打造 Anthropic 首席执行官有人称他为“悲观论者”,也有人视他为负责任的技术专家。以下是达里奥·阿莫代伊关于人工智能的思考是如何形成的。
The Making of Anthropic’s C.E.O. Some call him a doomer. Others see him as a responsible technologist. Here’s how Dario Amodei’s thinking on artificial intelligence was shaped.
2007 年的一个晚上,达里奥·阿莫代伊在普林斯顿大学向来访的朋友科林·赫克曼滔滔不绝地谈论《奇点临近》——计算机科学家雷·库兹韦尔 2005 年出版的一部著作。
One night in 2007, Dario Amodei regaled a friend who was visiting him at Princeton University with his thoughts on “The Singularity Is Near,” a 2005 treatise by the computer scientist Ray Kurzweil.
阿莫代伊博士刚读完这本书,书中库兹韦尔预测人工智能最终将在被称为“奇点”的时刻超越人脑。阿莫代伊博士一边用手指拨弄头发,一边告诉赫克曼,这一里程碑可能在他们有生之年实现。
Dr. Amodei had just read the book, in which Mr. Kurzweil predicted that artificial intelligence would eventually surpass the human brain in a moment known as the Singularity. As he twisted his hair with his fingers, Dr. Amodei told his friend, Colin Heckman, that the milestone could happen within their lifetimes.
“他变得着迷了,”从旧金山高中起就与阿莫代伊博士相识的赫克曼说。“他说人工智能必须正确构建,否则我们就完了。”当时 24 岁、攻读物理学博士学位的阿莫代伊博士正处于大量吸收信息的时期。他沉迷于“加速未来”和“克服偏见”这两个鲜为人知的博客,那里的作者探讨人工智能消灭人类的潜在可能。他还深入研究了 GiveWell 的在线论坛,该组织旨在最大化慈善捐赠的效果,知识分子们在那里辩论如何为慈善事业赚取最多资金。
“He became obsessed,” said Mr. Heckman, who went to high school with Dr. Amodei in San Francisco. “He said A.I. had to be built right, or we’re going to be done.”Dr. Amodei, then 24 and a doctoral student in physics, was in a period of soaking up information. He had become absorbed in the obscure blogs “Accelerating Future” and “Overcoming Bias,” where writers discussed A.I.’s potential to wipe out humanity. And he had delved into the online forums of GiveWell, an organization that aimed to maximize the effectiveness of charitable giving, where intellectuals debated how to make the most money for philanthropy.
据对他的二十多位同学、朋友、老师和同事的采访,这些书籍、博客和在线社区至今仍在影响阿莫代伊博士。它们有助于解释驱动这位 43 岁 Anthropic 首席执行官的思想——该公司将于本月进行全球最大规模的公开上市之一——并让人理解这位处于人工智能革命前沿的人物究竟如何看待人工智能。
These books, blogs and online communities have influenced Dr. Amodei to this day, according to interviews with more than two dozen of his classmates, friends, teachers and colleagues. They help explain the ideas that drive the 43-year-old chief executive of Anthropic, a company that is set to conduct one of the world’s largest public listings this month, and offer an understanding of how the man at the forefront of the A.I. revolution actually thinks about A.I.
据他的朋友和同事称,阿莫代伊博士世界观的核心,是两种交织在一起的哲学,它们共享一个不妥协的信念:数学推理能够改变人类的未来。
At the heart of Dr. Amodei’s worldview, his friends and colleagues said, are two interwoven philosophies that share an uncompromising belief that mathematical reasoning can change the future of humanity.
一种是“理性主义”,旨在将冷静的思考应用于世界。这一观点由埃利泽·尤德考斯基等人倡导,后者是一位自学成才的人工智能研究员。理性主义认为,人工智能将变得足够强大,足以毁灭社会。另一种是“有效利他主义”,它与理性主义同步演变,寻求用数学般的精确度实现社会利益最大化。
One is “rationalism,” which aims to apply dispassionate thinking to the world. Espoused by people like Eliezer Yudkowsky, a self-taught A.I. researcher, it argues that A.I. will become powerful enough to destroy society. The other is “effective altruism,” which evolved alongside rationalism and seeks to maximize social good with mathematical precision.
