Time特稿：萨姆·奥特曼，年度CEO

By Naina Bajekal/San Francisco and Billy Perrigo/San Francisco | Photograph by Joe Pugliese for TIME
作者:NAINA BAJEKAL/旧金山 和 BILLY PERRIGO/旧金山 | TIME杂志摄影师Joe Pugliese拍摄

2023年12月6日 上午7:42 美国东部标准时间

It was a strange Thanksgiving for Sam Altman. Normally, the CEO of OpenAI flies home to St. Louis to visit family. But this time the holiday came after an existential struggle for control of a company that some believe holds the fate of humanity in its hands. Altman was weary. He went to his Napa Valley ranch for a hike, then returned to San Francisco to spend a few hours with one of the board members who had just fired and reinstated him in the span of five frantic days. He put his computer away for a few hours to cook vegetarian pasta, play loud music, and drink wine with his fiancé Oliver Mulherin. “This was a 10-out-of-10 crazy thing to live through,” Altman tells TIME on Nov. 30. “So I’m still just reeling from that.”
这对萨姆·奥特曼来说是个不同寻常的感恩节。通常,OpenAI的首席执行官会飞回圣路易斯探望家人。但这一次,假期之后,他经历了一场关于控制一家被一些人认为掌握着人类命运的公司的生死斗争。奥特曼感到疲惫。他去了纳帕谷的牧场徒步,然后回到旧金山与一位董事会成员共度了几个小时,就在五天的狂乱中,这位成员刚刚解雇又重新任命了他。他把电脑收起来几个小时,煮了素食意大利面,和他的未婚夫奥利弗·穆尔赫林一起大声播放音乐,喝酒。奥特曼在11月30日对《时代》杂志说:“这是我经历过的疯狂程度十分满分的事情,所以我现在还在从中缓过劲来。”

0 seconds of 30 secondsVolume 0%
0秒至30秒音量0%

We’re speaking exactly one year after OpenAI released ChatGPT, the most rapidly adopted tech product ever. The impact of the chatbot and its successor, GPT-4, was transformative—for the company and the world. “For many people,” Altman says, 2023 was “the year that they started taking AI seriously.” Born as a nonprofit research lab dedicated to building artificial intelligence for the benefit of humanity, OpenAI became an $80 billion rocket ship. Altman emerged as one of the most powerful and venerated executives in the world, the public face and leading prophet of a technological revolution.
我们正好在OpenAI发布ChatGPT一周年之际进行对话,这是有史以来采用速度最快的技术产品。这款聊天机器人及其继任者GPT-4的影响是变革性的——对公司和世界都是如此。Altman说,对许多人来说,2023年是“他们开始认真对待人工智能的一年”。OpenAI诞生为一个致力于为人类利益构建人工智能的非营利研究实验室,它已经成为一家价值800亿美元的火箭般的公司。Altman成为世界上最有权力和最受尊敬的高管之一,是技术革命的公众面孔和领军先知。

Until the rocket ship nearly imploded. On Nov. 17, OpenAI’s nonprofit board of directors fired Altman, without warning or even much in the way of explanation. The surreal maneuvering that followed made the corporate dramas of Succession seem staid. Employees revolted. So did OpenAI’s powerful investors; one even baselessly speculated that one of the directors who defenestrated Altman was a Chinese spy. The company’s visionary chief scientist voted to oust his fellow co-founder, only to backtrack. Two interim CEOs came and went. The players postured via selfie, open letter, and heart emojis on social media. Meanwhile, the company’s employees and its board of directors faced off in “a gigantic game of chicken,” says a person familiar with the discussions. At one point, OpenAI’s whole staff threatened to quit if the board didn’t resign and reinstall Altman within a few hours, three people involved in the standoff tell TIME. Then Altman looked set to decamp to Microsoft—with potentially hundreds of colleagues in tow. It seemed as if the company that catalyzed the AI boom might collapse overnight.
直到火箭飞船几乎内爆。11月17日,OpenAI的非营利董事会在没有任何预警或多少解释的情况下解雇了阿特曼。随后发生的超现实操纵使得《继承之战》中的企业剧情显得平淡无奇。员工起义。OpenAI的强大投资者也是如此;其中一位甚至毫无根据地猜测,将阿特曼推出窗外的董事之一是中国间谍。公司的远见卓识首席科学家投票赞成驱逐他的联合创始人,随后又回头。两位临时首席执行官来了又走。参与者通过自拍、公开信和社交媒体上的心形表情进行姿态展示。与此同时,公司的员工和董事会在“一场巨大的博弈”中对峙,一位熟悉讨论的人士说。有一次,OpenAI的全体员工威胁说,如果董事会不辞职并在几小时内重新任命阿特曼,他们就会集体辞职,三位参与对峙的人士告诉《时代》杂志。然后阿特曼看起来准备转投微软——可能带着数百名同事。这家催化了人工智能热潮的公司似乎可能一夜之间崩溃。

In the end, Altman won back his job and the board was overhauled. “We really do feel just stronger and more unified and more focused than ever,” Altman says in the last of three interviews with TIME, after his second official day back as CEO. “But I wish there had been some other way to get there.” This was no ordinary boardroom battle, and OpenAI is no ordinary startup. The episode leaves lingering questions about both the company and its chief executive.
最终,阿特曼赢回了他的工作,董事会也进行了重组。“我们确实感觉比以往任何时候都更强大、更团结、更专注,”阿特曼在与《时代》杂志的三次采访中的最后一次采访中说,那是他正式回归CEO职位的第二天。“但我希望有其他方式可以达到这个目的。”这不是一场普通的董事会之争,OpenAI也不是一家普通的初创公司。这一事件留下了关于公司及其首席执行官的持续疑问。

Altman, 38, has been Silicon Valley royalty for a decade, a superstar founder with immaculate vibes. “You don’t fire a Steve Jobs,” said former Google CEO Eric Schmidt. Yet the board had. (Jobs, as it happens, was once fired by Apple, only to return as well.) As rumors swirled over the ouster, the board said there was no dispute over the safety of OpenAI’s products, the commercialization of its technology, or the pace of its research. Altman’s “behavior and lack of transparency in his interactions with the board” had undermined its ability to supervise the company in accordance with its mandate, though it did not share examples.
阿特曼,38岁,已经是硅谷的皇室成员十年了,一个拥有无懈可击氛围的超级明星创始人。“你不会解雇一个史蒂夫·乔布斯,”前谷歌首席执行官埃里克·施密特说。然而董事会却解雇了他。(碰巧的是,乔布斯曾经被苹果解雇,后来又回来了。)随着关于解雇的传言四起,董事会表示,对OpenAI产品的安全性、其技术的商业化或其研究的进度并无争议。尽管没有分享例子,但董事会表示,阿特曼在与董事会互动时的“行为和缺乏透明度”破坏了其根据授权对公司进行监督的能力。

Interviews with more than 20 people in Altman’s circle—including current and former OpenAI employees, multiple senior executives, and others who have worked closely with him over the years—reveal a complicated portrait. Those who know him describe Altman as affable, brilliant, uncommonly driven, and gifted at rallying investors and researchers alike around his vision of creating artificial general intelligence (AGI) for the benefit of society as a whole. But four people who have worked with Altman over the years also say he could be slippery—and at times, misleading and deceptive. Two people familiar with the board’s proceedings say that Altman is skilled at manipulating people, and that he had repeatedly received feedback that he was sometimes dishonest in order to make people feel he agreed with them when he did not. These people saw this pattern as part of a broader attempt to consolidate power. “In a lot of ways, Sam is a really nice guy; he’s not an evil genius. It would be easier to tell this story if he was a terrible person,” says one of them. “He cares about the mission, he cares about other people, he cares about humanity. But there’s also a clear pattern, if you look at his behavior, of really seeking power in an extreme way.”
采访了奥特曼圈子里的20多人——包括现任和前任OpenAI员工、多位高级执行官,以及多年来与他密切合作的其他人——揭示了一个复杂的画像。认识他的人描述奥特曼为和蔼可亲、聪明绝顶、非比寻常的有驱动力,并且擅长团结投资者和研究人员,共同致力于他创造普适人工智能(AGI)以造福整个社会的愿景。但是,有四个曾与奥特曼合作多年的人也说,他可能会变得难以捉摸——有时,甚至是误导性和欺骗性的。两个熟悉董事会程序的人说,奥特曼擅长操纵人,他反复收到反馈,有时他为了让人觉得他同意他们的观点,而实际上并不同意,这是不诚实的。这些人将这种模式视为巩固权力的更广泛尝试的一部分。“在很多方面,萨姆是个真正的好人;他不是邪恶的天才。如果他是个可怕的人,讲这个故事会更简单,”其中一个人说。“他关心使命,关心他人,关心人类。” 但如果你观察他的行为,也会发现一个明显的模式,那就是他以一种极端的方式真正寻求权力。"

Buy the Person of the Year issue here
在这里购买年度人物杂志

An OpenAI spokesperson said the company could not comment on the events surrounding Altman’s firing. “We’re unable to disclose specific details until the board’s independent review is complete. We look forward to the findings of the review and continue to stand behind Sam,” the spokesperson said in a statement to TIME. “Our primary focus remains on developing and releasing useful and safe AI, and supporting the new board as they work to make improvements to our governance structure.”
OpenAI的一位发言人表示,公司无法对Altman被解雇的事件发表评论。“在董事会的独立审查完成之前,我们无法披露具体细节。我们期待审查的结果,并继续支持Sam,”该发言人在一份给时代杂志的声明中说。“我们的主要焦点仍然是开发和发布有用且安全的人工智能,并支持新董事会在改善我们的治理结构方面所做的工作。”

Altman has spent much of the past year assuring the public that OpenAI takes seriously the responsibility of shepherding its powerful technology into the world. One piece of evidence he gave was OpenAI’s unusual hybrid structure: it is a for-profit company governed by a nonprofit board, with a mandate to prioritize the mission over financial interests. “No one person should be trusted here,” Altman told a Bloomberg Technology conference in June. “The board can fire me. I think that’s important.” But when that happened only for Altman to maneuver his way back, it seemed to underscore that this accountability was a mirage. How could a company that had brought itself to the brink of self-destruction overnight be trusted to safely usher in a technology that many believe could destroy us all?
奥特曼在过去的一年里花了很多时间向公众保证,OpenAI非常重视将其强大技术引入世界的责任。他提供的一个证据是OpenAI不同寻常的混合结构:它是一个由非营利组织董事会管理的营利性公司,其任务是将使命置于财务利益之上。奥特曼在6月的彭博科技大会上说:“这里不应该只信任一个人。董事会可以解雇我。我认为这很重要。”但当这种情况发生,只是为了让奥特曼设法回归,似乎强调了这种问责制是一种幻觉。一个几乎在一夜之间把自己推到自我毁灭边缘的公司,如何能够被信任安全地引领一项许多人认为可能毁灭我们所有人的技术?

It’s not clear if Altman will have more power or less in his second stint as CEO. The company has established itself as the field’s front runner since the launch of ChatGPT, and expects to release new, more capable models next year. But there’s no guarantee OpenAI will maintain the industry lead as billions of dollars pour into frontier AI research by a growing field of competitors. The tech industry is known for its hype cycles—bursts of engineered excitement that allow venture capital to profit from fads like virtual reality or cryptocurrency. It’s possible the breakneck pace of AI development slows and the lofty promises about AGI don’t materialize.
目前尚不清楚奥特曼在第二次担任CEO时是否会拥有更多或更少的权力。自从ChatGPT推出以来,该公司已经确立了自己作为该领域的领跑者,并且预计明年将发布新的、更强大的模型。但没有保证OpenAI能够保持行业领先地位,因为数十亿美元正涌入由越来越多竞争者组成的前沿人工智能研究领域。科技行业以其炒作周期而闻名——这是一种工程化的兴奋爆发,让风险资本能够从虚拟现实或加密货币等时尚潮流中获利。人工智能发展的快速步伐可能会放缓,而关于通用人工智能的宏伟承诺可能不会实现。

But one of the big reasons for the standoff at OpenAI is that everyone involved thinks a new world is not just coming, but coming fast. Two people familiar with the board’s deliberations emphasize the stakes of supervising a company that believes it is building the most important technology in history. Altman thinks AGI—a system that surpasses humans in most regards—could be reached sometime in the next four or five years. AGI could turbocharge the global economy, expand the frontiers of scientific knowledge, and dramatically improve standards of living for billions of humans—creating a future that looks wildly different from the past. In this view, broadening our access to cognitive labor—“having more access to higher-quality intelligence and better ideas,” as Altman puts it—could help solve everything from climate change to cancer.
但OpenAI僵局的一个主要原因是,所有相关人士都认为一个新世界不仅即将到来,而且来势迅猛。两位了解董事会讨论的人士强调,监管一个认为自己正在构建史上最重要技术的公司的风险。奥特曼认为,AGI——一个在大多数方面超越人类的系统——可能在未来四到五年内实现。AGI可以加速全球经济增长,拓展科学知识的边界,并显著提高数十亿人的生活水平——创造一个与过去截然不同的未来。在这种观点中,扩大我们对认知劳动的获取——正如奥特曼所说的“获得更多高质量智能和更好的想法”——可以帮助解决从气候变化到癌症的一切问题。

Read More: The AI Arms Race Is Changing Everything
阅读更多:人工智能军备竞赛正在改变一切

But it would also come with serious risks. To many, the rapid rise in AI’s capabilities over the past year is deeply alarming. Computer scientists have not solved what’s known in the industry as the “alignment problem”—the task of ensuring that AGI conforms to human values. Few agree on who should determine those values. Altman and others have warned that advanced AI could pose “existential” risks on the scale of pandemics and nuclear war. This is the context in which OpenAI’s board determined that its CEO could not be trusted. “People are really starting to play for keeps now,” says Daniel Colson, executive director of the Artificial Intelligence Policy Institute (AIPI) and the founder of an Altman-backed startup, “because there’s an expectation that the window to try to shift the trajectory of things is closing.”
但这也会带来严重的风险。对许多人来说,人工智能能力在过去一年的迅速提升令人深感不安。计算机科学家尚未解决业内所称的“对齐问题”——确保通用人工智能符合人类价值观的任务。很少有人同意应由谁来确定这些价值观。奥特曼和其他人警告说,先进的人工智能可能会带来与大流行病和核战争同等级别的“存在性”风险。这是OpenAI董事会决定其首席执行官不可信任的背景。“现在人们真的开始认真对待了,”人工智能政策研究所(AIPI)执行董事兼奥特曼支持的初创公司创始人丹尼尔·科尔森说,“因为人们预期影响事物发展轨迹的窗口正在关闭。”

On a bright morning in early November, Altman looks nervous. We’re backstage at a cavernous event space in downtown San Francisco, where Altman will soon present to some 900 attendees at OpenAI’s first developer conference. Dressed in a gray sweater and brightly colored Adidas Lego sneakers, he thanks the speech coach helping him rehearse. “This is so not my thing,” he says. “I’m much more comfortable behind a computer screen.”
在十一月初一个明亮的早晨,阿特曼看起来很紧张。我们在旧金山市中心一个巨大的活动空间的后台,阿特曼不久将在OpenAI的首次开发者大会上向大约900名与会者发表演讲。他穿着灰色毛衣和色彩鲜艳的阿迪达斯乐高运动鞋,感谢帮他排练的演讲教练。“这真的不是我的强项,”他说。“我在电脑屏幕后面要舒服得多。”

That’s where Altman was to be found on Friday nights as a high school student, playing on an original Bondi Blue iMac. He grew up in a middle-class Jewish family in the suburbs of St. Louis, the eldest of four children born to a real estate broker and a dermatologist. Altman was equal parts nerdy and self-assured. He came out as gay as a teenager, giving a speech in front of his high school after some students objected to a National Coming Out Day speaker. He enrolled at Stanford to study computer science in 2003, as memories of the dot-com crash were fading. In college, Altman got into poker, which he credits for inculcating lessons about psychology and risk. By that point, he knew he wanted to become an entrepreneur. He dropped out of school after two years to work on Loopt, a location-based social network he co-founded with his then boyfriend, Nick Sivo.
高中时期的周五晚上,可以在原版邦迪蓝色iMac上找到奥特曼在玩耍的身影。他在圣路易斯郊区一个中产阶级犹太家庭长大,是一位房地产经纪人和一位皮肤科医生所生的四个孩子中的长子。奥特曼既书呆子气又自信。他在十几岁时出柜,此后在高中发表演讲,因为一些学生反对国家出柜日的演讲者。2003年,随着对互联网泡沫破灭记忆的淡化,他进入斯坦福大学学习计算机科学。在大学期间,奥特曼开始玩扑克牌,他认为这让他学到了关于心理学和风险的课程。到那时,他已经知道自己想成为一名企业家。两年后,他辍学,与当时的男友尼克·西沃共同创办了一家基于位置的社交网络Loopt。

