Wewnątrz chaotycznego rozstania Elona Muska z OpenAI

As OpenAI was ironing out a fresh deal with Microsoft in 2016 — 1 that would nab the young startup critical compute to build what would become ChatGPT — Sam Altman needed the blessing of his biggest investor, Elon Musk.

“$60MM of compute for $10MM, and input from us on what they deploy in the cloud,” Altman messaged Musk in September 2016, according to newly revealed emails. Microsoft wanted OpenAI to supply feedback on and advance (in tech circles, “evangelize”) Microsoft AI tools like Azure Batch. Musk hated the idea, saying it made him “feel nauseous.”

Altman came back with another offer: “Microsoft is now willing to do the agreement for a full $50m with ‘good religion effort at OpenAI’s sole discretion’ and full common termination rights at any time. No evangelizing. No strings attached. No looking like lame Microsoft marketing pawns. Ok to decision ahead?”

“Fine by me if they don’t usage this in active messaging,” Musk responded. “Would be worth way more than $50M not to seem like Microsoft’s marketing bitch.”

Musk released these emails and others last week as part of a suit he’s filed against OpenAI and Microsoft. They are ostensibly meant to show an anticompetitive partnership between the 2 companies. But primarily, they exposure the details of early collaborations and power struggles between Altman and Musk, who invested between $50 million and $100 million in the earliest iteration of OpenAI. They trace OpenAI’s evolution from an open-source nonprofit to what the suit calls a “closed-source de facto subsidiary” of Microsoft that abandoned its mission to make AI for good. And they lay bare the complete and utter unraveling of Musk and Altman’s once-promising partnership.

“Elon’s 3rd effort in little than a year to reframe his claims is even more baseless and overreaching than the erstwhile ones,” OpenAI spokesperson Hannah Wong wrote in a message to The Verge. “His prior emails proceed to talk for themselves.”

Musk and Altman launched OpenAI united by fears of human-level intelligence in the hands of tech giants like Google — only to see it become the kind of tech juggernaut they feared. After winning a CEO position that Musk coveted, Altman chose to keep OpenAI’s cutting-edge AI behind closed doors, claiming it was too dangerous to be openly released. The decision incensed Musk, who left OpenAI’s board to found his own competitor, xAI. Nearly a decade after the pair founded OpenAI, the 2 companies are amassing billions of dollars and Musk is taking the fight to court — in a race to own what both men see as the inevitable future of computing.

“Been reasoning a lot about whether it’s possible to halt humanity from developing AI,” Altman wrote in 2015 in an email to Musk as a pitch to start OpenAI. “If it’s going to happen anyway, it seems like it would be good for individual another than Google to do it first.”

From its inception, OpenAI was caught between 2 conflicting forces: an idealistic mission to benefit humanity and a cutthroat race against tech behemoths. Musk and Altman agreed that whatever their motivations, securing top talent (along with piles of cash) would be a paramount concern. This early compromise would set the phase for what Musk would later call the startup’s pursuit of profit over principle.

In 2015, the startup was known as YC AI — a laboratory tucked inside Y Combinator’s nonprofit investigation division, YCR. Altman, then president of the startup incubator, leveraged its extended network and resources to attract researchers and money. Musk urged Altman and CTO (now president) Greg Brockman to search over $100 million in funding, cautioning them that anything little would appear paltry compared to the deep pockets of tech giants like Google and Facebook.

“I think we should say that we are starting with a $1B backing commitment. This is real. I will cover whatever anyone else doesn’t provide,” Musk said in 2015 emails revealed by OpenAI earlier this year in consequence to Musk’s lawsuit.

Still, despite Musk’s support and a war chest of millions of dollars, the fledgling organization faced an early challenge that plagues most startups: the fierce competition for top talent. OpenAI might be the hottest place to work in Silicon Valley today, but a decade ago (and long before the launch of ChatGPT), many top AI researchers were improbable to give it a second glance.

In their aggressive bid for the best AI researchers, Altman and his squad devised an different compensation package: a base wage of $175,000, a “part-time partner” title at YC, and 0.25 percent equity in each YC startup batch. (Now, it’s more common for AI researchers to be compensated closer to $1 million annually.) Altman billed it as a “Manhattan task for AI,” per 1 email to Musk, and sensed he could get many of the top 50 researchers to join and “structure it so that the tech belongs to the planet via any kind of nonprofit but the people working on it get startup-like compensation.”

The goal was to assemble an elite founding squad of 7 to 10 members — whatever it took to win the industry’s best minds. Still, Google’s AI lab, DeepMind, was on their heels.

“DeepMind is going to give everyone in OpenAI massive counteroffers next day to effort to kill it,” Altman wrote to Musk in December 2015. “Do you have any objection to me proactively expanding everyone’s comp by 100-200k per year? I think they’re all motivated by the mission here but it would be a good signal to everyone we are going to take care of them over time.”

“Sounds like DeepMind is planning to go to war over this,” Altman added.

Musk approved of the wage bumps, and by February 2016, OpenAI’s founding squad was offered a $275,000 wage plus YC equity, while subsequent hires received a $175,000 wage with performance-based bonuses of $125,000 or equivalent stock in YC or SpaceX. Brockman added that there were 3 peculiar cases: himself, along with cofounders Ilya Sutskever and Trevor Blackwell. It was later reported that Sutskever earned more than $1.9 million in 2016, and he told The New York Times that he “turned down offers for multiple times the dollar amount” he accepted from OpenAI. “I don’t know what will happen if/when Google starts throwing around the numbers they threw at Ilya,” Brockman wrote to Musk as he outlined a plan to poach researchers.

