What AI Is, Ethics, and How To Prompt It
Meet AI. For Real.
What it is, what it isn't, and how to bend it to your will.
AI is a machine that learned to guess the next word — really well.
That single trick unlocks writing, coding, art, video, music, and reasoning.
Rules vs learned behavior
A human writes exact rules. 'IF button clicked → open menu.' It only does what it was told.
Nobody wrote rules. It read the internet and learned patterns. It writes NEW things it was never told to write.
Under the hood
- 01Predicts — 'given all this, what comes next?'
- 02Patterns — finds relationships humans miss
- 03Generates — combines learned patterns into brand-new stuff
- 04That's it. There's no ghost inside.
70 years in 60 seconds
Can machines think? First serious question.
Neural nets learn to see. Cats. Then everything.
Google invents the architecture behind every modern AI.
1 million users in 5 days. The world notices.
This is not a fad
Weekly ChatGPT users
Predicted AI market by 2030
Since GPT-4 shipped
The full AI family
- 01Text: ChatGPT, Claude, Gemini — reasoning + writing
- 02Image: Midjourney, Ideogram, Nano Banana — pictures
- 03Voice: ElevenLabs — clone or generate any voice
- 04Video: Sora, Veo, Runway — full clips from text
- 05Code: Cursor, Lovable — build apps by chatting
- 06Agents: AI that clicks buttons and does tasks for you
Sign up for these 4
- 01ChatGPT — default thinking partner
- 02Claude — the writer's AI, huge context
- 03Perplexity — research with real sources
- 04Gemini — image + doc understanding
The barrier between having an idea and shipping it just collapsed.
One kid + AI = what a 20-person team used to be.
AI is confidently wrong. A lot.
It doesn't 'know' things. It predicts the most likely-sounding next words.
When AI just makes stuff up
- 01Invents citations that don't exist
- 02Confidently states wrong dates, names, math
- 03Feels totally right — that's the danger
- 04Rule: verify anything with numbers, names, or laws
Every serious AI user must think about these
- 01Bias — trained on the internet = inherits its prejudices
- 02Misinformation — makes fake photos, voices, articles trivial
- 03Privacy — where does your prompt data go?
- 04Jobs — some work disappears; new work appears
- 05Environment — huge energy + water cost per query
Use vs abuse
Draft ideas, brainstorm, translate, learn, prototype, code. Always disclose when it matters. Fact-check output.
Submit AI work as your own on graded assignments. Fake someone's voice or face without consent. Spread synthetic 'news.' Pretend AI is a human.
With great tools comes great responsibility to not be a jerk with them.
Every founder's rule
Before you ship anything with AI
- 01Is anyone being fooled who shouldn't be?
- 02Did I check the AI's facts?
- 03Did I get consent for voices, faces, or private data?
- 04Would I be proud if my name was on this?
Prompting is talking to AI on purpose.
Vague in → vague out. Specific in → magic out.
R.A.C.E. — memorize this today
Tell the AI who to be. 'You are a Gen-Z copywriter…'
What exactly do you want? One clear job.
Paste the info it needs. Don't summarize — paste.
Show what good looks like. Bullets? Table? JSON? 8 lines?
Weak vs sharp — same idea
'Give me Instagram captions for my brownie business.'
'You are a Gen-Z social media manager. Write 5 Instagram captions for @midnightbrownies, a home-baked brand in Lahore. Playful, 1 emoji max, under 15 words, soft DM-to-order CTA.'
Stop doing these
- 01'Make it good' — AI has no idea what good means to you
- 02Asking two things at once — split into two prompts
- 03Not pasting the real material — AI can't read your mind
- 04Not naming the format — you'll get random paragraphs
The career-coach prompt
Act as my personal career coach. Ask me 5 sharp questions to understand what excites me. Then propose 3 wild but realistic AI-powered side projects I could launch this month. Format as a table.
Notice the Role, the Ask, and the Format.
The 'be specific' drill
You are a stand-up comedian who teaches physics. Explain black holes to a 14-year-old who loves basketball. Under 120 words. Include exactly one joke and one analogy. End with one question that makes me curious.
Feel how Role + Constraints changes the whole output.
What did you actually use AI for last week?
Poll voting opens when this slide is live.
Live activity: Prompt Battle
Same starting question — sharpest prompt wins.
- 01Split into pairs.
- 02Everyone starts with: 'Give me a startup idea that sounds dumb but is actually brilliant.'
- 03Each pair rewrites it using R.A.C.E. — role, ask, context, format.
- 04Run both versions. Vote on the best output. Winner reads their prompt.
Which R.A.C.E. letter do you find hardest?
Poll voting opens when this slide is live.
Meet the kids and companies bending AI to their will.
Three real people. Three real products. Three insane outcomes — all built in the last 24 months.
Chungin "Roy" Lee — got kicked out, raised $30M.
21-year-old dropped out of Columbia after building an AI that whispers answers into your ear during interviews. Went viral on X, raised from a16z within weeks. His moat isn't the tech — it's the audacity.
- →One post. One demo. Millions of views in 48 hours.
- →The tech: a Whisper transcript + GPT prompt + a stealth overlay.
- →Lesson: distribution + a spicy demo > 10 years of R&D.
- →You could rebuild v1 of Cluely in a weekend.
Four teens. A photo-of-your-plate app. $12M/year.
Zach Yadegari + friends, all 17-19. They wrapped GPT-4 vision around 'take a photo of what you're eating' and shipped it to the App Store. Sits in the US top 20 health apps.
- →Zero marketing budget. Grew on TikTok organically.
- →The prompt is literally 6 lines. That's the whole product.
- →One of the founders is still in high school.
- →Lesson: the fastest path is boringly obvious problems + AI + a nice UI.
The AI code editor eating VS Code alive.
Anysphere (the team behind Cursor) went from 4 MIT grads to a $10B valuation in under 2 years. Every top engineer you know is switching to it right now.
- →They didn't invent an AI model. They wrapped Claude beautifully.
- →$500M+ ARR. Team of ~50 people.
- →Proves: the interface layer captures the value, not the model.
- →Same tool you'll use tomorrow to build your first app.
Small team. Boring problem. AI wrapper. Ruthless taste.
This is the 2026 playbook, and you already have everything you need to run it.
Turn to your neighbor: what's ONE task you did this week that AI could have done in 60 seconds?
Name the exact task (not 'homework' — 'summarizing my history chapter').
What would you have used the saved time for?
What's stopping you from letting AI do it next time?
Which prompt is most likely to give a sharp, on-brand result?
You know what AI is, where it lies, and how to talk to it.
Tomorrow: pro moves — chain-of-thought, self-critique, personas, prompt chains.