Loopt became part of the first cohort of eight companies to join Y Combinator, the now vaunted startup accelerator. The company was sold in 2012 for $43 million, netting Altman $5 million. Though the return was relatively modest, Altman learned something formative: “The way to get things done is to just be really f-cking persistent,” he told Vox’s Re/code. Those who know him say Altman has an abiding sense of obligation to tackle issues big and small. “As soon as he’s aware of a problem, he really wants to solve it,” says his fiancé Mulherin, an Australian software engineer turned investor. Or as Altman puts it, “Stuff only gets better because people show up and work. No one else is coming to save the day. You’ve just got to do it.”
Loopt 成为了首批加入 Y Combinator 的八家公司之一,Y Combinator 现在是一个备受推崇的创业加速器。该公司在2012年以4300万美元的价格出售,Altman 净赚了500万美元。尽管回报相对适中,但Altman 学到了一些形成性的东西:“完成事情的方法就是要非常非常地坚持不懈,”他告诉 Vox 的 Re/code。那些了解他的人说,Altman 有一种持久的责任感去解决大大小小的问题。“一旦他意识到一个问题,他真的很想解决它,”他的未婚妻 Mulherin 说,她是一位澳大利亚的软件工程师转投资者。或者正如 Altman 所说:“事情只有因为人们的出现和工作才会变得更好。没有其他人会来拯救这一天。你只能自己去做。”

YC’s co-founder Paul Graham spotted a rare blend of strategic talent, ambition, and tenacity. “You could parachute him into an island full of cannibals and come back in five years and he’d be the king,” Graham wrote of Altman when he was just 23. In February 2014, Graham tapped his protégé, then 28, to replace him as president of YC. By the time Altman took the reins, YC had incubated unicorns like Airbnb, Stripe, and Dropbox. But the new boss had a bigger vision. He wanted to expand YC’s remit beyond software to “hard tech”—the startups where the technology might not even be possible, yet where successful innovation could unlock trillions of dollars and transform the world.
YC的联合创始人保罗·格雷厄姆发现了一种罕见的战略才能、雄心和坚韧的结合。格雷厄姆在谈到当时只有23岁的奥特曼时写道:“你可以把他空降到一个食人族的岛屿上,五年后回来,他会成为那里的国王。”2014年2月,格雷厄姆选择了他的门徒,当时28岁的奥特曼,来接替他成为YC的总裁。当奥特曼接过 reins 时,YC已经孵化出了像Airbnb、Stripe和Dropbox这样的独角兽公司。但新任老板有着更宏大的愿景。他想要将YC的业务范围扩展到软件之外的“硬科技”领域——那些技术可能尚未成熟,但成功的创新能够解锁数万亿美元并改变世界的初创公司。

Soon after becoming the leader of YC, Altman visited the headquarters of the nuclear-fusion startup Helion in Redmond, Wash. CEO David Kirtley recalls Altman showing up with a stack of physics textbooks and quizzing him about the design choices behind Helion’s prototype reactor. What shone through, Kirtley recalls, was Altman’s obsession with scalability. Assuming you could solve the scientific problem, how could you build enough reactors fast enough to meet the energy needs of the U.S.? What about the world? Helion was among the first hard-tech companies to join YC. Altman also wrote a personal check for $9.5 million and has since forked over an additional $375 million to Helion—his largest personal investment. “I think that’s the responsibility of capitalism,” Altman says. “You take big swings at things that are important to get done.”
成为YC领导人后不久,阿特曼就访问了位于华盛顿州雷德蒙德的核聚变初创公司Helion的总部。CEO大卫·柯特利回忆说,阿特曼带着一叠物理教科书来到这里,询问他关于Helion原型反应堆设计选择的问题。柯特利回忆说,阿特曼对可扩展性的迷恋显而易见。假设你能解决科学问题,你如何能够足够快地建造足够多的反应堆来满足美国的能源需求?世界的需求呢?Helion是最早加入YC的硬科技公司之一。阿特曼还亲自出资950万美元,并且此后又向Helion投资了额外的3.75亿美元——这是他个人最大的一笔投资。“我认为这是资本主义的责任,”阿特曼说,“你要在重要的事情上大胆尝试。”

Subscribe now and get the Person of the Year issue
立即订阅并获取年度人物特刊

Altman’s pursuit of fusion hints at the staggering scope of his ambition. He’s put $180 million into Retro Biosciences, a longevity startup hoping to add 10 healthy years to the human life-span. He conceived of and helped found Worldcoin, a biometric-identification system with a crypto-currency attached, which has raised hundreds of millions of dollars. Through OpenAI, Altman has spent $10 million seeding the longest-running study into universal basic income (UBI) anywhere in the U.S., which has distributed more than $40 million to 3,000 participants, and is set to deliver its first set of findings in 2024. Altman’s interest in UBI speaks to the economic dislocation that he expects AI to bring—though he says it’s not a “sufficient solution to the problem in any way.”
奥特曼对融合技术的追求显示了他雄心壮志的惊人范围。他向Retro Biosciences投资了1.8亿美元,这是一家致力于将人类寿命健康增加10年的长寿创业公司。他构想并帮助创立了Worldcoin,这是一个附带加密货币的生物识别系统,已经筹集了数亿美元。通过OpenAI,奥特曼投入了1000万美元资助美国迄今为止持续时间最长的全民基本收入(UBI)研究,该研究已向3000名参与者分发了超过4000万美元,并计划在2024年发布第一批研究结果。奥特曼对UBI的兴趣表明了他预期AI将带来的经济错位——尽管他说这绝不是“以任何方式对问题的充分解决方案”。

The entrepreneur was so alarmed at America’s direction under Donald Trump that in 2017 he explored running for governor of California. Today Altman downplays the endeavor as “a very lightweight consideration.” But Matt Krisiloff, a senior aide to Altman at the time, says they spent six months setting up focus groups across the state to help refine a political platform. “It wasn’t just a totally flippant idea,” Krisiloff says. Altman published a 10-point policy platform, which he dubbed the United Slate, with goals that included lowering housing costs, Medicare for All, tax reform, and ambitious clean-energy targets. He ultimately passed on a career switch. “It was so clear to me that I was much better suited to work on AI,” Altman says, “and that if we were able to succeed, it would be a much more interesting and impactful thing for me to do.”
这位企业家对唐纳德·特朗普领导下的美国方向感到非常担忧,以至于在2017年他考虑竞选加利福尼亚州州长。如今,阿特曼淡化了这一尝试,称其为“非常轻率的考虑”。但是,当时作为阿特曼高级助手的马特·克里西洛夫说,他们花了六个月的时间在全州设立焦点小组,以帮助完善政治平台。“这不仅仅是一个轻率的想法,”克里西洛夫说。阿特曼发布了一个包含10点政策的平台,他将其称为“联合名单”,其目标包括降低住房成本、全民医保、税收改革和雄心勃勃的清洁能源目标。他最终放弃了职业转换。“我非常清楚,我更适合从事人工智能工作,”阿特曼说,“如果我们能够成功,这对我来说将是一件更有趣且更有影响力的事情。”

But he remains keenly interested in politics. Altman’s beliefs are shaped by the theories of late 19th century political economist Henry George, who combined a belief in the power of market incentives to deliver increasing prosperity with a disdain for those who speculate on scarce assets, like land, instead of investing their capital in human progress. Altman has advocated for a land-value tax—a classic Georgist policy—in recent meetings with world leaders, he says.
但他仍然对政治保持着浓厚的兴趣。阿特曼的信仰受到了19世纪末政治经济学家亨利·乔治理论的影响,乔治结合了对市场激励能力带来持续繁荣的信念以及对那些投机稀缺资产如土地而不是将资本投资于人类进步的人的鄙视。阿特曼表示,在最近与世界领导人的会晤中,他倡导了一种土地价值税——这是一项经典的乔治主义政策。

Asked on a walk through OpenAI’s headquarters whether he has a vision of the future to help make sense of his various investments and interests, Altman says simply, “Abundance. That’s it.” The pursuits of fusion and superintelligence are cornerstones of the more equitable and prosperous future he envisions: “If we get abundant intelligence and abundant energy,” he says, “that will do more to help people than anything else I can possibly think of.”
在OpenAI总部散步时,当被问及他是否有一个未来愿景来帮助理解他的各种投资和兴趣时,阿特曼简单地说:“丰富。就是这样。”追求聚变和超级智能是他设想的更加公平繁荣未来的基石:“如果我们获得了丰富的智能和丰富的能源,”他说,“那将比我能想到的任何其他事情更有助于帮助人们。”