对于阿莫代伊博士而言,这两种思想流派融合成了一种坚定不移的信念:人工智能可能极其危险,因此只有他和志同道合的技术专家才有资格构建它。他的朋友和同事表示,这种信念催化了他明确提出的努力——即在不关心人类安全的人之前创造出人工智能,并激发了他想要控制这项技术的愿望。
These schools of thought fused for Dr. Amodei into an unyielding conviction that A.I. could be so dangerous that only he and like-minded technologists could be entrusted with building it. That catalyzed his stated efforts to create A.I. before people without concerns about the safety of humanity did, and spurred a desire to control the technology, his friends and colleagues said.
阿莫代伊博士在9月的一篇文章中写道:“只有当我们以正确的方式构建这项技术时,其益处才能实现——只要我们善用争取来的时间,付出异常审慎的努力来确保万无一失就是值得的。”该文呼吁放缓该技术的发展。
With A.I., “the benefits will only be achieved if we build the technology in the right way, and — so long as we use the time we gain well — it is worth taking unusually deliberate care to get it right,” Dr. Amodei wrote in a September essay that called for a slowdown in the technology’s development.
这些思想构成了Anthropic的基石。阿莫代伊博士于2021年与妹妹丹尼拉·阿莫代伊及其他五人共同创立了该公司,这些人曾在理性主义和有效利他主义社群中活动。随着公司的成长,他们的观点——阿莫代伊博士一直大声疾呼且孜孜不倦地推动这些观点——变得日益突出。
These ideas formed the foundation for Anthropic, which Dr. Amodei established in 2021 with his sister, Daniela Amodei, and five others who spent time around rationalist and effective altruist communities. And as the company sprouted, their views — which Dr. Amodei has pushed vocally and relentlessly — have become increasingly prominent.
阿莫代伊博士在地缘政治辩论中表态,称美国应限制向中国出售芯片,以限制该国在人工智能领域的竞争。他宣称应对人工智能进行监管,以防止技术失控。在与国防部的争论中,他宣称该技术不应用于大规模监视美国人,也不应部署在自主武器系统中。
Dr. Amodei has weighed in on geopolitical debates and said the United States should restrict chip sales to China to limit that nation from competing in A.I. He has proclaimed that A.I. should be regulated to stop the technology from going rogue. And in disputes with the Department of Defense, he has declared that the technology should not be used for the mass surveillance of Americans or deployed in autonomous weapons.
“他确实意识到,这里摆着巨大的生存风险,”Nathan Avery 说。他曾与 Amodei 博士在高中时一起参加越野跑,多年来曾有过许多深度交谈,尽管两人后来疏远了。“他需要让自己处于一个能够对此采取行动的位置。”Amodei 博士是一位亿万富翁,随着 Anthropic 上市,其净资产有望增长。他通过公司代表拒绝接受采访。
“He actually sees that there’s a huge existential risk on the table here,” said Nathan Avery, who ran high school cross-country with Dr. Amodei and had many deep conversations with him over the years, though they have drifted apart. “And he needs to put himself in a position to do something about it.”Dr. Amodei, a billionaire whose net worth is set to grow with Anthropic’s public offering, declined through a company representative to be interviewed.
他的想法赢得了支持者,尤其是一些从事 AI 安全工作的人,他们表示 Amodei 博士正在负责任地开发 AI。这也让他在硅谷和华盛顿树敌不少,批评者——包括 Meta 的 Mark Zuckerberg 和 Nvidia 的 Jensen Huang——指责他是“厄运论者”,并试图将 AI 权力集中在自己手中。9 月,美国国防部在社交媒体上发文:“美国主义,而非有效利他主义。”
His thinking has won him supporters, especially among some who work in A.I. safety and who say Dr. Amodei is developing A.I. responsibly. It has also made him enemies in Silicon Valley and Washington, with critics — including Meta’s Mark Zuckerberg and Nvidia’s Jensen Huang — attacking him for being a “doomer” and trying to centralize power in A.I. In September, the Defense Department posted on social media: “Americanism, not effective altruism.”
