Today is about one shift: going AI-native. AI isn't just for writing code. It helps you think through an idea, search for what's been done, organize what you find, and build the real thing — and you do it all in plain language. Increasingly, you just talk.
Everything you make this way leaves a trail — a version from the AI, a version from you, draft after draft. GitHub is where all of it lives. That's why we end today by setting one up.
Every step below is a prompt you can copy — drop it straight into Cursor, Claude Code, or Codex. Keep this page open all class. Stuck on setup? Mia, our code assistant, is here for exactly that.
01 · Icebreaker
Before any keyboards — a quick game to warm up the room. Everyone shares one number.
SayThe number that means the most to youAny number at all — a date, a jersey number, a score, an age. Just pick the one that's yours.
TellThe story behind itWhy that number? Give us the thirty-second version — that's how we get to know each other.
RuleNo wrong answers, no project talk yetThis one's purely for fun. We'll get to your idea soon enough.
02 · What good looks like
Here's last year's bar. A student named Jenny built "Designing Inclusive Fashion Tools" — a generative-AI tool that lets non-expert creators turn a character idea into a fashion design, by writing the hard image-prompts for them. Fill in a simple form (style, materials, colors…) → it generates a good prompt → Stable Diffusion makes the design.
Notice: it wasn't a game — it was a tool, for a real, overlooked group. That's completely fair. Your thing can be a game, a website, or a tool — as long as it helps real people and you study how they use it.
ArtifactA working tool real people usedBuilt with AI, online, usable — not a slideshow, not a demo on a laptop.
StudyA 25-person user studyShe surveyed real users, found what they actually needed, and tested her tool against it.
PaperA ~4–5k word paper, 15 refs, APALit review · method · results · ethics — defended in a deck at the Stanford × Peking University conference.
YouAim here, starting todayEverything in this class is the first step toward that.
03 · Talk, don't type
AI-native means you work in natural language — and the fastest natural language is your voice. You already have the best tool: WeChat. Turn on its voice input and dictate your prompts instead of typing them.
FirstLog into WeChat (update to 4.1.7+)You already have it. On Mac, make it work everywhere: WeChat → bottom-left menu → Settings › Shortcuts › Scope → set to "All Windows".
MacHold the Fn key and talkPress and hold Fn to dictate, release to finish. (Fn + Space = hands-free.) Real-time, any input box, Chinese + English + dialects.
WindowsPress Ctrl + WinCtrl + Win starts voice input in any text box — chat, Word, browser, your AI tool. (WeChat PC 4.1.7+.)
Stuck?Mia — our code assistantInstalls, accounts, "it won't open." Mia handles setup so we keep moving.
05 · Brainstorm with AI
First, how to actually talk to AI so it's useful:
Be specificSay exactly what you want, and the format"Give me 3 options, one sentence each" beats "help me think."
Give contextWho you are, what you're building, the limitsThe more it knows, the less generic it sounds.
Make it ask"Ask me questions before you answer"Turns a one-shot reply into a real conversation.
IterateYour first prompt is a draftRead the reply, tell it what's off, run it again. That's the loop.
Now don't brainstorm alone — make the AI argue with you. Your idea can be a small game or an interactive website; the one rule is that it should collect the most user feedback you can get, because that feedback becomes the evidence in your paper.
我是「AI for Social Sciences」课程的高中生。我想做一个关于我在意的某个社会议题的小项目:[在这里描述你的议题]。
请当我的头脑风暴搭子,并质疑我的想法。帮我决定:
1. 这该做成一个小游戏,还是一个交互式网站?
2. 这周我能搭出来并上线的最小版本是什么?
3. 它怎么才能从真实用户那里收集到最有用的反馈?
4. 这些反馈怎么变成论文里的证据?
回答前先反问我几个问题。挑我想法里薄弱的地方挑战我。
06 · Talk it out — small groups
We're nine. Split into three groups of three. Ideas get sharp when you say them out loud to people who aren't your AI.
PitchOne minute eachYour issue, whether it's a game or a site, and how it'll collect feedback.
RespondOne sharp question per ideaFor each person: what one question would make their idea stronger?
Leave withOne concrete change to your ideaWalk out of your group having improved it by at least one thing.
07 · Frame it as research
You're not just making something cool. This is social science: your project is an instrument for finding something out about people. Keep this arc in mind from day one —
QuestionA sharp research questionWhat do you actually want to find out about your issue?
InstrumentYour game or site is the toolNot the final product — it's how you gather evidence from real people.
DataUser feedback is your dataEvery choice, reaction, and answer you log is a data point.
PaperIt all becomes a paperLit review · method · a small user study (~25 people) · results — APA.
Paste this to turn your idea into a research-shaped plan — and to find what's already been studied:
Help me think about my project as social-science research, not just a build.
My issue / idea: [paste your idea here]
1. What's a sharp research question I could actually answer with this?
2. What topics, theories, or search terms should I read up on for this issue? Give me 3–5 real leads to look into.
3. What could I measure from how people use my game/site? (my data)
4. How do I build in feedback collection from the start, so I end up with real evidence for my paper?
Be concrete and push me toward something measurable.
Don't get fooled by AI. It sounds confident even when it's wrong, and it invents sources, names, and statistics. Treat anything it cites as a lead to verify yourself — open the real paper, check the real number. In research, the verifying is the work.
08 · Set up GitHub
Remember the trail? Everything you and the AI make — every draft, a version from each of you — needs a home where you can see every version. That home is GitHub. Here's what you'll set up:
You'll haveAn account that's yoursgithub.com — your home base for everything you ship.
You'll haveA repo for your projectYour project's folder in the cloud, public so people can see it.
You'll haveYour file, live on GitHubYour idea HTML, pushed — real, shareable, and yours.
Here's the AI-native move: you don't learn git. You hand the whole setup to your AI tool — it'll use the GitHub CLI (gh) and run it with you, step by step. Paste this. (Or just ask Mia.)
Your deliverable is a single index.html describing your preliminary idea, pushed to your repo. Why HTML, not Markdown? HTML is the most AI-native format there is — AI reads it best, every browser opens it, and it can grow straight into your real project. Paste this to draft it: