Hand raise or chat
Ask for the most useful explanation someone has received. If nobody has tried it, ask which option they would start with.
Let’s start with you 03 / 05
Where does AI enter your homework process?
Choose any that fit.
Hand raise or chat
Keep this neutral. We are comparing ways of working, not asking anyone to disclose something they would rather keep private.
Let’s start with you 04 / 05
What have you asked AI to help you write?
Choose any that fit.
Hand raise or chat
Invite one example. Ask whether they used the first draft or changed it to sound like themselves.
Let’s start with you 05 / 05
What’s the most unexpected thing you’ve asked AI?
Or share your own answer.
Hand raise or chat
Give everyone a moment. Invite their own answer in chat; these options are just examples.
A useful starting point
What is AI?
Formal definition
A machine-based system works toward stated or implied goals by inferring how to turn inputs into outputs that can affect digital or physical environments.
Its outputs may be predictions, recommendations, decisions, or generated content.
Paraphrased from OECD ↗Already part of everyday business
Spam detection
Which messages belong in junk?
Product recommendations
What might this customer like?
Demand prediction
How much stock will we need?
Start with the broad category. AI has been used to predict, recommend, and make decisions long before today’s chat apps.
The definition is paraphrased from OECD. The examples show familiar business uses of this broader category.
The everyday meaning
What people usually mean
by “AI” today
Apps powered by generative models: systems that produce new text, images, audio, video, or code.
A model trained to work with language.
Understand
Explain a
case study.
Write
Draft an
email.
Create
Generate a
drawing.
Generative AI is one part of the broader AI category.
In casual conversation, “AI” often means an app we can ask to make something. Generative AI is a part of AI, not a replacement definition.
An app gives users access to models and tools. ChatGPT, Gemini, and Claude are examples of apps. Claude Code is an agent tool, not just a chatbot.
Models writing drawing instructions
One prompt. Six generations.
Generate an SVG of a pelican riding a bicycle.
The prompt asks the model to write SVG drawing instructions. This is not a separate image generator.
Look at the bird, bicycle, and their relationship. What improves? What is still odd? Click a drawing to inspect it.
These are selected examples, with high effort settings for the newest pair. The early GPT-3.5 Turbo and GPT-4o files come from Simon Willison’s 2024 comparison, not ChatGPT’s 2022 launch.
AI-generated depictions
Will Smith eating spaghetti
Watch what happens to the face, hands, and movement.
2023
ModelScope2025
Veo 3Play the 2023 clip first. Before playing 2025, ask the room to notice whether the face, hands, and spaghetti stay consistent as they move.
The 2023 example is an edited montage of ModelScope outputs. The 2025 example is attributed to Veo 3 by its creator, Javi Lopez. This compares selected examples, not identical testing conditions.
Creator demo · GPT-6 Astra
A 3D house from photos
From listing to model
Photos of a house.
A 3D reconstruction.
A promotional video.
Yunfan Ye shared this result from a Zillow listing.
The creator notes that some details are still wrong.
Play the original creator’s demo. The input is a listing with photos; the output shown is a 3D house and a promotional video.
Yunfan Ye describes a one-shot result and acknowledges wrong details. Treat this as a creator’s demonstration, not a guarantee that every listing will work the same way.
This is a coding and 3D tool workflow, distinct from generating video directly like the previous slide.
Creator demo · GPT-6 Astra
A playable game
Built to run in a browser
Code, 3D assets,
and a game
you can play.
Theo shared this browser game as a one-shot result.
A creator’s example, not a promise of the same result in one prompt.
The clip is a recording of a working browser game. This example combines code and 3D reasoning, rather than producing a video of a game.
Theo calls it a one-shot result. We do not know every condition behind the demo, so focus on what is visibly running.
Switch to the second example to see op7418’s 3D Roguelike level using GPT-6 Astra, Blender, and Godot.
From answering to acting · 1 / 3
Chat: ask for help, get a response.
Illustrative conversation
How should I find a venue for a student-club event with 40 people?
Start with these steps:
- Find venues near campus.
- Compare capacity and total cost.
- Email your shortlist about availability.
What happens next?
You carry out the plan.
You search the websites.
You work out the costs.
You open email and write.
The output is an answer.
The next actions are yours.
About 45 seconds. Start with a familiar experience: you ask a question and receive useful text. In this example, AI gives a plan, but you still move between websites, calculations, and email.
This is an illustrated text-only exchange. The same chat app may also offer tools and agent features; the distinction is what it does in this task, not the brand or the chat interface.
From answering to acting · 2 / 3
Agentic AI: work toward a goal.
The model chooses actions, uses tools, and checks results to decide what to do next.
Find three venues for 40 people within our budget.
Compare costs and prepare inquiry emails.
Browser
Search venue websites.
Look for location, capacity, and prices.
Calculator or script
Work out the total cost.
Include listed fees; flag missing prices.
Spreadsheet
Create a comparison.
Save the options, costs, and source links.
Email app
Prepare inquiry drafts.
Save messages for you to review and send.
Tool call = asking another program to do something.
Illustrative workflow · Requires connected tools and permissionAbout 75 seconds. An agent can take steps toward a goal using tools. These four boxes show a possible path, not a fixed recipe. If a venue has no prices, it may search another page, record that gap, or ask you for guidance. If all options exceed the budget, it should bring that back to you.
It needs your location, event date, budget, and access to the relevant tools. A tool call is the model requesting an action from another program. The software executes it and returns a result. The model uses that result to choose its next step.
Drafting and sending are different actions. This example stops at drafts because that is the requested goal. How agents use tools ↗.
From answering to acting · 3 / 3
An agent can build the website, too.
Build a website for our student-club event,
with the schedule, venue, and an RSVP link.
Editor
Create
the files.
Write the page, layout, and buttons.
Terminal
Run
commands.
Use a shell such as Bash to build and start the site.
Browser
Open
the website.
Inspect the layout and try the links.
Editor + browser
Fix issues.
Check again.
Revise the files and test the result.
Act → see what happened → adjust.
The agent can repeat steps, or ask you for help.
You can review a working website, then decide whether to publish it.
About 60 seconds. In a text-only exchange, AI might give you code to copy. With access to an editor, terminal, and browser, an agent can write files, run the site, and check its own work.
A terminal is a way to give the computer written commands. Bash is one kind of shell used to run them. Students do not need to understand these commands to follow the idea: the agent is using software, not only describing what to do.
The feedback loop matters. A broken link or failed command can cause another edit and another check. This does not guarantee a correct result. The site can be a local preview; publishing is a separate action. This example links to an existing RSVP destination, so it does not imply a working registration database.
Transition: “Let me show you what I have actually been using this for.” How agents use tools ↗.
Personal examples
What I’m using
AI for.
It’s more than chat.
Switch to your personal examples.