Amodei 博士似乎意识到自己的观点激怒了他人,曾试图公开将自己的信念与理性主义和有效利他主义区分开来。2023 年,他在《纽约时报》的“Hard Fork”播客节目中表示,他不认为自己是有效利他主义者。
Dr. Amodei, who appears to be aware that his ideas have riled others, has tried to publicly differentiate his beliefs from rationalism and effective altruism. In 2023, he told The New York Times’s “Hard Fork” podcast that he did not consider himself to be an effective altruist.
“我同情其中许多观点,”他说,但反对“将人们划分为不同部落”。“Dario,别那么严肃!”
“I’m sympathetic to a number of the ideas,” he said, but against “dividing people up into tribes.”‘Dario, Lighten Up!’
Amodei 博士和他年幼的妹妹 Daniela 在旧金山 Mission 区长大。他们的父亲 Riccardo Amodei 是一名皮革工匠,母亲 Elena Engel 负责管理公共图书馆翻新工程。根据离婚档案,这对夫妇于 1991 年分居。
Dr. Amodei and his younger sister Daniela grew up in San Francisco’s Mission District. Their father, Riccardo Amodei, was a leather craftsman, and their mother, Elena Engel, managed public library renovations. The couple separated in 1991, according to divorce filings.
在 Lawton Alternative School 读八年级时,Amodei 博士就已脱颖而出。1997 年,他和一位同学被评为“最有可能成功”,并手持美元钞票出现在毕业纪念册照片中。
By eighth grade at Lawton Alternative School, Dr. Amodei stood out. In 1997, he and a peer were voted “Most Likely to Succeed” and were pictured in their yearbook holding dollar bills.
在旧金山顶尖公立高中洛厄尔高中,阿莫代伊是一位贪婪的学习者,曾在加州大学伯克利分校修读课程。同学们说,他思考过功利主义等观念——主张人们应为最多的生命体最大化幸福——并担忧千年虫问题及其可能毁掉全球计算机的潜在风险。
At Lowell High School, San Francisco’s top public high school, Dr. Amodei was a voracious learner who took classes at the University of California, Berkeley. He pondered ideas like utilitarianism, which argues that people should maximize happiness for the most living beings, and worried about issues like the Y2K bug and its potential to ruin the world’s computers, classmates said.
当时担任洛厄尔高中物理教师的理查德·夏皮罗说,阿莫代伊经常认为自己考试表现不佳。“我会对他大喊:‘达里奥,别太较真!’”夏皮罗说。
Richard Shapiro, then a physics teacher at Lowell, said Dr. Amodei had frequently thought he underperformed on tests. “I’d yell at him, ‘Dario, lighten up!’”Mr. Shapiro said.
2001年,阿莫代伊入读位于帕萨迪纳的加州理工学院,但常返回旧金山探望朋友和家人。一次行程中,他与艾弗里和赫克曼去了一家寿司餐厅,三人在那里辩论功利主义。阿莫代伊是素食主义者,却辩称吃螃蟹没问题,因为它们是无脊椎动物。
In 2001, Dr. Amodei enrolled at the California Institute of Technology in Pasadena, but often returned to San Francisco to see friends and family. On one trip, he went with Mr. Avery and Mr. Heckman to a sushi restaurant, where they debated utilitarianism. Dr. Amodei was a vegetarian but argued it was fine to eat crabs because they are invertebrates.
用餐时,赫克曼提出一个思想实验:如果有人开始尽可能多地杀鸡,什么时候杀掉那个人以防止更多鸡被杀才说得通?阿莫代伊笑了笑,然后说“在功利主义意义上,会有一定数量的鸡,其价值超过一个人类”,艾弗里回忆道。
During the meal, Mr. Heckman presented a thought experiment: If someone started killing as many chickens as possible, when would it make sense to kill that person to prevent more chickens from dying? Dr. Amodei chuckled, then said that “there was some number of chickens that would outweigh a human in the utilitarian sense,” Mr. Avery recalled.
