July 3, 2026

70 Years of AI History in Under 8 Minutes | EP 2

In 1956, ten researchers at Dartmouth College promised to build a thinking machine in two months. Seventy years later, ChatGPT hit 100 million users faster than any product in history, DeepSeek wiped half a trillion dollars off Nvidia in a single day, and autonomous AI agents started booking flights, filing taxes, and writing code. This episode is the complete story of how we got here.

Alex Smith, founder of Instant AI and host of Super Confident AI, walks through the full history of artificial intelligence from Alan Turing's 1950 paper to the transformer architecture that changed everything, OpenAI's scaling bet, the DeepSeek shock, and the agent era of 2026. He explains why the "AI is a fad" argument collapses under seven decades of math and trillions in infrastructure. If you use AI tools daily but never understood the story behind them, this is your episode.

Chapters:

(00:00) Introduction

(00:54) The Dartmouth Promise: Two Months

(02:11) AlexNet

(03:50) ChatGPT Goes Public

(05:52) 2025: AI Starts Reasoning

(06:36) The Agent Era Arrives

(07:04) Still Early Innings

(07:34) The Super Confident Take

Ten men in 1956 said two months. It took 70 years. Which moment in AI history surprised you the most? Comment below.

Sign up and get your free tokens: https://www.myinstantai.com

Follow me on Instagram: https://www.instagram.com/captainmakeithappn

00;00;00;02 - 00;00;02;07

AI is never going away.

00;00;02;13 - 00;00;04;12

how did it even fucking get here?

00;00;04;15 - 00;00;07;21

Here's the history. You're going to want to know this.

00;00;07;24 - 00;00;19;09

the summer of 1956. Ten men locked themselves in a building at Dartmouth College and promised they build a thinking machine in two months. It took 70 years.

00;00;19;11 - 00;00;22;03

What's crazy is they weren't actually wrong.

00;00;22;07 - 00;00;24;24

They just had no idea what they'd unleashed.

00;00;24;24 - 00;00;29;07

To understand 2026, we need to rewind to 1950.

00;00;29;07 - 00;00;36;29

A British mathematician named Alan Turing publishes a paper asking a question that frankly sounded insane for its time.

00;00;37;03 - 00;00;48;23

Can machines think? He proposes a test. If you chat with something behind a wall and can't tell if it's a person or a machine, for all practical purposes, it's thinking.

00;00;49;01 - 00;00;54;09

Six years later, the Dartmouth gang gives this thing a name artificial intelligence.

00;00;54;16 - 00;00;59;07

Two months. That was the project timeline. Yeah. About that.

00;00;59;08 - 00;01;11;28

What followed was 60 years of what we politely call progress, and what historians less politely call the AI winters plural. Funding dried up, researchers got laughed out of rooms.

00;01;11;28 - 00;01;24;07

Entire careers ended on the phrase, so what does it actually do? In 1958, a psychologist named Frank Rosenblatt built something called the perceptron, the original neural network.

00;01;24;12 - 00;01;37;11

A New York Times headline declared it the embryo of an electronic computer that would walk, talk, see right, and reproduce itself. Problem was, it could barely recognize a triangle.

00;01;37;12 - 00;02;11;18

Then for 40 years, basically nothing till we taught computers to play chess. Deep blue beat Kasparov in 1997, and everyone thought that was the AI moment. Spoiler alert it wasn't. Deep Blue. Couldn't do anything else. Ask it to write a haiku. It would just stare at you in binary. The first real holy shit moment 2012 A grad student named Alex enters a computer vision competition with something called AlexNet, a deep neural network running on gamer graphics cards.

00;02;11;24 - 00;02;27;16

He doesn't just win, he demolishes second place by an absolutely embarrassing margin. That's the moment the lights flicker back on the AI. Winter is over, and almost nobody outside of academia notices.

00;02;27;19 - 00;02;53;26

Fast forward to 2017. A team at Google publishes a paper with one of the most illuminating titles in academic history. Attention is all you need. Okay, let's walk in here for a moment. This is when AI, as you understand it, starts to take shape. Eight researchers casually drop a new neural network architecture called the transformer. It learns language by paying attention to which words matter in a sentence.

00;02;53;26 - 00;02;56;04

That's it. That's the trick.

00;02;56;04 - 00;03;00;09

That paper, in retrospect, was the heist of the century.

00;03;00;09 - 00;03;02;05

Because here's what happened next.

00;03;02;05 - 00;03;21;16

A scrappy startup called OpenAI, which at the time was a nonprofit, by the way. Remember, that takes the transformer idea and starts scaling it bigger, more data, more compute. They released GPT 2 in 2019 and refuse to publish the full model because they claim it's too dangerous.