Altman began thinking seriously about AGI nearly a decade ago. At the time, “it was considered career suicide,” he says. But Altman struck up a running conversation with Elon Musk, who also felt smarter-than-human machines were not only inevitable, but also dangerous if they were built by corporations chasing profits. Both feared Google, which had bought Musk out when it acquired the top AI-research lab DeepMind in 2014, would remain the dominant player in the field. They imagined a nonprofit AI lab that could be an ethical counterweight, ensuring the technology benefited not just shareholders but also humanity as a whole.
阿特曼大约十年前开始认真思考通用人工智能(AGI)。他说,当时这被认为是职业自杀。但是阿特曼与埃隆·马斯克展开了持续的对话,马斯克也认为比人类更聪明的机器不仅是不可避免的,而且如果这些机器是由追求利润的公司建造的,那么它们也是危险的。两人都担心谷歌会继续在这一领域占据主导地位,因为谷歌在2014年收购了顶尖的人工智能研究实验室DeepMind时买下了马斯克的股份。他们想象着一个非营利的人工智能实验室,它可以成为一个道德的平衡力量,确保技术不仅惠及股东,而且惠及整个人类。

In the summer of 2015, Altman tracked down Ilya Sutskever, a star machine-learning researcher at Google Brain. The pair had dinner at the Counter, a burger bar near Google’s headquarters. As they parted ways, Altman got into his car and thought to himself, I have got to work with that guy. He and Musk spent nights and weekends courting talent. Altman drove to Berkeley to go for a walk with graduate student John Schulman; went to dinner with Stripe’s chief technology officer Greg Brockman; took a meeting with AI research scientist Wojciech Zaremba; and held a group dinner with Musk and others at the Rosewood hotel in Menlo Park, Calif., where the idea of what a new lab might look like began to take shape. “The montage is like the beginning of a movie,” Altman says, “where you’re trying to establish this ragtag crew of slight misfits to do something crazy.”
2015年夏天,Altman追踪到了谷歌大脑的明星机器学习研究员Ilya Sutskever。两人在谷歌总部附近的Counter汉堡酒吧共进晚餐。分别时,Altman上了车,心想,我得和这家伙合作。他和Musk花费夜晚和周末招募人才。Altman开车去伯克利与研究生John Schulman散步;与Stripe的首席技术官Greg Brockman共进晚餐;与人工智能研究科学家Wojciech Zaremba会面;并在加利福尼亚州门洛帕克的Rosewood酒店与Musk和其他人共进晚餐,新实验室的构想开始成形。“这就像电影开头的蒙太奇,”Altman说,“你试图建立一个由些微不合群的人组成的杂乱无章的团队去做一些疯狂的事情。”

OpenAI launched in December 2015. It had six co-founders—Altman, Musk, Sutskever, Brockman, Schulman, and Zaremba—and $1 billion in donations pledged by prominent investors like Reid Hoffman, Peter Thiel, and Jessica Livingston. During OpenAI’s early years, Altman remained YC president and was involved only from a distance. OpenAI had no CEO; Brockman and Sutskever were its de facto leaders. In an office in a converted luggage factory in San Francisco’s Mission district, Sutskever’s research team threw ideas at the wall to see what stuck. “It was a very brilliant assembly of some of the best people in the field,” says Krisiloff. “At the same time, it did not necessarily feel like everyone knew what they were doing.”
OpenAI于2015年12月成立。它有六位联合创始人——Altman、Musk、Sutskever、Brockman、Schulman和Zaremba——以及由像Reid Hoffman、Peter Thiel和Jessica Livingston这样的知名投资者承诺的10亿美元捐款。在OpenAI的早期,Altman仍然是YC的总裁,并且只是从远处参与。OpenAI没有CEO;Brockman和Sutskever是其实际领导者。在旧金山Mission区一个改造过的行李厂的办公室里,Sutskever的研究团队抛出想法,看看哪些能够成型。Krisiloff说:“这是一群非常杰出的,领域内一些最优秀人才的聚集。与此同时,它并不一定感觉像是每个人都知道自己在做什么。”

In 2018, OpenAI announced its charter: a set of values that codified its approach to building AGI in the interests of humanity. There was a tension at the heart of the document, between the belief in safety and the imperative for speed. “The fundamental belief motivating OpenAI is, inevitably this technology is going to exist, so we have to win the race to create it, to control the terms of its entry into society in a way that is positive,” says a former employee. “The safety mission requires that you win. If you don’t win, it doesn’t matter that you were good.” Altman disputes the idea that OpenAI needs to outpace rival labs to deliver on its mission, but says, “I think we care about a good AGI outcome more than others.”
2018年,OpenAI宣布了其宪章:一套凝练了其构建人类利益AGI方法的价值观。该文件的核心存在一种紧张关系,即安全信念与速度的紧迫性之间的冲突。“驱动OpenAI的基本信念是,这项技术终将存在,因此我们必须赢得创造它的竞赛,以一种积极的方式控制它进入社会的条件,”一位前员工说。“安全使命要求你必须赢。如果你没赢,那么你的善良也无关紧要。”Altman对于OpenAI需要超越竞争对手实验室以实现其使命的观点表示异议,但他说,“我认为我们比其他人更关心一个良好的AGI结果。”

One key to winning was Sutskever. OpenAI’s chief scientist had an almost religious belief in the neural network, a type of AI algorithm that ingested large amounts of data and could independently detect underlying patterns. He believed these networks, though primitive at the time, could lead down a path toward AGI. “Concepts, patterns, ideas, events, they are somehow smeared through the data in a complicated way,” Sutskever told TIME in August. “So to predict what comes next, the neural network needs to somehow become aware of those concepts and how they leave a trace. And in this process, these concepts come to life.”
获胜的关键之一是Sutskever。OpenAI的首席科学家几乎对神经网络有一种宗教般的信仰,这是一种AI算法,能够吸收大量数据并独立检测出潜在的模式。他相信,尽管当时这些网络还很原始,但它们可以引领我们走向通用人工智能(AGI)的道路。“概念、模式、想法、事件,它们以某种复杂的方式散布在数据中,”Sutskever在8月份对时代杂志说。“因此,为了预测接下来会发生什么,神经网络需要以某种方式意识到这些概念以及它们如何留下痕迹。在这个过程中,这些概念变得鲜活起来。”

Read More: How We Chose the TIME100 Most Influential People in AI
阅读更多:我们如何选择《时代》杂志百大人工智能领域最具影响力人物

To commit to Sutskever’s method and the charter’s mission, OpenAI needed vast amounts of computing power. For this it also needed cash. By 2019, OpenAI had collected only $130 million of the original $1 billion committed. Musk had walked away from the organization—and a planned donation of his own—after a failed attempt to insert himself as CEO. Altman, still running YC at the time, was trying to shore up OpenAI’s finances. He initially doubted any private investor could pump cash into the project at the volume and pace it required. He assumed the U.S. government, with its history of funding the Apollo program and the Manhattan Project, would be the best option. After a series of discussions—“you try every door,” Altman says—he was surprised to find “the chances of that happening were exactly zero.” He came to believe “the market is just going to have to do it all the way through.”
为了致力于Sutskever的方法和章程的使命,OpenAI需要大量的计算能力。为此,它还需要现金。到2019年,OpenAI只筹集到了最初承诺的10亿美元中的1.3亿美元。在尝试插手成为CEO失败后,马斯克已经离开了该组织——以及他自己计划的捐款。当时仍在运营YC的Altman正在努力稳固OpenAI的财务状况。他最初怀疑任何私人投资者是否能以项目所需的速度和规模注入现金。他认为,鉴于美国政府资助阿波罗计划和曼哈顿计划的历史,它将是最佳选择。经过一系列讨论——Altman说:“你尝试每一扇门”——他惊讶地发现“那发生的可能性完全是零”。他开始相信“市场将不得不一路做到底。”

Wary of the perverse incentives that could arise if investors gained sway over the development of AGI, Altman and the leadership team debated different structures and landed on an unusual one. OpenAI would establish a “capped profit” subsidiary that could raise funds from investors, but would be governed by a nonprofit board. OpenAI’s earliest investors signed paperwork indicating they could receive returns of up to 100 times their investment, with any sums above that flowing to the nonprofit. The company’s founding ethos—a research lab unshackled from commercial considerations—had lasted less than four years.
考虑到如果投资者对人工通用智能(AGI)的发展产生影响可能带来的不良激励,奥特曼和领导团队讨论了不同的结构,并最终确定了一个不同寻常的模式。OpenAI将成立一个“有限盈利”的子公司,该公司可以从投资者那里筹集资金,但将由一个非营利组织的董事会管理。OpenAI的最早期投资者签署了文件,表明他们可以获得最高达到投资额100倍的回报,任何超出该数额的资金将流向非营利组织。公司的创始理念——一个不受商业考虑束缚的研究实验室——不到四年就结束了。