阿莫代伊自称无神论者,朋友们说他信念坚定。2003年,他开车载着艾弗里等人前往旧金山,参加抗议美国入侵伊拉克的示威活动。
Mr. Amodei, who has said he is an atheist, had strong convictions, his friends said. In 2003, he drove Mr. Avery and others to protests in San Francisco against the U.S. invasion of Iraq.
次年,阿莫代伊转学到位于硅谷核心地带的斯坦福大学。这一搬迁让他离妹妹阿莫代伊更近了,妹妹当时在附近上高中。
The next year, Dr. Amodei transferred to Stanford University, in the heart of Silicon Valley. The move brought him closer to Ms. Amodei, who was in high school nearby.
“我们一直想一起拯救世界,”他在2023年的一档播客节目中谈到妹妹时说。时间有限2006年从斯坦福大学毕业后,阿莫迪博士进入普林斯顿大学攻读物理学博士学位,最终将研究重点放在将物理学应用于生物学和计算神经科学上。那一年,患有丙型肝炎的父亲去世。阿莫迪博士全身心投入视网膜细胞研究,并对开发加速科学进步的技术产生了兴趣。
“We always wanted to save the world together,” he said of his sister on a 2023 podcast. The Finiteness of Time After graduating from Stanford in 2006, Dr. Amodei joined a physics doctoral program at Princeton, eventually focusing on applying physics to biology and computational neuroscience. That year, his father, who suffered from hepatitis C, died. Dr. Amodei threw himself into researching retina cells and became interested in developing technology to accelerate scientific progress.
曾与阿莫迪博士交往并共同发起一个科学讨论小组的王洁德(Jade Wang)表示,他们经常探讨:毕生致力于研究单一疾病的科学家,若转而为整个领域开发技术或方法,影响力会更大。她透露,阿莫迪博士以父亲为例——父亲在一种能让大多数丙肝病例可治疗的药物问世前去世。
Jade Wang, who dated Dr. Amodei and started a science discussion group with him, said they had often talked about how scientists who dedicated their lives to studying one disease would have a greater impact developing technology or methods for their entire field. He cited his father, who died before the discovery of a drug that made most hepatitis C cases treatable, she said.
那时,阿莫迪博士已深受库兹韦尔先生和尤德考斯基先生著作的影响,后者是理性主义者的领军人物。他对库兹韦尔先生的观点尤感兴趣,即先进技术将让人类超越自然极限生存,这种信念被称为“超人类主义”。
By then, Dr. Amodei was embracing the writings of Mr. Kurzweil and Mr. Yudkowsky, a leader of the rationalists. He was especially intrigued by Mr. Kurzweil’s idea that advanced technologies would let humans exist beyond natural limits, a belief known as transhumanism.
2007年赫克曼先生访问普林斯顿期间,阿莫迪博士告诉他,奇点将在2040年前到来,人类未来或许能将思想与机器融合。据赫克曼先生回忆,阿莫迪博士说:“随着时间的推移,我们将成为新的超级智能AI,或AI-人类混合体。”
During Mr. Heckman’s 2007 visit to Princeton, Dr. Amodei told him that the Singularity would happen by 2040 and that humans might someday meld their minds with machines. “We would over time become the new superintelligent A.I. or A.I.-human hybrid,” Dr. Amodei said, according to Mr. Heckman.
阿莫迪博士深深着迷于这种思维,甚至鼓励朋友和妹妹参加尤德考斯基先生“奇点研究所”的年度会议(该组织现称“机器智能研究所”)。据《纽约时报》看到的一封2008年发给朋友的电子邮件,阿莫迪博士写道:“尤德考斯基是‘我遇到过的最深刻的思想家之一’,且‘明确致力于带来一个造福人类的奇点’。”
Dr. Amodei became so fascinated by this thinking that he encouraged his friends and sister to attend the annual conference of Mr. Yudkowsky’s Singularity Institute. (The group is now called the Machine Intelligence Research Institute.) Mr. Yudkowsky was “one of the deepest thinkers I have ever encountered” and “explicitly trying to bring about a singularity that will be beneficial to humanity,” Dr. Amodei told friends in a 2008 email seen by The Times.