“We request to do what it takes to get the top talent. Let’s go higher. If, at any point, we request to revisit what existing people are getting paid, that’s fine,” Musk replied. “Either we get the best people in the planet or we will get whipped by DeepMind. Whatever it takes to bring on ace talent is fine by me.” He warned that a triumph by DeepMind, which was causing him “extreme intellectual stress,” would be truly bad news with their “one head to regulation the world” philosophy. “They are evidently making major advancement and well they should, given the talent level over there,” Musk added.

It didn’t take long for things to get contentious between the cofounders.

In August 2017, OpenAI was ironing out the specifics of an first backing circular of between $200 million and $1 billion. Shivon Zilis, an ex-OpenAI board associate and Neuralink operations manager who would later bear 3 of Musk’s 12 children, wrote to Musk that Brockman and Sutskever were concerned. They were worried about how a recently founded for-profit branch of OpenAI would distribute equity and control as well as whether Musk — who wanted the occupation of CEO there — would commit adequate time to it. “This is very annoying,” Musk responded, according to 1 of the recently released emails. “Please encourage them to go start a company. I’ve had enough.”

The next month, Sutskever and Brockman escalated with a joint email to Musk and Altman. They expressed fears that Musk would seize “unilateral absolute control” over artificial general intelligence (AGI) if he took power as CEO. At the same time, they questioned Altman’s motivations, asking why “the CEO title is so important” to him. “Is AGI truly your primary motivation? How does it connect to your political goals? How has your thought process changed over time?” the pair asked. (The email doesn’t elaborate on what “politics” refers to, but Altman had become vocally active in California political campaigning earlier that year.) They said that they had let the promise of money cloud their judgement during earlier negotiations, blinding them to concerns they should have raised.

“The goal of OpenAI is to make the future good and to avoid an AGI dictatorship. You are afraid that [DeepMind CEO Demis Hassabis] could make an AGI dictatorship. So do we,” the pair wrote. “So it is simply a bad thought to make a structure where you could become a dictator if you chose to, especially given that we can make any another structure that avoids this possibility.”

The email echoed a common refrain from OpenAI’s founders: that superintelligent AI was a serious threat to humanity, and any single entity controlling that power was even greater. But Musk was unimpressed.

“I will no longer fund OpenAI until you have made a firm commitment to stay or I’m just being a fool who is fundamentally providing free backing for you to make a startup. Discussions are over,” Musk replied. Altman replied that he remains “enthusiastic about the non-profit structure,” which yet led Sutskever and Brockman to back down.

Shortly after the confrontation, Zilis relayed a conversation she had with Altman to Musk. Zilis revealed that Altman “admitted that he lost quite a few trust with Greg and Ilya through this process” and “felt their messaging was inconsistent and felt childish at times.” Altman decided to take 10 days off to process the incident, Zilis added, due to the fact that he “needs to figure out how much he can trust them and how much he wants to work with them.”

Just 5 months after Brockman and Sutskever’s email expressing fears of a power struggle, the situation reached another inflection point. In an altercation that was reported years later, Musk became convinced OpenAI had fallen irreparably behind Google and proposed taking control of the company himself — the very script Brockman and Sutskever had cautioned against.

“My probability assessment of OpenAI being applicable to DeepMind/Google without a dramatic change in execution and resources is 0%. Not 1%. I want it were otherwise,” Musk said in 2018, per emails revealed by OpenAI earlier this year.

OpenAI’s leadership rejected his offer, and Musk departed the board in February 2018, cutting off backing but continuing to offer his support as an adviser.

The failure of Musk, who had by that point reportedly invested $100 million, put OpenAI’s nonprofit model in peril. erstwhile Musk was inactive mostly bankrolling the operation in 2017, Zilis explained to him that OpenAI leadership wanted to rise “$100M out of the gate” due to the fact that “they are of the opinion that the datacenter they request alone would cost that.” So, in 2019, desperate to fund the training data center and reduce reliance on Musk, the squad crafted a unique structure: a capped for-profit company controlled by the nonprofit. LinkedIn cofounder Reid Hoffman and venture capitalist Vinod Khosla participated in the first backing round, which secured pledges of nearly $1 billion but a far smaller first backing of $130 million.

In March 2019, Musk sent Altman an article that implied his engagement in the fresh for-profit structure. “Please be explicit that I have no financial interest in the for-profit arm of OpenAI,” Musk said in the email, which he would later submit for inclusion in the suit. Altman responded simply: “On it.”

OpenAI wields immense influence and power in the AI industry, and the conflict for control was not lost on either Musk or Altman. In the end, Altman emerged victorious — then consolidated his power into near-total control over OpenAI.

The legal merits of Musk’s case are questionable. While he’s accused OpenAI and Microsoft of myriad offenses, much of his suit boils down to accusing Altman of hypocrisy, not typically something that’s punished in a court of law. The case is being heard in California, not in Texas, where Musk has been able to number on a sympathetic ear from a Tesla-stock-owning judge. Still, a suit that accuses OpenAI and Microsoft of anticompetitive practices could garner sympathy while Musk has the ear of US president-elect Donald Trump.

But whatever its outcome, the suit gives Musk a chance to uncover details that form the communicative of OpenAI’s origins and his own role. The exhibits show Altman securing power in the company’s early days, possibly despite the wishes of his cofounders. They underline Altman’s willingness to go toe-to-toe with his for-profit competitors from the beginning. And they supply the public with a clear image of what powers OpenAI: Altman’s willingness to do whatever it takes to get what he wants.

How complete is this narrative? We don’t know. It’s likely quite a few crucial conversations happened offline or in emails that aren’t included. And Musk, obviously, isn’t any less power-hungry; if anything, this suit demonstrates his sheer petty desire to retaliate erstwhile slighted. But as both leaders are competing for a finite amount of venture capitalist cash, he may be betting that he can teardrop down Altman’s reputation — and cement himself as the rightful steward of AGI.