Altman was spending an increasing amount of time thinking about OpenAI’s financial troubles and hanging out at its office, where Brockman and Sutskever had been lobbying him to come on full time. “OpenAI had never had a CEO,” he says. “I was kind of doing it 30% of the time, but not very well.” He worried the lab was at an inflection point, and without proper leadership, “it could just disintegrate.” In March 2019, the same week the company’s restructure was announced, Altman left YC and formally came on as OpenAI CEO.
阿特曼花越来越多的时间思考OpenAI的财务问题,并在其办公室里闲逛,布罗克曼和苏茨克弗一直在游说他全职加入。“OpenAI从未有过CEO,”他说。“我大概有30%的时间在做这个工作,但做得不太好。”他担心实验室正处于一个转折点,如果没有适当的领导,“它可能就会瓦解。”2019年3月,就在公司重组宣布的同一周,阿特曼离开了YC,并正式成为OpenAI的CEO。

Altman insists this new structure was “the least bad idea” under discussion. In some ways, the solution was an elegant one: it allowed the company to raise much-needed cash from investors while telegraphing its commitment to conscientiously developing AI. Altman embodied both goals—an extraordinarily talented fundraiser who was also a thoughtful steward of a potentially transformative technology.
奥特曼坚称这种新结构是讨论中“最不坏的主意”。在某些方面,这个解决方案是优雅的:它让公司能够从投资者那里筹集到急需的资金,同时传达出其对于认真发展人工智能的承诺。奥特曼身兼两者——既是一位非凡才能的筹资人,也是一位对潜在改变游戏规则的技术负有深思熟虑的管理者。

It didn’t take long for Altman to raise $1 billion from Microsoft—a figure that has now ballooned to $13 billion. The restructuring of the company, and the tie-up with Microsoft, changed OpenAI’s complexion in significant ways, three former employees say. Employees began receiving equity as a standard part of their compensation packages, which some holdovers from the nonprofit era thought created incentives for employees to maximize the company’s valuation. The amount of equity that staff were given was very generous by industry standards, according to a person familiar with the compensation program. Some employees fretted OpenAI was turning into something more closely resembling a traditional tech company. “We leave billion-dollar ideas on the table constantly,” says VP of people Diane Yoon.
奥特曼从微软筹集了10亿美元的时间并不长,这个数字现在已经膨胀到130亿美元。三位前员工表示,公司的重组以及与微软的合作,显著改变了OpenAI的面貌。员工开始将股权作为薪酬套餐的标准部分,而来自非营利时代的一些老员工认为这会激励员工最大化公司的估值。一位熟悉薪酬计划的人士表示,员工获得的股权按行业标准来看非常慷慨。一些员工担心OpenAI正变得更像传统科技公司。“我们经常放弃价值十亿美元的想法,”人力资源副总裁黛安娜·尹说。

Buy a print of the Person of the Year covers now
立即购买年度人物封面的印刷品

Microsoft’s investment supercharged OpenAI’s ability to scale up its systems. An innovation from Google offered another breakthrough. Known as the “transformer,” it made neural networks far more efficient at spotting patterns in data. OpenAI researchers began to train the first models in their GPT (generative pre-trained transformer) series. With each iteration, the models improved dramatically. GPT-1, trained on the text of some 7,000 books, could just about string sentences together. GPT-2, trained on 8 million web pages, could just about answer questions. GPT-3, trained on hundreds of billions of words from the internet, books, and Wikipedia, could just about write poetry.
微软的投资极大地加速了OpenAI扩展其系统的能力。谷歌的一项创新提供了另一个突破。这项被称为“变压器”的技术,使得神经网络在数据模式识别方面变得更加高效。OpenAI的研究人员开始训练他们的GPT(生成预训练变压器)系列中的第一个模型。随着每一次迭代,模型的性能都有了显著的提升。GPT-1,训练了大约7000本书的文本,勉强能够串联句子。GPT-2,训练了800万个网页,勉强能够回答问题。GPT-3,训练了来自互联网、书籍和维基百科的数千亿单词,勉强能够写诗。

Altman recalls a breakthrough in 2019 that revealed the vast possibilities ahead. An experiment into “scaling laws” underpinning the relationship between the computing power devoted to training an AI and its resulting capabilities yielded a series of “perfect, smooth graphs,” he says—the kind of exponential curves that more closely resembled a fundamental law of the universe than experimental data. It was a cool June night, and in the twilight a collective realization dawned on the assembled group of researchers as they stood outside the OpenAI office: AGI was not just possible, but probably coming sooner than any of them previously thought. “We were all like, this is really going to happen, isn’t it?” Altman says. “It felt like one of these moments of science history. We know a new thing now, and we’re about to tell humanity about it.”
奥特曼回忆起2019年的一个突破,揭示了前方巨大的可能性。他说,一项关于支撑人工智能训练所投入的计算能力与其最终能力之间关系的“规模定律”的实验产生了一系列“完美、平滑的图表”——这种指数曲线更像是宇宙的基本法则,而不是实验数据。那是一个凉爽的六月夜晚,在黄昏时分,当研究人员们站在OpenAI办公室外时,他们集体意识到了一件事:通用人工智能(AGI)不仅有可能实现,而且可能比他们之前认为的来得更快。“我们都在想,这真的要发生了,不是吗?”奥特曼说。“感觉像是科学史上的那些时刻之一。我们现在知道了一个新事物,我们即将告诉人类。”

The realization contributed to a change in how OpenAI released its technology. By then, the company had already reneged on its founding principle of openness, after recognizing that open-sourcing increasingly powerful AI could be great for criminals and bad for business. When it built GPT-2 in 2019, it initially declined to release the model publicly, fearing it could have a devastating impact on public discourse. But in 2020, the company decided to slowly distribute its tools to wider and wider numbers of people. The doctrine was called “iterative deployment.” It enabled OpenAI to collect data on how AIs were used by the public, and to build better safety mechanisms in response. And it would gradually expose the public to the technology while it was still comparatively crude, giving people time to adapt to the monumental changes Altman saw coming.
这一认识促使OpenAI改变了其技术发布方式。那时,公司已经背离了其开放原则的创立宗旨,因为它认识到开源越来越强大的人工智能可能对犯罪分子有利,对商业不利。在2019年构建GPT-2时,它最初拒绝公开发布该模型,担心这可能对公共话语产生毁灭性的影响。但到了2020年,公司决定逐步将其工具分发给越来越多的人。这种做法被称为“迭代部署”。它使OpenAI能够收集关于公众如何使用AI的数据,并据此构建更好的安全机制。同时,它会逐渐让公众在技术还相对粗糙时接触到它,给人们时间适应Altman所看到的巨大变化。

On its own terms, iterative deployment worked. It handed OpenAI a decisive advantage in safety-trained models, and eventually woke up the world to the power of AI. It’s also true that it was extremely good for business. The approach bears a striking resemblance to a tried-and-tested YC strategy for startup success: building the so-called minimum viable product. Hack together a cool demo, attract a small group of users who love it, and improve based on their feedback. Put things out into the world. And eventually—if you’re lucky enough and do it right—that will attract large groups of users, light the fuse of a media hype cycle, and allow you to raise huge sums. This was part of the motivation, Brockman tells TIME. “We knew that we needed to be able to raise additional capital,” he says. “Building a product is actually a pretty clear way to do it.”
按照自身的条件,迭代部署取得了成功。它为OpenAI在安全训练模型方面提供了决定性的优势,并最终唤醒了世界对人工智能力量的认识。同样确实的是,这对商业来说极为有利。这种方法与YC策略中用于创业成功的经过验证的方法极为相似:构建所谓的最小可行产品。拼凑出一个很酷的演示,吸引一小群喜欢它的用户,并根据他们的反馈进行改进。将产品推向世界。如果你足够幸运并且做得正确,最终这将吸引大批用户,点燃媒体炒作周期的导火索,并让你能够筹集巨额资金。布洛克曼在接受《时代》杂志采访时说,这是动机的一部分。“我们知道我们需要能够筹集额外的资本,”他说。“构建一个产品实际上是一个相当清晰的方式。”

