Generated from the production script. Stage directions removed. The final mix may differ slightly from the script; check against the audio before publishing.

Disclaimer: a dramatization based on public records, including court filings, published emails and company posts. Some scenes and dialogue are reconstructed or composited. Voices are AI-generated and perform words from the public record; they are not intended to resemble any real person. Not affiliated with or endorsed by OpenAI or any person depicted. Allegations in court filings are unproven claims. The Role Models is produced by Will Sigmon Media with AI-generated voices (ElevenLabs). Quoted emails, posts and filings are read verbatim and attributed; where the parties disagree, both accounts are given, and a company's post written during a lawsuit is treated as advocacy by one side. Sources for each episode are listed in its show notes.

[Season 1: OpenAI · Episode 2: "The $1 Billion Handshake"]

[COLD OPEN]

Narrator: Saturday. December 12th, 2015. Montreal is waking up hungover. NIPS is winding down. Poster tubes, checkout lines, taxis to the airport. And somewhere in this city, eleven people are about to find out what they're worth. Yesterday afternoon, at half past one Pacific time, they announced to the world that they were walking away from Google, from Facebook, from the richest research labs on Earth, to join a nonprofit that is exactly three days old. It has no office, no computers, and no whiteboards. It has a name that a billionaire picked because the dot-com was available. Yesterday morning, a warning reached Sam Altman, and he forwarded it to Elon Musk. You heard it last time.

Altman (email): "just got word that deepmind is going to give everyone in openAI massive counteroffers tomorrow to try to kill it."

Narrator: Tomorrow. That's today.

Narrator: We don't know every number that was offered that weekend. Nobody has published the letters. But one of the eleven, a Polish researcher named Wojciech Zaremba, later described the money to a reporter.

Zaremba (Wired): "The amount of money was borderline crazy."

Narrator: Borderline crazy. By Wired's account, he was offered two or three times his market value. And his market value was already the kind of number most people never see. So here's the situation. Eleven researchers. Every one of them is holding, or about to be holding, an offer that could change their family's life. The only thing on the other side of the scale is a blog post, a mission, and a billion-dollar pledge that nobody has actually wired to anybody. The founders did something. It's in the record, written two months later in Greg Brockman's own words, in an email about salaries. One sentence of it reads like a thriller.

Brockman (email): "The day after the lab launched, we proactively increased everyone's salary by $100k, telling them that we are financially committed to them as the lab becomes successful, and asking for a personal promise to ignore all counteroffers and trust we'll take care of them."

Narrator: A personal promise. Not a contract. Not stock. A promise, looking someone in the eye: ignore the money, trust us. That's the handshake. The public would hear about a billion dollars. What actually held OpenAI together that weekend was a hundred thousand dollars a head and a promise. Years from now, reporters will open this organization's tax filings and add up every dollar that ever arrived.

Narrator: The number will not be a billion.

Narrator: But this is December 2015, and the ink on the handshake is still wet. This is the story of the most expensive promise in Silicon Valley. And of the people who had to keep it before the money was real.

Title Card: THE ROLE MODELS. Season One: OpenAI. Episode Two: "The One-Billion-Dollar Handshake."

[CHAPTER 2 – "Borderline Crazy"]

Narrator: To understand why that promise worked, why nine brilliant people said no to borderline-crazy money, you have to understand what they'd said yes to. And for that, we go back a few weeks.

Narrator: Fall, 2015. Greg Brockman has a problem that money can't fix. He's the former CTO of Stripe. He's resigned to build an AI lab that doesn't exist, on the strength of a few dinners and an email thread. He has backers, and he has a mission. What he doesn't have is scientists. So he goes to one of the three godfathers of deep learning, Yoshua Bengio in Montreal, the only one of the three who hasn't gone to work for a tech giant. Hinton is at Google. LeCun is at Facebook. Bengio stayed academic. And according to Wired's reconstruction of that fall, Bengio drew up a list of the best researchers in the field. Picture it as a treasure map with every X marked on someone else's property. Nearly every name on it works for Google, Facebook, or a university that Google and Facebook are circling. Brockman calls them. They like the idea. They're wary of the leap. Everybody wants to go if everybody else goes, and nobody wants to go first. So Brockman does something that isn't in any startup playbook.