在给《纽约时报》的一封电子邮件中,尤德科斯基表示,阿莫代伊“目前既不寻求也不采纳我的建议,我不记得有过他这样做的任何时候”。大约在那时,阿莫代伊也开始与埃隆·马斯克通信,后者创立了火箭公司SpaceX。两位知情人士透露,阿莫代伊曾告诉朋友,他钦佩这位科技大亨,并曾考虑为他工作,但他从未在该太空初创公司任职。
In an email to The Times, Mr. Yudkowsky said Dr. Amodei “does not currently ask or take my advice and I do not remember any time when he did.”Around then, Dr. Amodei also began corresponding with Elon Musk, who had founded the rocket company SpaceX. Dr. Amodei told friends that he admired the tech mogul and had considered working for him, but he never did at the space start-up, two people with knowledge of the men’s communications said.
2008年,阿莫代伊在阅读经济学家泰勒·考恩的博客时了解到GiveWell,考恩在博文中写道,该组织在慈善方面提供了“有趣的建议”。阿莫代伊开始活跃于GiveWell的电子邮件列表和论坛,在那里他讨论了有效利他主义在慈善和科学领域的应用。
In 2008, Dr. Amodei came across GiveWell while reading the blog of the economist Tyler Cowen, who posted that the organization offered “interesting tips” on charities. Dr. Amodei became active on GiveWell’s email lists and forums, where he discussed an effective altruist approach to philanthropy and science.
据《纽约时报》看到的一封电子邮件显示,2010年,他写道,自己“正在考虑从事金融职业,这将使我能够捐出比现在更多的钱”。
In 2010, he wrote that he was “considering a career in finance, which would allow me to give away more money than I currently do,” according to an email seen by The Times.
那一年,阿莫代伊在GiveWell的一篇博文中探讨了一个捐赠1万美元的决定。他在“终结结核病伙伴关系”(Stop TB Partnership)——该组织专注于预防成人结核病——和“村庄触达”(VillageReach)——该组织帮助非洲村庄的儿童接种疫苗——之间做选择。
In a blog post for GiveWell that year, Dr. Amodei discussed a decision to give away $10,000. He was choosing to donate to either the Stop TB Partnership, which focused on preventing adult tuberculosis, or VillageReach, which helped vaccinate children in African villages.
阿莫代伊写道,“终结结核病伙伴关系”在“成本效益”方面更优,因为它服务于成年人,而“成年人比婴儿能够拥有更深刻、更有意义的体验”。他补充道:“我认为成年人的死亡可能比婴儿的死亡严重2到3倍。”
The Stop TB Partnership, Dr. Amodei wrote, was superior in “cost-effectiveness” because it served adults and “adults are capable of deeper and more meaningful experiences” than infants. “I think an adult death is perhaps 2 or 3 times worse than an infant’s death,” he added.
在考虑了这些组织执行任务的方式后,阿莫代伊最终改变了主意,认为“村庄触达”更有效。
Dr. Amodei ultimately changed his mind after considering how the groups executed their missions and deeming VillageReach to be more effective.
2011年,他在斯坦福大学开始博士后工作,研究癌细胞。朋友们说,在个人生活中,他尝试服用抗衰老补充剂,保持严格的游泳计划,并对热量限制产生兴趣,一些研究表明这能延长实验小鼠的寿命。
In 2011, he started postdoctoral work at Stanford studying cancer cells. In his personal life, he experimented with anti-aging supplements, maintained a strict swimming regimen and became interested in caloric reduction, which was shown in some studies to prolong the lives of lab mice, his friends said.
