One of the strangest episodes in the history of the tech industry ended as startup events often do: with a party in San Francisco’s eclectic Mission District.
On Tuesday evening, OpenAI announced that Sam Altman was returning as chief executive, five days after the artificial intelligence startup’s board forced him out. At the company’s San Francisco offices, giddy employees snacked on chicken tenders, drank boba tea and champagne and celebrated Mr. Altman’s return late into the night.
Mr. Altman’s reinstatement culminated a corporate drama involving piles of cash, a pressure campaign from allies, intense media attention and an unwavering conviction among some members of the community. AI that they should proceed with caution in what they build.
Now OpenAI, which for two days seemed on the verge of collapse just a year after introducing the popular chatbot ChatGPT, will replace a much-criticized board of directors with a more traditional group including former Treasury Secretary Lawrence Summers and a former executive at software giant Salesforce.
Other board members, who could be drawn from OpenAI’s largest investor, Microsoft, and the AI research community, are expected to join soon. Mr. Altman was not appointed to the board as of Tuesday evening, and it is unclear whether he ever will be.
What seemed to emerge from the mess on Wednesday was a company better able to handle the billions of dollars that have been thrown at it and the attention it has received since ChatGPT’s release. But some are already arguing that it won’t be as suited to OpenAI’s original mission of creating safe AI for the world.
The OpenAI debacle showed that building AI systems is about testing whether businessmen who want to make money can work in sync with researchers who fear that what they develop could eventually eliminate jobs or become a threat if technologies such as autonomous weapons spiral out of control.
The tech industry — perhaps even the world — will be watching to see whether OpenAI is any closer to balancing these conflicting aspirations than it was a week ago.
“We will look back on this period as a very brief, very dramatic incident that gave us a public and dramatic reset,” said Aaron Levie, chief executive of Box, an online data storage provider. “It has to be an organization that is trustworthy and aligned with its board, and at the end of the day, OpenAI is a more valuable organization than it was a week ago.”
When Mr. Altman, 38, was fired shortly after noon on Friday, OpenAI was thrown into chaos. Its employees and Microsoft, which invested $13 billion in the company, were caught off guard.
The AI company has an unusual governance structure. It is controlled by the board of directors of a nonprofit organization, and its investors have no formal way to influence decisions. But no one expected that four board members — including OpenAI’s chief scientist, co-founder Ilya Sutskever — would suddenly fire Mr. Altman, saying he could no longer be trusted with the mission of the company to build artificial intelligence that “benefits”. all humanity. »
The consequences were immediate. OpenAI President Greg Brockman, who also helped found the company eight years ago, resigned in protest.
The board had become increasingly frustrated with Mr. Altman’s behavior over the last year and believed it needed to rein it in, according to two people familiar with the board’s thinking. One episode, in particular, illustrates how strained the relationship between the board and Mr. Altman had become.
Both sides focused on one month of October research paper co-authored by Helen Toner, OpenAI board member and director of strategy at the Center for Security and Emerging Technologies at Georgetown University.
Mr. Altman complained to Ms. Toner that the paper appeared to criticize OpenAI’s efforts to keep its technologies secure while praising a rival. He argued that “any criticism from a board member carries a lot of weight,” he wrote in an email to colleagues.
Ms. Toner defended the paper as academic research, but Mr. Altman and other OpenAI executives, including Mr. Sutskever, later discussed whether to remove her from the board, said a person involved in the conversations.
But Mr. Sutskever, who fears that AI could one day destroy humanity, unexpectedly sided with Ms. Toner and two other board members: Adam D’Angelo, chief executive of the question-and-answer site Quora, and Tasha McCauley, a senior associate scientist at the RAND Corporation.
In a video call on Friday, Mr. Sutskever read Mr. Altman a statement that said Mr. Altman was fired because he was not “always candid in his communications with the board.”
Over the next five days, Mr. Altman and his allies pressured the board to bring him back and for the board to resign. On Sunday, he and company executives negotiated at OpenAI’s offices. Early in the afternoon, a delivery man with a dozen drinks from the Boba Guys chain arrived on a motorcycle outside with two bags. Then a second delivery driver appeared.
That night, talks broke down and the board named Twitch co-founder Emmett Shear as interim CEO.
But Microsoft came up with a plan B: hiring Mr. Altman to run a new AI research lab for Microsoft with Mr. Brockman. OpenAI executives orchestrated a letter from employees saying they would follow Mr. Altman to Microsoft if he was not reinstated. More than 700 of OpenAI’s 770 employees have signed, including Mr. Sutskever, who said in an article on
The pressure caused other board members to become stubborn, said three people familiar with their thinking. They were dismayed that Mr. Altman and his allies were encouraging a mutiny and wondered whether it might be illegal because employees had a contractual obligation to the company and not its chief executive. And they believed that as a board, they were acting with integrity and fulfilling their obligations to the nonprofit’s mission.
The board was still determined to force Mr. Altman to change his behavior, two people familiar with the board’s deliberations said. He was also concerned about some of his recent efforts to raise money for personal interests, such as a drug development startup, even as he raised money for OpenAI.
Discussions from Saturday to Tuesday focused on how to create a board that everyone could trust. For current members, that meant finding administrators who could check Mr. Altman’s power and push for an independent investigation into his behavior.
While Microsoft supported Mr. Altman’s return to OpenAI, the company was working on backup plans, a person familiar with the matter said. Microsoft employees have begun preparing offer letters and recruiting immigration lawyers for OpenAI staff on work visas, the person said.
OpenAI’s three board members spent most of Tuesday on Google Meet video calls, discussing the board’s options. They spoke several times with Microsoft Chief Executive Satya Nadella, one of the people said.
Mr. Altman’s allies proposed a slate of board members consisting of Mr. D’Angelo, Mr. Summers and Bret Taylor, a veteran Silicon Valley executive. Mr. Taylor, who will be the new chairman of the board, oversaw the sale of Twitter to Elon Musk for $44 billion when he led Twitter’s board last year.
Mr. Taylor and Ms. McCauley did not respond to requests for comment. No one involved in the discussions explained how Mr. Summers became an option, and he did not respond to requests for comment Wednesday.
But he has recently established himself as an authority on AI and economics. Mr. Summers you have warned that ChatGPT would come for the “cognitive class,” changing the way doctors make diagnoses, editors work on books, and Wall Street traders invest. He also served on the boards of other technology companies, including financial services company Block, formerly known as Square.
The board views Mr. Summers as an independent thinker with enough management experience to stand up to Mr. Altman, two of the people familiar with the negotiations said.
Tuesday evening, they reached an agreement. Thanksgiving helped. For all their disagreements, everyone agreed the chaos should not extend into Thursday, one person said.
But there is still a lot of work to be done. Over the next six months, the board will analyze and potentially change OpenAI’s unusual structure, one of these people said.
After the decision to bring back Mr. Altman, OpenAI workers filled employee Slack channels with heart emojis and images of a frog, known as a “frog,” which became an unofficial mascot of the company, three employees said.
Late Tuesday, employees gathered at the company office to drink boba tea — an internal reference to the weekend’s media coverage. Mr. Brockman posted a selfie with dozens of smiling employees in the office around midnight.
The caption read: “we are so back.”
Erin Griffith And Yiwen Lu reports contributed.