Narrator: He picks the ten he wants most and invites them to spend a Saturday in Napa Valley. A winery. And the part Brockman remembered afterward wasn't the wine. It was the drive.

Brockman (Wired): "An underrated way to bring people together are these times where there is no way to speed up getting to where you're going. You have to get there, and you have to talk."

Narrator: Think about that as a recruiting strategy. You can't recruit the best AI researchers on Earth with money, because Google will always have more. So you trap them in a car on Highway 29 for an hour and let them talk to each other. The pitch isn't come work for me. It's look who else is in this car.

Brockman (Wired): "It was one of those days where you could tell the chemistry was there."

Narrator: And Ilya Sutskever, the researcher everyone in that car was secretly measuring themselves against, put it in seven words.

Sutskever (Wired): "the wine was secondary to the talk."

Narrator: At the end of the day, Brockman asks all ten to join. He gives them three weeks. Nine say yes.

Narrator: Those nine, plus the people already inside, are the ones in Montreal now with offers in their pockets. And the offers aren't subtle. Wired would later report that the last-minute offers, some made at the conference itself, were large enough to force Musk and Altman to delay the announcement. The launch was supposed to happen sooner. It slid because the other side was bidding while the founders were still writing the press release. Here's Zaremba's read on why.

Narrator: According to Wired, he felt the money was at least as much an effort to prevent the creation of OpenAI as it was to win him. That changes the math. It's the difference between we want you and we want this thing not to exist. And it had the opposite effect.

Zaremba (Wired): "I realized that OpenAI was the best place to be."

Narrator: Sutskever felt the pull too. Google had built its reputation on people like him, and it didn't let go politely.

Sutskever (Wired): "They did make it very compelling for me to stay, so it wasn't an easy decision. But in the end, I decided to go with OpenAI, partly of because of the very strong group of people and, to a very large extent, because of its mission."

Narrator: To a very large extent, because of its mission. Remember that phrase. This whole season is about what happens to a mission when it gets expensive. Now, the founders' side of that weekend. On launch day, Elon Musk told the New York Times something that was simple and true.

Musk (NYT): "The people on the team have all been offered substantially more to work at other places."

Narrator: And Sam Altman added the line that would become the lab's recruiting creed:

Altman (NYT): "It is lucky for us the best people in any field generally care about what is best for the world."

Narrator: Generally. It's a careful word, and a smart one. Generally isn't always. And the day after launch, the founders didn't bet everything on generally. They added a hundred thousand dollars a year to every founding salary, from a hundred and seventy-five to two seventy-five, and asked for a promise. Mission first. But mission with a raise.

Narrator: It's also worth hearing what Musk wrote to the team at 4:41 that launch-day afternoon, past the part we heard last time. After "I like the odds," after "the vector sum of the people," the email keeps going.

Musk (email): "I would recommend paying close attention to people who haven't completed their grad or even undergrad, but are obviously brilliant. Better to have them join before they achieve a breakthrough."

Narrator: Better to have them join before they achieve a breakthrough. That's the arbitrage. You can't outbid Google for the person who has already made the breakthrough. So you find the person who's about to, and you hire them while they're still cheap. It's a talent scout's instinct. It's also an early admission. Even on day one, with a billion-dollar headline, the founders knew they couldn't win a bidding war. They'd have to win before the bidding started.

[CHAPTER 3 – "Some Number in Excess of a Billion"]

Narrator: Launch day. While the team was fending off offers, the founders were doing press, and one conversation from that day holds almost every tension this season will explore. Steven Levy, one of the most respected technology journalists alive, interviewed Musk and Altman together for Backchannel. The piece published December 11th. He asks the question everyone wants answered. Is there really a billion dollars?

Musk (Backchannel interview): "I think it's fair to say that the commitment actually is some number in excess of a billion. We don't want to give an exact breakdown but there are significant contributions from all the people mentioned in the blog piece."

Narrator: Some number in excess of a billion. Hold on to that sentence. We'll come back to it with a calculator. Levy asks about oversight. Who watches the watchers? Altman's answer is honest about how early it is.