“他经常谈论时间的有限性,以及时间是唯一不可再生的资源,”艾弗里说,并补充道,阿莫代伊“对待时间极其无情”。两位了解他的人士透露,2014年确诊患有睾丸癌,这让这种特质更加突出。阿莫代伊接受了化疗并完全康复。
“He would often talk about the finiteness of time and how it was the one nonrenewable resource,” Mr. Avery said, adding that Dr. Amodei was “ruthless with his time.”That was compounded by a diagnosis for testicular cancer in 2014, two people who know him said. Dr. Amodei underwent chemotherapy and made a full recovery.
到那时,他已与霍尔登·卡诺夫斯基关系密切,后者是GiveWell的创始人之一。GiveWell已开始与Good Ventures合作,后者是Facebook联合创始人达斯汀·莫斯科维茨及其妻子卡里·图纳的慈善机构,两人资助有效利他主义事业。阿莫代伊开始为这项合作项目GiveWell Labs提供生命科学资助方面的建议。
By then, he had become close to Holden Karnofsky, a GiveWell founder. GiveWell had started working with Good Ventures, the philanthropy of the Facebook co-founder Dustin Moskovitz and his wife, Cari Tuna, who finance effective altruist causes. Dr. Amodei began advising the partnership, GiveWell Labs, on life sciences funding.
当时,卡诺夫斯基先生正与阿莫代伊女士交往。这对伴侣搬进了旧金山的一所合租屋,最终住进的人员包括阿莫代伊博士;理论物理学家贾里德·卡普兰;以及大学辍学生克里斯托弗·奥拉,后者曾接受过阿莫代伊博士的数学辅导。
At the time, Mr. Karnofsky was dating Ms. Amodei. The couple moved into a group house in San Francisco, which eventually included Dr. Amodei; Jared Kaplan, a theoretical physicist; and Christopher Olah, a college dropout who had been tutored in math by Dr. Amodei.
三位曾到访该住所的人士表示,室友们讨论尤德考斯基的文章,探讨理性主义和有效利他主义。他们还参加了“火人节”,这是在内华达沙漠举行的年度反主流文化聚会,有时会装扮成动物造型。
The housemates discussed Mr. Yudkowsky’s essays and explored rationalism and effective altruism, three people who visited the home said. They also attended Burning Man, the annual counterculture gathering in the Nevada desert, sometimes dressing up in animal costumes.
卡普兰博士和奥拉先生后来与阿莫代伊博士和阿莫代伊女士共同创立了Anthropic。2025年,卡诺夫斯基先生也开始为该公司工作。
Dr. Kaplan and Mr. Olah later co-founded Anthropic with Dr. Amodei and Ms. Amodei. In 2025, Mr. Karnofsky also started working for the company. An A.I.
人工智能复兴 2010年代初,机器学习的进步——即计算机算法通过分析数据自主学习执行任务——促使科技公司加大对人工智能的投资。
Revival In the early 2010s, advancements in machine learning — computer algorithms that learn to perform tasks on their own by analyzing data — led to new investments by tech companies in A.I.
投身其中的一家公司是中国互联网巨头百度,它在加利福尼亚州桑尼维尔设立实验室,并从谷歌挖角人工智能先驱吴恩达担任首席科学家。阿莫代伊于2014年加入百度,因经常在办公室过夜和直言不讳而迅速为人所知。
One company that jumped in was the Chinese internet giant Baidu, which opened a lab in Sunnyvale, Calif., and hired an A.I. pioneer, Andrew Ng, from Google as chief scientist. Dr. Amodei joined Baidu in 2014 and quickly became known for staying overnight at the office and sharing his opinions.
“达里奥几乎在每一个机会都会激烈地争论,”曾与阿莫代伊在百度共事的研究员格雷戈里·迪亚莫斯说。
“Dario argued very aggressively at pretty much every opportunity,” said Gregory Diamos, a researcher who worked with Dr. Amodei at Baidu.
即使在人工智能发展的早期,阿莫代伊博士就被这项技术的潜在危险所震惊。迪亚莫斯博士回忆说,他在2015年告诉同事,人工智能很快就能达到或超越人类大脑的能力,因此开发它的公司不应回避其出错的可能性。
Even in the early days of this A.I. period, Dr. Amodei was struck by the technology’s potential dangers. He told colleagues in 2015 that A.I. could soon match or exceed the abilities of the human brain, so companies developing it should not shy away from the possibility of its going awry, Dr. Diamos recalled.
