In 1978, a 17-year-old Yann LeCun was told by his math teacher that he wasn't cut out for France's most elite schools.

Instead, he enrolled in engineering school ESIEE PARIS to study electrical engineering. A perfectly respectable career.

Then he picked up a book.

In 2016, Niki Parmar had no PhD, no formal research title, and no one asking her to co-author what would become the most cited AI paper of the 21st century.

At Google, a small research team was trying to figure out how to make AI understand language faster. They had a radical idea: what if AI could read an entire sentence at once, instead of one word at a time?

Niki wasn't supposed to be central to that effort. The youngest on her team, she was a software engineer from Pune, India who had first learned AI through free online courses in college.

In 2004, Sal Khan's 12-year-old cousin Nadia was placed into her school's slower math track. She told him, "I'm just not a math person."

Khan, a hedge fund analyst with 3 degrees from Massachusetts Institute of Technology, knew that one placement in middle school follows you for life.

He made her a deal: if she'd give him half an hour a day after school, he'd tutor her over the phone and Yahoo Doodle.

In 2018, Mark Cuban opened a cold email he would normally delete. The subject line read "Cold pitch."

A 33-year-old radiologist named Alex Oshmyansky, MD, PhD was asking him to help sell cheap drugs.

Despite getting hundreds of pitches a day, Cuban replied to this one in five minutes. "I started reading it, and I'm like, 'This guy gets it.'"

In 2011, Jelani Nelson taped a math puzzle to bulletin boards across Addis Ababa, capital of Ethiopia:

"700 people enter a room and each shakes every other person's hand exactly once. How many handshakes? If the answer is X, email algorithmsX@mit.edu."

Nelson was 27, visiting family between finishing his PhD at Massachusetts Institute of Technology and starting a postdoc in Berkeley. With six weeks to fill, he decided to teach a free coding course to local high schoolers. He called it AddisCoder.

In 2022, Aravind Srinivas sat outside Yann LeCun's office for five hours with no appointment.

LeCun was a Turing Award winner, one of the "Godfathers of AI," and a professor at NYU.

Srinivas was a 28-year-old with a prototype for an AI search engine held together with conviction.

In 2012, Andrew Ng launched Coursera to bring world-class education to millions. The first power user is his 66-year-old father.

Dr. Ronald Ng was a hematologist in Singapore. When his son's platform went live, he signed up for a logic course from the University of Michigan.

He got hooked.

In 2002, Gwynne Shotwell walked into the SpaceX office and told Elon Musk he had no idea how to sell rockets.

She was at SpaceX to celebrate a friend's new job over lunch and ran into Musk by accident. The next day, Musk's assistant called. He wanted to hire her.

She was going through a divorce and raising two young children. Joining a rocket startup with zero launches felt reckless. She spent weeks agonizing.

In 2020, a 20-year-old Dwarkesh Patel sat in his UT Austin dorm room and cold-emailed economist Bryan Caplan about a podcast that didn't exist yet.

He'd just finished Caplan's book, The Case Against Education, and wrote him a fan email. Caplan replied. Dwarkesh asked to interview him.

"He assumed I had an actual podcast," Dwarkesh later said. "I didn't even have a name for one."

In 1994, Terence Tao spent two weeks preparing for a qualifying exam his Princeton classmates had studied months for. He walked in with confidence. He walked out shaken.

Tao was 18 and already a math PhD student. He had won a Math Olympiad gold medal before turning 13 and finished a master's degree at 17. He had never needed to study.

So he coasted at Princeton University. He played computer games late into the night. While his classmates formed study groups and drilled each other with mock exams, Tao skimmed his notes and figured raw ability would carry him through. It always had.

In 2006, Sundar Pichai walked into Eric Schmidt’s office and told Google’s then CEO they needed to build a browser. Schmidt looked at him cold. "Do you know what it takes to build a browser?"

Schmidt had spent six years shutting down the idea. He had lived through the brutal browser wars of the ‘90s and watched Netscape get destroyed by Microsoft. Most leaders inside Google agreed.

Pichai saw the opposite reality. As Google Toolbar’s product director, he watched Microsoft build Internet Explorer 7 that could potentially block Google search from hundreds of millions of computers overnight. An executive later called it a "doomsday-like scenario."

In 2001, a 26 year old engineer named John Ternus stood in a supplier's factory late at night, holding a magnifying glass over a tiny screw.

It was his first project at Apple: a desktop monitor called the Cinema Display.

The back panel screws were supposed to have 25 grooves on their heads. The supplier had cut 35. Ternus wouldn't let it go.