Altman (Backchannel): "We do want to build out an oversight for it over time. It'll start just with Elon and me."

Narrator: It'll start just with Elon and me. Two men. That's the governance on day one of the organization built to make sure no small group controls the future. They mean it as a temporary scaffold. File it anyway. The question of who sits on the board of this thing, who can overrule whom, will come back at the worst possible moment eight years from now. Then Levy asks the question at the heart of the whole idea. If you give AI to everyone, you give it to the bad guys too.

Narrator: His phrasing, and I promise I'm not making this up, was: If I'm Dr. Evil and I use it, won't you be empowering me?

Musk (Backchannel): "I think that's an excellent question and it's something that we debated quite a bit."

Altman (Backchannel): "We think its far more likely that many, many AIs, will work to stop the occasional bad actors than the idea that there is a single AI a billion times more powerful than anything else."

Narrator: Here's how that theory is supposed to work, because it isn't crazy. Think of it like an immune system. One all-powerful AI is a single point of failure. If it goes wrong, or the wrong person gets it, nothing can stop it. But if there are millions of AIs, held by millions of people, then any one bad actor faces a crowd. Power checks power. It's the same logic as the internet, and as democracy: don't make anything too big to fail. Musk put it in one line.

Musk (Backchannel): "I think the best defense against the misuse of AI is to empower as many people as possible to have AI. If everyone has AI powers, then there's not any one person or a small set of individuals who can have AI superpower."

Narrator: Not any one person. That's the whole founding philosophy of OpenAI in two sentences. Keep it in mind for two years from now, when the man who said it sits at a negotiating table. And then Altman says the most honest thing anyone says all launch week.

Altman (Backchannel): "I wish I could count the hours that I have spent with Elon debating this topic and with others as well and I am still not a hundred percent certain."

Narrator: Not a hundred percent certain. On launch day. About the core premise. That isn't weakness. It's a man who has thought about this enough to know where the floor is soft. Before we leave the interview, there's one more exchange, and it will matter later. Levy asks whether Musk's other companies will benefit. Musk says there are no specific plans, and then adds this almost in passing:

Musk (Backchannel): "Probably Tesla will have more real world data than any other company in the world."

Narrator: Tesla. On OpenAI's launch day, in OpenAI's launch interview. Just a data point. For now.

Narrator: Levy asked Musk how he'd find the time, as CEO of two companies and chairman of a third. Musk said he'd spend an afternoon in the office every week or two, and then:

Musk (Backchannel): "AI safety has been preying on my mind for quite some time, so I think I'll take the trade-off in peace of mind."

Narrator: Peace of mind. That's what the billion was buying him. Hold that thought too.

[CHAPTER 4 – "No Whiteboards"]

Narrator: Early January, 2016. San Francisco. The Mission District. A three-unit Victorian on 21st Street. Up the stairs is Greg Brockman's apartment, freshly renovated, with a skylit living room where the attic used to be. This is OpenAI's headquarters.

Narrator: The billion-dollar AI lab walks in carrying laptops. Some of the best machine-learning researchers on the planet stand in a living room and look around for the one tool every research lab since the dawn of research has had. There isn't one. According to Wired, the project was so new that they didn't even have whiteboards.

Narrator: So they went out and bought some. That same day.

Narrator: If you want the whole story of OpenAI in a single image, it might be this one. The headline says a billion dollars. Nine people are standing in a living room waiting for someone to come back from the office-supply store. And the first thing written on those whiteboards? We don't know. But we know what they planned to work on, because Brockman told Wired. Two things. The first is reinforcement learning. Here's how the impossible thing works. You don't teach the machine the right answer. You don't even know the right answer. You give it a goal and a score, and you let it try. Over and over, a million times. Every time it scores higher, the behavior that led there gets a little stronger. Every time it fails, that behavior gets a little weaker. It's how a dog learns to sit and how a toddler learns to walk. Nobody hands the toddler a manual on balance. Gravity is the teacher and falling is the grade. The second is unsupervised learning: machines that learn without anyone labeling anything. Remember Google's cat neuron from last episode, the network that found cats in YouTube videos nobody had tagged. That, but for everything.