00;03;21;19 - 00;03;30;01

Everyone in the industry rolled their eyes. Then in 2020, they released GPT three and the eye rolling stops.

00;03;30;07 - 00;03;41;11

For two years, GPT three lives in developer land. Programmers play with it. Nobody else cares. Your mom has never heard of it. Your senator certainly hasn't.

00;03;41;11 - 00;03;50;09

Then on November 30th, 2022, OpenAI does something they call, quote, a low key research preview, a little chat bot.

00;03;50;11 - 00;03;51;28

They threw it online for free.

00;03;52;03 - 00;03;54;15

They thought maybe a few thousand people would try it.

00;03;54;15 - 00;04;12;14

ChatGPT hit 100 million users in two months, the fastest growing consumer product in human history. Faster than TikTok. Faster than Instagram. Faster than literally anything humans have ever built before. That's the day the 70 year heist went public.

00;04;12;14 - 00;04;15;24

What happened next was a complete loss of corporate composure.

00;04;15;29 - 00;04;34;18

Google declared a code red an internal panic mode that until that point had been reserved for like server outages. Microsoft poured $10 billion into OpenAI. Meta open sourced its own model called llama, and basically gave away the recipe to anyone who wanted it.

00;04;34;20 - 00;04;53;03

A new company called anthropic, founded by OpenAI researchers who were spooked about safety, released a little model called Claude. Google eventually fired back with their model Gemini. Within 12 months, you couldn't throw a rock in San Francisco without hitting $1 billion AI startup.

00;04;53;07 - 00;04;56;05

Okay, now it really starts getting out of hand.

00;04;56;05 - 00;05;10;08

In January 2025, a Chinese company nobody had heard of called Deep Seek released a model called R1. It matched the best American models it was trained for, allegedly a fraction of the cost.

00;05;10;13 - 00;05;13;06

No one knows how they built it, but it's good.

00;05;13;06 - 00;05;14;10

Really good.

00;05;14;10 - 00;05;21;28

Silicon Valley went so far as to accuse China of corporate espionage, which is obviously never happened before.

00;05;21;29 - 00;05;34;15

within 48 hours, Nvidia, the company that makes the chips powering this whole revolution, lost half $1 trillion in market cap in one day, half a trillion with a T in one day.

00;05;34;20 - 00;05;39;29

Because a company you'd never heard of in a country we'd been told was behind

00;05;40;01 - 00;05;43;08

shipped a model that was just as good.

00;05;43;08 - 00;05;49;10

Just like that, the American AI monopoly was now existential threatened an A Monday.

00;05;49;17 - 00;05;52;21

Which brings us finally to right now.

00;05;52;26 - 00;06;12;06

2025 was the year AI stopped predicting and started reasoning models, stop guessing the next word and started actually thinking, as in pausing, planning, checking their own work, remembering rules, utilizing tools, and creating its own code. Literally everything

00;06;12;11 - 00;06;19;06

GPT five dropped in August. Gemini three. In November, Claude went through three major revisions.

00;06;19;06 - 00;06;36;17

And then in January of 2026, something happened. Agents AI that doesn't just answer you, it goes out and just does things. Books your travel, reads, your inbox, files your taxes, argues with your insurance company,

00;06;36;19 - 00;06;40;06

runs tasks passively, codes your website,

00;06;40;06 - 00;06;41;08

everything.

00;06;41;08 - 00;06;51;29

ChatGPT crossed 800 million monthly users. Your parents now use it. Your kids teachers now uses it. Your doctor uses it, even if she won't admit it.

00;06;51;29 - 00;07;04;27

Anthropic then adds on tens of billions of dollars in revenue in a matter of months, becoming the fastest growing company in human history. And here's a wild part we're still in the early innings.

00;07;04;27 - 00;07;08;16

Two months. That was the original Dartmouth timeline.

00;07;08;16 - 00;07;19;19

70 years to overshoot it. But once it tipped, the curve didn't bend it vertically. An actual hockey stick moment played out in a matter of a couple of years.

00;07;19;22 - 00;07;34;04

So when somebody tells you that AI is a fad, ask them which part. The seven decades of math, the trillions of dollars of infrastructure, the 800 million people using it before their cup of coffee?

00;07;34;10 - 00;07;41;20

We didn't stumble into this. Ten guys in 1956 told us this was coming. We just didn't believe them.

00;07;41;25 - 00;07;56;03

I'm Alex Smith. This is super confident AI, the channel that pays you in tokens to actually understand what's happening in your world. Hit subscribe because the next 70 years, they're going to be a lot faster than the last 70.