In 2011, John Jumper started a chemistry PhD at the University of Chicago. He had never taken a college chemistry class in his life.

He wanted to pursue a PhD in physics, but he had missed Chicago's application deadline by hours. The department wouldn't let him apply late.

When someone suggested a PhD in chemistry, he said: "sure, how hard can this be?" He was accepted.

In 1990, Jeff Dean studied neural networks and said: "give me 32 more computers and I can make this work." He was directionally correct. He just needed 15,968 more.

Dean was a 21-year-old senior at the University of Minnesota, training neural networks on the CS department's beefiest machine. He believed in its potential. The hardware just hadn't caught up yet.

So he filed it away. In 1999, he joined Google as its 30th employee and spent the next decade scaling search with Sanjay Ghemawat, writing the systems that would become the backbone of the modern internet.

In 2008, Satya Nadella made a phone call that would cost him his own job title.

At that time, Nadella was a 16-year Microsoft veteran and an SVP running R&D for the search division.

The person on the other end of the call was Qi Lu, a computer scientist who had spent a decade at Yahoo managing 3,000 engineers.

In 2021, Anima Anandkumar told her team at NVIDIA they were going to teach AI to outperform supercomputers at simulating the physical world. "Many scientists discouraged us from pursuing this," she recalled.

She chose the hardest test case: weather.

For 50 years, global weather forecasting belonged to governments running hundreds-of-millions-dollar machines that filled entire buildings to simulate the atmosphere across millions of grid points.

In 2000, Jimmy Wales's newborn daughter Kira inhaled contaminated fluid and couldn't breathe. Doctors offered an experimental procedure. Wales searched the internet for reliable medical information and found nothing. "It was like sifting through the debris of a bombed-out library."

Doctors routed Kira’s blood through a machine, and flushed her lungs four times. Wales and his wife said yes in the dark, trusting a doctor they had never met. Kira survived. "I realized how precious the doctor's knowledge was," he later said. "There had to be a better way."

Before Kira, Wales spent two years building Nupedia, a free online encyclopedia where experts reviewed every draft. In two years, it produced 21 articles.

In 2011, Stéphane Bancel told his wife the odds of using Moderna's mRNA technology to treat diseases like cancer were barely 5%.

Brenda, a professional photographer, replied, "You have to do it. You are so relentless and bold and stubborn, you will find a path."

Bancel resigned from a 6,000-employee company and became employee number two of Moderna.

In 2015, Leonardo de Moura submitted a paper about his proof assistant, Lean. It was rejected. The reviewer asked, "Why do we need another proof assistant?"

At Microsoft, Leo and Nikolaj Bjorner had spent years developing Z3, a tool designed to catch software bugs. But when engineers tried using it to prove their code was correct, it produced unstable results. One engineer vented: "Come on, man, I just want to convince the system that my proof is correct."

So Leo started building something better. He called it Lean because Z3 had grown bloated, and he wanted the code to be clean.

In July 2023, Ben Gilbert hit delete on an hour of finished podcast recording. His cohost David Rosenthal agreed: "Losers make excuses. Winners make adjustments."

Gilbert and Rosenthal are the creators of Acquired, a podcast telling the stories behind the world's greatest companies. They started it in 2015 as two friends in Seattle looking for an excuse to hang out. The early episodes were 30-minute case studies analyzing tech acquisitions. Almost nobody listened.

In 1986, Aswath Damodaran walked into a classroom at NYU Stern, closed the door, and started teaching a finance class that didn't exist. "They have no idea what I do in the classroom," he later said. "I could be teaching cooking for all they know."

NYU had assigned him Security Analysis, Ben Graham's legendary course. Damodaran took one look: "The most boring collection of topics I can think of." He asked to teach Valuation instead. His department head told him there wasn't enough material to fill a semester.

Damodaran was 28 and untenured. He started teaching Valuation anyway.

In 1995, Katalin Karikó was given a choice by her university: stop researching mRNA, or accept a demotion and pay cut.

Most scientists had already abandoned mRNA research. It triggered dangerous immune reactions. Grant committees called it impractical.

The safe move was obvious: Pivot to something fundable. Chase the grants. Rebuild her career.

In 2016, Tony Xu told investors DoorDash wouldn't compete in New York. Or San Francisco. Or any major city.

He was sending DoorDash to the suburbs. They thought he was crazy.

Xu grew up washing dishes beside his mother in a Chinese restaurant in Champaign, Illinois. His family had immigrated from China with a few hundred dollars. McDonald's was a luxury.