Brockman (Wired): "If you have really good unsupervised learning, machines would be able to learn from all this knowledge on the Internet—just like humans learn by looking around—or reading books."

Narrator: Learn from all this knowledge on the Internet. Or reading books. In January 2016, in a living room, that's an aspiration. Seven years later it will be the most valuable sentence in technology. We aren't there yet. File it.

Narrator: Brockman had a model for what this place should be, and it wasn't Google. It was Xerox PARC, the legendary 1970s lab that invented the graphical user interface, the laser printer, and the modern personal computer, and then watched Steve Jobs walk out the door with the ideas.

Brockman (Wired): "You want a shared vision, without central control."

Narrator: Without central control. In a lab whose governance, on launch day, was "just Elon and me." Nobody in the living room would have called that a contradiction. It was January. The whiteboards were new. Everything was possible.

Narrator: Saturday, January 2nd, 2016. Musk is in a good mood. Eleven days ago, on December 21st, SpaceX landed a Falcon 9 booster upright on a pad at Cape Canaveral, the first time an orbital rocket had come back down and stood on its own legs. The world watched a skyscraper fall out of the sky and land like a gymnast. That morning, an email arrives from someone whose name is redacted in the court exhibits. It opens with congratulations on the landing. Then the tone changes. The correspondent says they've watched Musk, Altman, and the other OpenAI people giving interview after interview praising open-source AI, and asks whether they realize that isn't a cure-all for the safety problem. They say there are good arguments that the approach could be dangerous and could increase the risk to the world. And they attach a link to an essay published two weeks earlier on a blog called Slate Star Codex, by a psychiatrist writing as Scott Alexander. The title is a question: "Should AI Be Open?" The argument goes like this. If the first superintelligence is dangerous, and safety is harder than capability, then publishing everything is like handing out the blueprints to the bomb and hoping the good guys finish first. It's the Dr. Evil question again, three weeks later, from someone who clearly isn't satisfied with the launch-day answer.

Narrator: Musk forwards it to Altman, Sutskever, and Brockman. At 9:06 AM, Ilya Sutskever replies. This is the chief scientist of OpenAI, eleven days in, explaining what the word "open" means.

Sutskever (email): "As we get closer to building AI, it will make sense to start being less open. The Open in openAI means that everyone should benefit from the fruits of AI after its built, but it's totally OK to not share the science (even though sharing everything is definitely the right strategy in the short and possibly medium term for recruitment purposes)."

Narrator: Let's go through that slowly, because eight years later people will fight over it in federal court. As we get closer, it will make sense to be less open. The Open in OpenAI means everyone should benefit from the fruits of AI after it's built. It's totally OK to not share the science. And the parenthesis: sharing everything is definitely the right strategy in the short and possibly medium term, for recruitment purposes. For recruitment purposes. The same openness that three weeks earlier helped nine researchers turn down borderline-crazy money. Five minutes later, at 9:11 AM, Musk replies. The whole reply:

Musk (email): "Yup"

Narrator: One word. In the first episode, the story started with a seven-word email. Then a three-word one. Now it's down to one. We have to be careful here, the way we were careful last time, because this email is a weapon in a lawsuit. When OpenAI published it in March 2024, they presented it as proof that Musk understood from the very beginning that "open" never meant open-sourcing everything, and that he agreed with a single word. Musk's side sees the founding differently. His lawsuit treats open, public-benefit technology as the heart of the original bargain, the thing he says he was promised and that was later betrayed. Both things can be true at once. A man can reply "yup" to an email on a Saturday morning after landing a rocket, and still believe years later that the soul of the thing was openness. We'll hear both sides in full when the lawsuit arrives. For now, just notice the date. January 2nd, 2016. The lab is three weeks old. It doesn't have furniture. And its chief scientist has already written that the "Open" is about who benefits, not who gets to see.

Narrator: March 2016. Brockman is answering fact-check questions for that same Wired story. One of them asks, point-blank, whether it would be accurate to say OpenAI is giving away all of its research. He sends his draft answer to Musk first.

Brockman (email): "In the short term, we believe the best approach is giving away our research. But longer-term, this might not be the best approach: for example, it might be better not to immediately share a potentially dangerous technology."