Some worried that iterative deployment would accelerate a dangerous AI arms race, and that commercial concerns were clouding OpenAI’s safety priorities. Several people close to the company thought OpenAI was drifting away from its original mission. “We had multiple board conversations about it, and huge numbers of internal conversations,” Altman says. But the decision was made. In 2021, seven staffers who disagreed quit to start a rival lab called Anthropic, led by Dario Amodei, OpenAI’s top safety researcher.
有些人担心迭代部署会加速危险的人工智能军备竞赛,商业关切正在遮蔽OpenAI的安全优先事项。一些接近公司的人认为OpenAI正在偏离其最初的使命。“我们就此进行了多次董事会讨论,以及大量的内部讨论,”奥特曼说。但决定已经做出。2021年,七名不同意的员工辞职,以Dario Amodei为首成立了一个名为Anthropic的竞争实验室,Amodei是OpenAI的顶级安全研究员。

In August 2022, OpenAI finished work on GPT-4, and executives discussed releasing it along with a basic, user-friendly chat interface. Altman thought that would “be too much of a bombshell all at once.” He proposed launching the chatbot with GPT-3.5—a model that had been accessible to the public since the spring—so people could get used to it, and then releasing GPT-4 a few months later. Decisions at the company typically involve a long, deliberative period during which senior leaders come to a consensus, Altman says. Not so with the launch of what would eventually become the fastest-growing new product in tech history. “In this case,” he recalls, “I sent a Slack message saying, Yeah, let’s do this.” In a brainstorming session before Nov. 30 launch, Altman replaced its working title, Chat With GPT-3.5, with the slightly pithier ChatGPT. OpenAI’s head of sales received a Slack message letting her know the product team was silently launching a “low-key research preview,” which was unlikely to affect the sales team.
2022年8月,OpenAI完成了GPT-4的开发工作,高管们讨论是否应该连同一个基础的、用户友好的聊天界面一起发布。Altman认为这样做“一次性会引起太大的轰动”。他建议先用GPT-3.5发布聊天机器人——这个模型自春季以来已经对公众开放——这样人们可以先适应它,然后几个月后再发布GPT-4。Altman说,在公司,决策通常需要一个漫长、深思熟虑的过程,高级领导们达成共识。但推出最终成为科技史上增长最快的新产品并非如此。“在这种情况下,”他回忆说,“我发了一条Slack消息说,嗯,我们这么做吧。”在11月30日发布前的一个头脑风暴会议上,Altman将其工作标题从Chat With GPT-3.5更改为更简洁的ChatGPT。OpenAI的销售负责人收到一条Slack消息,告知她产品团队正在悄悄推出一个“低调的研究预览版”,这不太可能影响销售团队。

Nobody at OpenAI predicted what came next. After five days, ChatGPT crossed 1 million users. ChatGPT now has 100 million users—a threshold that took Facebook 41⁄2 years to hit. Suddenly, OpenAI was the hottest startup in Silicon Valley. In 2022, OpenAI brought in $28 million in revenue; this year it raked in $100 million a month. The company embarked on a hiring spree, more than doubling in size. In March, it followed through on Altman’s plan to release GPT-4. The new model far surpassed ChatGPT’s capabilities—unlike its predecessor, it could describe the contents of an image, write mostly reliable code in all major programming languages, and ace standardized tests. Billions of dollars poured into competitors’ efforts to replicate OpenAI’s successes. “We definitely accelerated the race, for lack of a more nuanced phrase,” Altman says.
OpenAI没有预料到接下来发生的事情。仅仅五天后,ChatGPT的用户就突破了100万。现在,ChatGPT拥有1亿用户——Facebook用了4年半的时间才达到这个门槛。突然间,OpenAI成为了硅谷最热门的初创公司。2022年,OpenAI的收入为2800万美元;今年它的月收入飙升至1亿美元。公司开始大规模招聘,员工数量增加了一倍多。3月份,它按照奥特曼的计划发布了GPT-4。这个新模型远远超越了ChatGPT的能力——与前代产品不同,它能描述图片内容,能够在所有主要编程语言中编写大体可靠的代码,并且能够轻松通过标准化考试。数十亿美元涌入竞争对手的努力中,试图复制OpenAI的成功。“我们肯定加速了这场竞赛,用一个不那么细腻的词来说,”奥特曼说。

The CEO was suddenly a global star. He seemed unusually equipped to navigate the different factions of the AI world. “I think if this technology goes wrong, it can go quite wrong, and we want to be vocal about that,” Altman told lawmakers at a U.S. Senate hearing in May. That month, Altman embarked on a world tour, including stops in Israel, India, Japan, Nigeria, South Korea, and the UAE. Altman addressed a conference in Beijing via video link. So many government officials and policy-makers clamored for an audience that “we ended up doing twice as many meetings than were scheduled for any given day,” says head of global affairs Anna Makanju. AI soared up the policy agenda: there was a White House Executive Order, a global AI Safety Summit in the U.K., and attempts to codify AI standards in the U.N., the G-7, and the African Union.
这位首席执行官突然成为了全球明星。他似乎异常擅长在人工智能世界的不同派系之间导航。奥特曼在五月份的美国参议院听证会上告诉立法者:“我认为如果这项技术出了问题,可能会出大问题,我们想对此发出响亮的声音。”那个月,奥特曼开始了一次世界巡回,包括在以色列、印度、日本、尼日利亚、韩国和阿联酋的停留。奥特曼通过视频链接在北京的一个会议上发表了演讲。全球事务负责人安娜·马坎朱说:“政府官员和政策制定者争相会见他,以至于我们最终进行的会议是任何给定日程的两倍。”人工智能在政策议程上急剧上升:有白宫的行政命令、在英国的全球人工智能安全峰会,以及在联合国、七国集团和非洲联盟中尝试制定人工智能标准。

By the time Altman took the stage at OpenAI’s developer conference in November, it seemed as if nothing could bring him down. To cheers, he announced OpenAI was moving toward a future of autonomous AI “agents” with power to act in the world on a user’s behalf. During an interview with TIME two days later, he said he believed the chances of AI wiping out humanity were not only low, but had gone down in the past year. He felt the increase in awareness of the risks, and an apparent willingness among governments to coordinate, were positive developments that flowed from OpenAI’s iterative-deployment strategy. While the world debates the probabilities that AI will destroy civilization, Altman is more sanguine. (The odds are “nonzero,” he allows, but “low if we can take all the right actions.”) What keeps him up at night these days is something far more prosaic: an urban coyote that has colonized the grounds of his $27 million home in San Francisco. “This coyote moved into my house and scratches on the door outside,” he says, picking up his iPhone and, with a couple of taps, flipping the screen around to reveal a picture of the animal lounging on an outdoor sofa. “It’s very cute, but it’s very annoying at night.”
到11月OpenAI开发者大会Altman登台时,似乎没有什么能够击败他。在欢呼声中,他宣布OpenAI正朝着一个自主AI“代理”的未来迈进,这些代理有能力代表用户在世界上采取行动。两天后,在接受《时代》杂志采访时,他说他相信AI消灭人类的可能性不仅很低,而且在过去一年中已经降低了。他感到对风险的认识增加,以及政府之间明显愿意协调,是源于OpenAI迭代部署策略的积极发展。当世界在辩论AI毁灭文明的可能性时,Altman更为乐观。(他承认几率是“非零”的,但如果我们能采取所有正确的行动,“几率会很低”。)如今让他夜不能寐的是一件更加平常的事情:一只定居在他位于旧金山价值2700万美元家中的城市郊狼。“这只郊狼搬进了我的房子,在外面的门上刮擦,”他说着,拿起iPhone,轻轻几下,就把屏幕翻转过来,显示出一张动物悠闲地躺在户外沙发上的照片。 它非常可爱,但晚上真的很烦人。"

As Altman radiated confidence, unease was growing within his board of directors. The board had shrunk from nine members to six over the preceding months. That left a panel made up of three OpenAI employees—Altman, Sutskever, and Brockman—and three independent directors: Adam D’Angelo, the CEO of question-and-answer site Quora; Tasha McCauley, a technology entrepreneur and Rand Corp. scientist; and Helen Toner, an expert in AI policy at Georgetown University’s Center for Security and Emerging Technology.
随着奥特曼散发出自信,他的董事会内部却开始感到不安。在过去的几个月里,董事会成员从九人减少到了六人。这导致董事会由三名OpenAI员工——奥特曼、苏茨克弗和布罗克曼——以及三名独立董事组成:问答网站Quora的首席执行官亚当·丹吉洛;科技企业家兼兰德公司科学家塔莎·麦考利;以及乔治敦大学安全与新兴技术中心的人工智能政策专家海伦·托纳。

