1 / 16
← → navigate  ·  N notes  ·  F fullscreen

Career Discovery Week — AI Workshop

The AI Story

From Dreams to the Agentic Era

What AI actually is, where it came from,
and why it matters to YOU

Dr. Nick Gu  ·  Google  ·  Columbia University  ·  Pioneering Minds AI

What do you think AI is?

Raise your hand if you've used
ChatGPT, Gemini, or Claude.

What did you use it for?

You're already living in the AI era.
Let me show you what's really happening.

The dream started 80+ years ago

A fast-forward through AI history

1943
First mathematical model of a neuron — “Can we build a brain?”
1950
Alan Turing asks “Can machines think?” — proposes the Turing Test
1956
Dartmouth Workshop coins “Artificial Intelligence”
2012
Deep learning explodes — neural nets crush image recognition
2017
Google's “Attention Is All You Need” — the Transformer
2022
ChatGPT launches — 100M users in 2 months.

They asked the same question
65 years ago.

CBS · MIT · 1961

“The Thinking Machine” — featuring Claude Shannon, the father of information theory.
The questions haven't changed. The answers have.

What changed: three ingredients

All three had to converge — and they did, around 2012–2017

🌍

Data

The internet gave AI billions of examples to learn from

⚡

Compute

Gaming GPUs turned out to be perfect for AI math

🧠

Algorithms

Transformers (2017) — the architecture behind every modern AI

All three hit critical mass between 2012 and 2017.
Then everything changed.

2022–2026: The Explosion

100M ChatGPT users in 2 months — fastest-adopted product in history.
Instagram: 2.5 years · Netflix: 10 years
320ms GPT-4o voice response time — same speed as human conversation
1M tokens in one context window — ~10 novels at once (Gemini)
96.7% on ARC-AGI reasoning benchmark — from near-zero two years ago (o3)
In January 2024, AI could write text and look at images. By March 2026, it generates cinema-quality video with sound, composes music, controls computers, and reasons through PhD-level math.

Beyond Chatbots:
The Agentic Era

Then: AI That Talks

  • You type a question, get an answer
  • Can't see its own output
  • Can't fix its own mistakes
  • Waits for your next instruction
  • Like a genius chained to a desk

Now: AI That Acts

  • Writes code, runs it, debugs it
  • Opens browsers, fills out forms
  • Controls your computer (mouse + keyboard)
  • Works for 25 hours straight autonomously
  • Like an employee you can manage
We went from AI that talks to AI that works.

The Master Pattern:
Feedback Loops

This one idea explains why some AI workflows succeed and others fail

👁

Perceive

Can AI see what it did?

⚖

Evaluate

Did it work?

🔧

Correct

What went wrong?

🔄

Execute

Try again.

💡
Chatbot = loop is broken. It can't see results or fix mistakes.
Agent (Claude Code, Cursor) = complete loop. Write → run → see error → fix → repeat.
Watch

AI That Sees & Acts

Pick one to show the class — all under 5 minutes

👁️ Project Astra

AI sees through your camera in real-time. Identifies objects, remembers where you left things.

Jaw-drop: “Where did I leave my glasses?”

▶ Watch on YouTube (~3 min)

🖥️ Claude Computer Use

AI controls a computer — moves mouse, clicks, types, reads errors, fixes them.

Jaw-drop: Hits a bug → reads terminal → fixes it.

▶ Watch on YouTube (~4 min)

🎬 Veo 3 — AI Video + Sound

Type a scene and AI generates video with synchronized dialogue.

Jaw-drop: A cowboy speaks your exact words.

▶ Watch on YouTube (~3 min)
Live Demo

Try It Live: AI That Creates

All free with a Google account — open these now

🎵 MusicFX

Type a description → get a music track in 10 seconds.

“lo-fi hip hop with vinyl crackle and soft piano”

▶ labs.google/musicfx

🧶 Whisk

Drag in images → AI remixes them into new creations. Combine styles, subjects, scenes.

Drop a photo of yourself + a style image → instant mashup

▶ labs.google/whisk

🎙️ NotebookLM

Upload any document → AI generates a podcast about it.

Upload a chapter, click “Audio Overview”

▶ notebooklm.google.com

🎬 Flow (Video)

Text → cinematic video clip with synchronized audio, dialogue, and music.

“Medium shot of a student standing alone in a vast library after hours. She reaches for an old book on the top shelf — as her fingers touch the spine, the pages burst open and golden light floods out, illuminating her face. The camera slowly pushes in as floating letters swirl around her. She whispers in quiet awe, ‘It was here the whole time.’ Cinematic, warm golden lighting. Audio: Soft piano, pages rustling, gentle orchestral swell.”

▶ labs.google/flow
Watch

AI That Codes

These agents write, run, debug, and fix — all by themselves

🤖 Claude Code

Terminal-based agent. Builds entire codebases — writing, running, reading errors, fixing bugs in a loop.

▶ 30-min Masterclass

🎮 OpenAI Codex

Ran for 25 hours straight. Built a 3D racing game from one prompt. 30,000 lines. Then played its own game to find bugs.

▶ Building a Game with Codex

💻 Gemini CLI

Google's terminal agent. Reads your codebase, writes files, runs commands, fixes bugs — all from the command line.

▶ Gemini CLI on GitHub
You won't compete with AI at writing code. You'll compete at directing it.

AI as Your First Teammate

Not a magic button — a teammate you need to work with

📋

Clear Instructions

Vague asks → vague work.

📁

Give Context

It doesn't know what you know.

🔍

Check the Work

It WILL make mistakes.

🔄

Give Feedback

“Not what I wanted. Change X.”

70–80% of AI failures trace back to the human not managing the AI properly.

Magic button → disappointment.   Your teammate → incredible results.

Your AI Toolkit: Free

Open these on your phone or laptop

Homework: Pick 3 tools. Use each for 15 minutes. Come back ready to share.

Seven things you can do today

  1. Define “done” before you ask AI.
    Not “help me with my essay” → “write a 500-word persuasive essay arguing X with 3 reasons”
  2. Use AI that does things, not just talks.
    NotebookLM generates podcasts. AI Studio builds apps. Move beyond chatting.
  3. Break big problems into small pieces.
    Not “help me study for bio” → “explain mitosis in 3 sentences, then 5 practice questions”
  4. Check everything.
    AI confidently states wrong things. Always verify facts.
  5. Learn the subject FIRST, then use AI to go faster.
    5x faster at things you understand. Dangerously confident about things you don't.
  6. Write things down.
    More context → better output. Notes, examples, background.
  7. Build throwaway tools freely.
    Use AI Studio to build a quiz app. Use NotebookLM to turn notes into a podcast.

Your competitive advantage

✗

Ignore AI

Fall behind as tools become standard

⚠

Blindly Rely

Produce mediocre, generic work

💡

Direct AI

Amplify your thinking. Keep your judgment.

You are the first generation that will manage AI employees before you manage human ones.

Questions & Discussion

What surprised you most today?

What's one thing about AI that worries you?

What will you try first?

Don't just be consumers of AI.
Be the ones who decide what it should do.