Narrator: Musk's reply:

Musk (email): "Sounds good"

Narrator: In the Wired story itself, Nick Bostrom, the philosopher whose book started Musk down this road, puts the objection in one line.

Bostrom (Wired): "If you have a button that could do bad things to the world, you don't want to give it to everyone."

Narrator: The man who wrote the warning label now has doubts about the cure.

[CHAPTER 5 – "Whipped by DeepMind"]

Narrator: February 2016. Two months in. The lab has a whiteboard now. It has some researchers. What it doesn't have yet is a result, and Elon Musk has started to count. Just after midnight on February 19th, he sends a long email to Sutskever, Brockman, and Altman. Part of it is redacted in the court exhibits. What remains reads like a man who can't sleep.

Musk (email): "The probability of DeepMind creating a deep mind increases every year. Maybe it doesn't get past 50% in 2 to 3 years, but it likely moves past 10%. That doesn't sound crazy to me, given their resources."

Narrator: DeepMind creating a deep mind. It sounds like a pun and isn't one. He's putting odds on the end of the world as he understands it, the unipolar mind from episode one, and the odds are rising.

Musk (email): "In any event, I have found that it is far better to overestimate than underestimate competitors."

Narrator: And then comes the deadline.

Musk (email): "It will be important for OpenAI to achieve something significant in the next 6 to 9 months to show that we are for real. Doesn't need to be a whopper breakthrough, but it should be enough for key talent around the world to sit up and take notice."

Narrator: Six to nine months. For a lab that got its whiteboards in January. Hours later, at 10:28 that morning, Sutskever answers. Calm. The scientist to the rocket man.

Sutskever (email): "We can't build AI today because we lack key ideas (computers may be too slow, too, but we can't tell). Powerful ideas are produced by top people. Massive clusters help, and are very worth getting, but they play a less important role."

Narrator: Listen to what Ilya believes in February 2016. Powerful ideas are produced by top people. Massive clusters help, but they play a less important role. Ideas over hardware. People over machines. Keep that sentence somewhere safe. In a little over a year, Ilya Sutskever is going to write another email, and he's going to sound very different. Two days later, February 21st. Brockman writes the email we heard in the cold open, subject line "compensation framework." The lab is making its first full-time offers since the launch, and Brockman admits he isn't comfortable deciding on his own. He lays out the market, and the numbers are extraordinary.

Brockman (email): "Wojciech was verbally offered ~$1.25M/year at FAIR (no concrete letter though)"

Narrator: FAIR is Facebook AI Research. One and a quarter million a year, verbally, for the man who told Wired the money was borderline crazy. That lines up.

Brockman (email): "Apple is a bit desperate and offering people $550k cash (plus stock, presumably). I don't think anyone good is saying yes."

Narrator: And then a name. Ian Goodfellow, at Google Brain, the inventor of generative adversarial networks, the two-network duel where one forges and one detects until the forgeries become perfect. Brockman calls him one of the top two scientists in the field that OpenAI doesn't have.

Brockman (email): "Ian Goodfellow is currently getting $165k cash + $600k stock/year at Google."

Narrator: Brockman's question to Musk comes down to this: do we stick to our flat, mission-first pay, and ask people to take a pay cut for the cause? Or do we slide when Google starts throwing numbers? The reply comes at 12:09 AM on February 22nd. Musk writes after midnight.

Musk (email): "We need to do what it takes to get the top talent. Let's go higher. If, at some point, we need to revisit what existing people are getting paid, that's fine. Either we get the best people in the world or we will get whipped by Deepmind. Whatever it takes to bring on ace talent is fine by me. Deepmind is causing me extreme mental stress. If they win, it will be really bad news with their one mind to rule the world philosophy."

Narrator: You've heard that last line before. In episode one, we quoted "extreme mental stress" from press coverage of Musk's own court filings, and we warned you that his lawyers chose which words to show you. Here's the whole email, and here's where it sits: the middle of the night, answering a question about salaries. The fear from 2014 hasn't faded. It's become a payroll policy. Whatever it takes. Eleven minutes later, Brockman:

Brockman (email): "Read you loud and clear."