The panel had argued over how to replace the three departing members, according to three people familiar with the discussions. For some time—little by little, at different rates—the three independent directors and Sutskever were becoming concerned about Altman’s behavior. Altman had a tendency to play different people off one another in order to get his desired outcome, say two people familiar with the board’s discussions. Both also say Altman tried to ensure information flowed through him. “He has a way of keeping the picture somewhat fragmented,” one says, making it hard to know where others stood. To some extent, this is par for the course in business, but this person says Altman crossed certain thresholds that made it increasingly difficult for the board to oversee the company and hold him accountable.
根据三位了解讨论情况的人士,小组就如何替换三名即将离职成员进行了争论。一段时间以来,三名独立董事和Sutskever逐渐对Altman的行为感到担忧。据两位了解董事会讨论的人士称,Altman有一种倾向,喜欢利用不同的人相互对抗,以达到他想要的结果。他们俩还说,Altman试图确保信息流通过他。一位说:“他有一种方式,让整个画面有些支离破碎”,这使得很难知道其他人的立场。在某种程度上,这在商业中是常有的事,但这位人士说,Altman越过了某些界限,使得董事会越来越难以监督公司并对他负责。

One example came in late October, when an academic paper Toner wrote in her capacity at Georgetown was published. Altman saw it as critical of OpenAI’s safety efforts and sought to push Toner off the board. Altman told one board member that another believed Toner ought to be removed immediately, which was not true, according to two people familiar with the discussions.
一个例子发生在10月下旬,当时Toner在乔治敦大学的职务上撰写的一篇学术论文被发表了。Altman认为这篇论文批评了OpenAI的安全工作,并试图将Toner赶出董事会。根据两位了解讨论情况的人士,Altman告诉一位董事会成员,另一位成员认为应该立即将Toner移除,但这实际上并不属实。

This episode did not spur the board’s decision to fire Altman, those people say, but it was representative of the ways in which he tried to undermine good governance, and was one of several incidents that convinced the quartet that they could not carry out their duty of supervising OpenAI’s mission if they could not trust Altman. Once the directors reached the decision, they felt it was necessary to act fast, worried Altman would detect that something was amiss and begin marshaling support or trying to undermine their credibility. “As soon as he had an inkling that this might be remotely on the table,” another of the people familiar with the board’s discussions says, “he would bring the full force of his skills and abilities to bear.”
这一事件并没有促使董事会决定解雇奥特曼,那些人说,但它代表了他试图破坏良好治理的方式,也是几起事件之一,让这四人确信如果他们不能信任奥特曼,就无法履行监督OpenAI使命的职责。一旦董事们做出了决定,他们觉得有必要迅速行动,担心奥特曼会察觉到有什么不对劲,开始集结支持或试图破坏他们的信誉。一位熟悉董事会讨论的人说:“只要他隐约感觉到这可能在考虑之中,他就会发挥出他全部的技能和能力。”

On the evening of Thursday, Nov. 16, Sutskever asked Altman to chat at noon the following day. At the appointed time, Altman joined Sutskever on Google Meet, where the entire board was present except Brockman. Sutskever told Altman that he was being fired and that the news would be made public shortly. “It really felt like a weird dream, much more intensely than I would have expected,” Altman tells TIME.
在11月16日星期四的晚上,Sutskever邀请Altman第二天中午聊天。到了约定的时间,Altman在Google Meet上加入了Sutskever,除了Brockman,整个董事会都在场。Sutskever告诉Altman他被解雇了,这个消息不久后会公之于众。“这真的感觉像一个奇怪的梦,比我预期的要强烈得多,”Altman告诉《时代》杂志。

The board’s statement was terse: Altman “was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities,” the announcement said. “The board no longer has confidence in his ability to continue leading OpenAI.”
董事会的声明简洁明了:声明称,阿特曼“在与董事会的沟通中并不总是坦率,这阻碍了董事会行使其职责。”“董事会不再对他继续领导OpenAI的能力抱有信心。”

Altman was locked out of his computer. He began reaching out to his network of investors and mentors, telling them he planned to start a new company. (He tells TIME he received so many texts his iMessage broke.) The board expected pressure from investors and media. But they misjudged the scale of the blowback from within the company, in part because they had reason to believe the executive team would respond differently, according to two people familiar with the board’s thinking, who say the board’s move to oust Altman was informed by senior OpenAI leaders, who had approached them with a variety of concerns about Altman’s behavior and its effect on the company’s culture.
奥特曼被锁在了自己的电脑外面。他开始联系自己的投资者和导师网络,告诉他们他计划创办一家新公司。(他告诉《时代》杂志,他收到的短信太多,导致他的iMessage崩溃了。)董事会预计会受到投资者和媒体的压力。但他们低估了公司内部的强烈反弹,部分原因是他们有理由相信高管团队会有不同的反应,两位熟悉董事会想法的人士表示,董事会决定罢免奥特曼是在得到OpenAI高级领导人的信息后作出的,这些领导人向他们提出了对奥特曼行为及其对公司文化影响的各种担忧。

Legal and confidentiality reasons have made it difficult for the board to share specifics, the people with knowledge of the proceedings say. But the absence of examples of the “lack of candor” the board cited as the impetus for Altman’s firing contributed to rampant speculation—that the decision was driven by a personal vendetta, an ideological dispute, or perhaps sheer incompetence. The board fired Altman for “nitpicky, unfireable, not even close to fireable offenses,” says Ron Conway, the founder of SVAngel and a mentor who was one of the first people Altman called after being terminated. “It is reckless and irresponsible for a board to fire a founder over emotional reasons.”
知情人士表示,出于法律和保密原因,董事会很难分享具体细节。但董事会在解雇阿特曼时所提到的“缺乏坦诚”缺乏实例,这导致了猜测的猖獗——决定是出于个人恩怨、意识形态争议,还是纯粹的无能。SVAngel的创始人、同时也是阿特曼被解雇后第一个接到他电话的导师罗恩·康威说:“董事会因为一些挑剔的、不应解雇的、甚至远非解雇理由的小过失就解雇一位创始人,这是鲁莽和不负责任的。”

Within hours, the company’s staff threatened to quit if the board did not resign and allow Altman to return. Under immense pressure, the board reached out to Altman the morning after his firing to discuss a potential path forward. Altman characterizes it as a request for him to come back. “I went through a range of emotions. I first was defiant,” he says. “But then, pretty quickly, there was a sense of duty and obligation, and wanting to preserve this thing I cared about so much.” The sources close to the board describe the outreach differently, casting it as an attempt to talk through ways to stabilize the company before it fell apart.
几小时内,如果董事会不辞职并允许Altman回归,公司员工威胁要辞职。在巨大的压力下,董事会在Altman被解雇后的第二天早上联系他,讨论可能的前进道路。Altman将其描述为他们请求他回来。“我经历了一系列情绪。起初我是抗拒的,”他说。“但是,很快,就有了一种责任和义务感,以及想要保护这个我如此在乎的东西。”接近董事会的消息人士对这次接触的描述则不同,他们将其描述为一种尝试,通过谈话来稳定公司,防止它崩溃。

For nearly 48 hours, the negotiations dragged on. Mira Murati, OpenAI’s chief technology officer who stepped in as interim CEO, joined the rest of the company’s leadership in advocating for Altman’s return. So on the night of Sunday, Nov. 19, the board appointed a new interim CEO, Emmett Shear, the former CEO of Twitch. Microsoft boss Satya Nadella announced Altman and Brockman would be joining Microsoft to start a new advanced AI unit; Microsoft made it known that any OpenAI staff members would be welcome to join. After a tearful confrontation with Brockman’s wife, Sutskever flipped his position: “I deeply regret my participation in the board’s actions,” he posted in the early hours of Nov. 20.
在将近48小时的谈判后,OpenAI的首席技术官Mira Murati临时担任首席执行官,加入公司其他领导层一起倡导Altman回归。因此,在11月19日星期日晚上,董事会任命了一位新的临时首席执行官,Twitch的前首席执行官Emmett Shear。微软老板萨提亚·纳德拉宣布Altman和Brockman将加入微软,开启一个新的高级AI部门;微软明确表示欢迎任何OpenAI的员工加入。在与Brockman的妻子发生了一场激动人心的对峙后,Sutskever改变了他的立场:“我深感遗憾参与了董事会的行动,”他在11月20日凌晨发帖说。