Narrator: The nonprofit that couldn't win a bidding war has just been told by its biggest backer to enter one. And it works. That spring, OpenAI lands Ian Goodfellow. Wired would report the hire that April, a senior researcher lured away from Google Brain. The price stays secret for two years.

Narrator: April 2018. OpenAI's 2016 tax filing becomes public, and the New York Times runs a headline that sounds like a contradiction: researchers making more than a million dollars, even at a nonprofit. Goodfellow, hired in March 2016, is listed at more than eight hundred thousand dollars for his partial year. And Ilya Sutskever, research director, at one million, nine hundred thousand. That number rhymes with the one Walter Isaacson reported Musk telling him, the one you heard in the first episode.

Narrator: So much for mission over money. Except, and this matters, Sutskever would tell the Times something every other number in this chapter backs up.

Sutskever (NYT): "I turned down offers for multiple times the dollar amount I accepted at OpenAI. Others did the same."

Narrator: Both things are true. The nonprofit paid more than any nonprofit had ever paid, and it still paid less than the market. It was the most expensive charity in Silicon Valley and the cheapest place the best researchers could possibly work.

Narrator: And then, in March, the enemy lands a punch the whole world sees. Seoul. The Four Seasons Hotel. March 2016.

Narrator: DeepMind's program AlphaGo is playing Lee Sedol, one of the greatest Go players alive, in a five-game match. Here's why Go was supposed to be safe. Chess was cracked in 1997. Go was the game humans kept. It has more possible positions than there are atoms in the observable universe, so you can't calculate your way through it. You need intuition, a feel for the board that the best players spend their whole lives building. March 10th. Game two. Move thirty-seven. AlphaGo places a stone on the fifth line, out near the edge, in a spot so strange the commentators assume it's a mistake.

Narrator: It isn't a mistake. It's a move no human master would have played, and it wins. Lee Sedol gets up and leaves the room, and when he comes back he's facing a different kind of opponent. AlphaGo wins the match four games to one. Now picture where OpenAI is that same week. A few months old. A few whiteboards. Promises of papers. And across the Pacific, the lab that Musk said was causing him extreme mental stress has just done something every newspaper on Earth calls historic. Remember his February line: show that we are for real. In March, DeepMind showed the world that it was real. The counterweight has barely been built, and the thing it's supposed to balance just got heavier.

[CHAPTER 6 – "The Calculator"]

Narrator: Back to the handshake. In the cold open I said the billion-dollar headline wasn't what held OpenAI together that first weekend. It was a raise and a promise. The founders promised they would take care of the team. To keep a promise like that, the money has to exist. So, as promised, let's come back to launch day with a calculator.

Narrator: The launch post said: In total, these funders have committed one billion dollars. The named funders were Sam Altman, Greg Brockman, Reid Hoffman, Jessica Livingston, Elon Musk, Peter Thiel, Amazon Web Services, Infosys, and YC Research. Musk told Levy the commitment was some number in excess of a billion. He'd told his co-founders three weeks earlier, in writing: I will cover whatever anyone else doesn't provide. And the founders told the New York Times, the same week, that they weren't ready to say who had given how much, or how fast the money would be spent, and that the funding would come year by year. That's the key. A commitment in philanthropy isn't a wire transfer. It's a handshake. It's a donor saying I intend to, and it becomes real one check at a time. So how many checks arrived?

Narrator: May 2023. TechCrunch reporters go through OpenAI's federal tax filings, the public forms every charity has to file, and add up every donation from the lab's first day. From inception through 2021, total donations to the OpenAI nonprofit came to one hundred and thirty-three point two million dollars.

Narrator: Not a billion. A hundred and thirty-three million. One of the named funders, YC Research, never gave a single dollar. And the dollars they could trace definitively to Elon Musk in those filings? About fifteen million.

Narrator: Now, and this is where we have to be most careful all episode, fifteen million is not the full story of what Musk gave. It's what reporters could trace with certainty in those filings, and money doesn't always move in straight lines. Here's one route we can document. In 2016, the Musk Foundation gave ten million dollars to YC.org, a charity run by Sam Altman's side of Y Combinator. YC.org didn't yet have its IRS nonprofit status, so the money couldn't go straight to OpenAI. So YC.org passed ten million on to OpenAI later that year. Money goes around corners, and the filings don't always show where it started. So what's the real number? Here's where each side lands, in their own words.

Narrator: OpenAI, in a March 2024 blog post signed by Altman, Brockman, Sutskever, Schulman, and Zaremba, written in response to Musk's lawsuit:

Openai (blog post): "In total, the non-profit has raised less than $45M from Elon and more than $90M from other donors."

Narrator: Musk's lawyers, in his federal complaint that August:

Musk'S Complaint (read): "Musk paid over $44 million to OpenAI, Inc. and invested substantial time and resources, including using his valuable track record and notoriety to recruit leading scientific talent and attract further financial contributions to the nonprofit."

Narrator: "Less than forty-five million." "Over forty-four million." Sworn enemies in a courtroom, and on this one point they're within a million dollars of each other. Somewhere between forty-four and forty-five million, give or take how you count. Musk himself has told a bigger story in public. At various times he's said he put in a hundred million, and later fifty. Those numbers don't match the filings or his own lawyers. But he's also right about something the filings can't measure: without his name, his fear, and his "I will cover whatever anyone else doesn't provide," there might never have been a launch to fund. Both are true. The money was smaller than the legend, and the legend is part of why there was any money at all.

Narrator: So add it up. The headline: one billion dollars. Some number in excess of a billion, in Musk's words. The reality, over the nonprofit's whole life, from every donor combined: one hundred and thirty-three million. About thirteen cents on the dollar. And for a while, in 2016, that was enough. It's worth saying clearly, because the math only turns cruel later. OpenAI's first full year, according to its tax filing, cost about eleven million dollars. Around fifty-two people. Seven million of the eleven went to salaries and benefits. People, not machines. Just as Sutskever wrote: powerful ideas are produced by top people. A nonprofit that spends eleven million a year can live for a long time on a hundred million. On that 2016 math the handshake holds. As long as the thing they're building keeps costing what it costs in 2016.

[OUTRO – CLIFFHANGER]

Narrator: Let's step back and look at the whole first season of OpenAI's life. The first few months. A billion-dollar headline. A launch delayed because the other side was bidding at the conference. A Saturday of counteroffers, and a founding team that turned them down for a hundred-thousand-dollar raise and a personal promise. A living room in the Mission with no whiteboards. A chief scientist who wrote, three weeks in, that the "Open" was about who benefits, and a founder who answered yup. A midnight email saying whatever it takes, because DeepMind was causing extreme mental stress. And across the ocean, a stone on the fifth line of a Go board that made the enemy look unbeatable. And underneath all of it, a number. Thirteen cents on the dollar. But here's what nobody in that living room could see yet. The billion was never really about the money. Musk said it himself back in November, and you heard it last time: a much bigger number than one hundred million, to avoid sounding hopeless. The billion was a flag you plant on a hill so people will climb it. It worked. They climbed. What actually kept OpenAI alive in 2016 was the thing Brockman and Altman did the day after launch. They looked eleven people in the eye and said trust us, we'll take care of you. That's the real handshake. It was never signed by anyone, and it's worth more than any pledge on the launch page. A handshake like that has a cost, though. Every year it has to be kept again. Every year, the people who gave their word have to believe the people who took it can still pay. In February, Ilya Sutskever wrote that massive clusters play a less important role. People over machines. Ideas over hardware. On that bet, eleven million dollars a year is plenty.

Narrator: But somewhere in the next eighteen months, the most important scientist at OpenAI is going to change his mind. He's going to look at the numbers coming out of Google, and at the experiments his own team can't afford to run, and he's going to write to Musk, in words you'll hear next time, that the history of AI might be a story of computers that were simply too slow. And when that bet replaces the old one, the math of the handshake breaks. Eleven million won't be enough. A hundred and thirty-three million won't be enough. A billion won't be enough, even if it had ever existed. A nonprofit is about to try to outspend Google. Next time, on The Role Models: a toolkit, a universe, and an army of bots that play forty-five thousand years of a video game. Microsoft calls with a deal that makes a founder feel sick. The compute bill starts doubling. And one of the founders has another AI project of his own to feed. Episode Three: "The Nonprofit Paradox."