By the end of that day, nearly all of OpenAI’s 770 employees had signed an open letter signaling their intention to quit if Altman was not reinstated. The same canniness that makes Altman such a talented entrepreneur also made him a formidable opponent in the standoff, able to command loyalty from huge swaths of the company and beyond.
到那天结束时,OpenAI的770名员工几乎全部签署了一封公开信,表明如果Altman不被重新任命,他们打算辞职。Altman作为一名才华横溢的企业家所具有的机智同样使他在对峙中成为一个强大的对手,能够从公司内部到外部指挥起大片的忠诚。

And while mission is a powerful draw for OpenAI employees, so too is money. Nearly every full-time OpenAI employee has financial interests in OpenAI’s success, including former board members Brockman and Sutskever. (Altman, who draws a salary of $65,000, does not have equity beyond an indirect investment through his stake in YC.) A tender offer to let OpenAI employees sell shares at an $86 billion valuation to outside investors was planned for a week after Altman’s firing; employees who stood to earn millions by December feared that option would vanish. “It’s unprecedented in history to see a company go potentially to zero if everybody walks,” says one of the people familiar with the board’s discussions. “It’s unsurprising that employees banded together in the face of that particular threat.”
尽管使命对OpenAI的员工来说是一个强大的吸引力,金钱同样如此。几乎每一位全职OpenAI员工都对OpenAI的成功拥有经济利益,包括前董事会成员布罗克曼和苏茨克弗。 (阿尔特曼领取65000美元的薪水,除了通过他在YC中的股份间接投资外,没有其他股权。)在阿尔特曼被解雇后一周,计划让OpenAI员工以860亿美元的估值向外部投资者出售股份;那些到12月可能赚取数百万美元的员工担心这个选择会消失。“如果每个人都走了,看到一家公司可能归零,在历史上是前所未有的,”一位熟悉董事会讨论的人说。“面对那种特殊威胁,员工团结起来也就不足为奇了。”

Unlike the staff, the three remaining board members who sought to oust Altman were employed elsewhere, had no financial stake in the company, and were not involved in its day-to-day operations. In contrast to a typical for-profit board, which makes decisions informed by quarterly earnings reports, stock prices, and concerns for shareholder value, their job was to exercise their judgment to ensure the company was acting in the best interests of humanity—a mission that is fuzzy at best, and difficult to uphold when so much money is at stake. But whether or not the board made a correct decision, their unwillingness or inability to offer examples of what they saw as Altman’s problematic behavior would ensure they lost the public relations battle in a landslide. A panel set up as a check on the CEO’s power had come to seem as though it was wielding unaccountable power of its own.
与员工不同,三名试图罢免奥特曼的董事会成员在其他地方就职,没有公司的财务利益,也不参与公司的日常运营。与典型的以盈利为目的的董事会不同,后者的决策是基于季度收益报告、股价和对股东价值的关注,他们的工作是运用自己的判断力确保公司行为符合人类最佳利益——这一使命本身就模糊不清,在涉及大量金钱时更难以维护。但无论董事会是否做出了正确的决定,他们不愿或无法提供他们认为奥特曼行为有问题的例子,这将确保他们在公关战中彻底失利。作为对CEO权力的一种制衡而设立的小组,似乎变成了自己行使不受问责的权力。

In the end, the remaining board members secured a few concessions in the agreement struck to return Altman as CEO. A new independent board would supervise an investigation into his conduct and the board’s decision to fire him. Altman and Brockman would not regain their seats, and D’Angelo would remain on the panel, rather than all independent members resigning. Still, it was a triumph for OpenAI’s leadership. “The best interests of the company and the mission always come first. It is clear that there were real misunderstandings between me and members of the board,” Altman posted on X. “I welcome the board’s independent review of all recent events.”
最终,剩余的董事会成员在重新任命奥特曼为首席执行官的协议中争取到了一些让步。一个新的独立董事会将监督对他的行为以及董事会解雇他的决定的调查。奥特曼和布罗克曼将不会重新获得他们的席位,而丹吉洛将继续留在小组中,而不是所有独立成员辞职。尽管如此,这对OpenAI的领导层来说仍是一次胜利。“公司和使命的最佳利益始终是第一位的。很明显,我和董事会成员之间存在真正的误解,”奥特曼在X上发布。“我欢迎董事会对所有最近事件的独立审查。”

Two nights before Thanksgiving, staff gathered at the headquarters, popping champagne. Brockman posted a selfie with dozens of employees, with the caption: “we are so back.”
感恩节前两晚,员工们聚集在总部,开香槟庆祝。布罗克曼发了一张自拍,照片中有几十名员工,配文:“我们又回来了。”

Ten days after the agreement was reached for their return, OpenAI’s leaders were resolute. “I think everyone feels like we have a second chance here to really achieve the mission. Everyone is aligned,” Brockman says. But the company is in for an overhaul. Sutskever’s future at the company is murky. The new board—former Twitter board chair Bret Taylor, former U.S. Treasury Secretary Larry Summers, and D’Angelo—will expand back to nine members and take a hard look at the company’s governance. “Clearly the current thing was not good,” Altman says.
在达成他们回归的协议十天后,OpenAI的领导层态度坚定。“我认为每个人都觉得我们有了第二次机会来真正实现使命。每个人都是一致的,”布罗克曼说。但公司将面临一次大改组。苏茨克弗在公司的未来前景不明朗。新董事会——前Twitter董事会主席布雷特·泰勒,前美国财政部长拉里·萨默斯,和D'Angelo——将扩大至九名成员,并将仔细审视公司的治理。“显然当前的情况不好,”奥特曼说。

OpenAI had tried a structure that would provide independent oversight, only to see it fall short. “One thing that has very clearly come out of this is we haven’t done a good job of solving for AI governance,” says Divya Siddarth, the co-founder of the Collective Intelligence Project, a nonprofit that works on that issue. “It has put into sharp relief that very few people are making extremely consequential decisions in a completely opaque way, which feels fine, until it blows up.”
OpenAI尝试了一种能够提供独立监督的结构,但最终未能达到预期。集体智能项目联合创始人Divya Siddarth表示:“一个非常明确的结论是,我们在解决人工智能治理问题上做得不好。这突显出极少数人以一种完全不透明的方式做出极其重大的决策,这看似没问题,直到一切失控。”集体智能项目是一个致力于该问题的非营利组织。

Back in the CEO’s chair, Altman says his priorities are stabilizing the company and its relationships with external partners after the debacle; doubling down on certain research areas after the massive expansion of the past year; and supporting the new board to come up with better governance. What that looks like remains vague. “If an oracle said, Here is the way to set up the structure that is best for humanity, that’d be great,” Altman says.
回到首席执行官的位置上,阿特曼表示,他的首要任务是在灾难之后稳定公司及其与外部合作伙伴的关系;在过去一年大规模扩张之后,加倍投入某些研究领域;以及支持新董事会提出更好的治理方案。这看起来还很模糊。“如果有个神谕说,这里是为人类设立最佳结构的方式,那就太好了,”阿特曼说。

Whatever role he plays going forward will receive more scrutiny. “I think these events have turned him into a political actor in the mass public’s eye in a way that he wasn’t before,” says Colson, the executive director of AIPI, who believes the episode has highlighted the danger of having risk-tolerant technologists making choices that affect all of us. “Unfortunately, that’s the dynamic that the market has set up for.”
无论他未来扮演什么角色都将受到更多审查。“我认为这些事件已经以一种他之前未曾有过的方式,将他变成了大众视野中的政治行为者,”AIPI的执行董事科尔森说,他认为这一事件凸显了让愿意承担风险的技术人员做出影响我们所有人的决策所带来的危险。“不幸的是,这正是市场设定的动态。”

But for now, Altman looks set to remain a leading architect of a potentially world-changing technology. “Building superintelligence is going to be a society-wide project,” he says. “We would like to be one of the shapers, but it’s not going to be something that one company just does. It will be far bigger than any one company. And I think we’re in a position where we’re gonna get to provide that input no matter what at this point. Unless we really screw up badly.” —With reporting by Will Henshall/Washington and Julia Zorthian/New York
但就目前而言,阿特曼似乎将继续作为一项可能改变世界的技术的主要建筑师。他说:“构建超级智能将是一个全社会的项目。我们希望成为塑造者之一,但这不会是一家公司就能做到的事情。它将远远超出任何一家公司的范畴。我认为我们现在的位置,无论如何都将提供那种输入。除非我们真的糟糕透顶。”——华盛顿的Will Henshall和纽约的Julia Zorthian报道。