同年,阿莫代伊博士加入了一个名为Google Brain的谷歌人工智能实验室,专注于人工智能安全研究。他和其他五位研究人员——包括他的室友奥拉先生——于2016年发表了一篇题为《人工智能安全的具体问题》的论文,阐述了如果开发不当,该技术可能会如何出现异常行为。
That same year, Dr. Amodei joined a Google A.I. lab called Google Brain, where he focused on A.I. safety. He and five other researchers — including his housemate Mr. Olah — published a 2016 paper, “Concrete Problems in A.I. Safety,” which laid out how the technology could misbehave if it was wrongly developed.
2016年,阿莫代伊博士加入了一个新的人工智能项目:一家名为OpenAI的非营利组织。他领导其安全团队和另一个研究“规模定律”的团队,即人工智能模型随着训练数据的增加而变得更强大这一观点。
In 2016, Dr. Amodei joined a new A.I. effort: a nonprofit called OpenAI. He led its safety team and another team that studied “scaling laws,” or the idea that A.I. models become more capable as they are trained with more data.
打造OpenAI 据七位曾与他共事的人士透露,在OpenAI,阿莫代伊博士关于开发人工智能的正确方式的信念变得更加坚定。他更公开地谈论人工智能的力量和潜在危害,并采取行动将他的思维注入该技术的开发中。
Building OpenAI At OpenAI, Dr. Amodei’s beliefs on the correct way to develop A.I. hardened, seven people who worked with him said. He became more vocal about A.I.’s power and potential harms and acted to infuse the technology’s development with his thinking.
例如,2016年11月唐纳德·J·特朗普当选总统三天后,阿莫代伊博士在OpenAI的内部通讯应用中写道,他担心美国正滑向威权主义。据《纽约时报》看到的信息显示,他说他“不愿意在一个威权国家从事军事强力技术的研究”。
Three days after Donald J. Trump was elected president in November 2016, for instance, Dr. Amodei wrote in OpenAI’s internal messaging app that he was concerned the United States was sliding toward authoritarianism. He said he was “not comfortable conducting research on militarily powerful technologies in an authoritarian country,” according to messages seen by The Times.
根据内部消息,阿莫代伊博士还写道,投资人彼得·蒂尔向特朗普竞选团队捐款125万美元,因该支持行为“在道德上值得谴责”,OpenAI不应再接受他的资金。(蒂尔创办的风投公司Founders Fund后来投资了Anthropic。)阿莫代伊博士还帮助促成了OpenAI与认同其理念的资助方之间的交易。据三位知情人士透露,2017年,他将OpenAI介绍给了卡诺夫斯基运营的慈善机构,阿莫代伊博士曾担任该机构的技术顾问。
Dr. Amodei also wrote that the investor Peter Thiel, who had donated $1.25 million to Mr. Trump’s campaign, was “morally worthy of condemnation” for that support and that OpenAI should no longer accept his money, according to internal messages. (Founders Fund, a venture capital firm started by Mr. Thiel, later invested in Anthropic.) Dr. Amodei also helped arrange a deal between OpenAI and funders who aligned with his thinking. In 2017, he connected OpenAI to the philanthropy run by Mr. Karnofsky, for which Dr. Amodei was a technical adviser, three people familiar with the arrangement said.
卡诺夫斯基的组织(现称为Coefficient Giving)向OpenAI捐赠了3000万美元。卡诺夫斯基也加入了OpenAI董事会。
Mr. Karnofsky’s organization, now known as Coefficient Giving, donated $30 million to OpenAI. Mr. Karnofsky also joined OpenAI’s board.
同年晚些时候,卡诺夫斯基与阿莫代伊女士结婚。据两人婚礼网站介绍,宾客收到了毛绒玩具,并因安排“不同角色之间最浪漫的邂逅”而获得奖励。
Later that year, Mr. Karnofsky and Ms. Amodei married. At their wedding, guests received stuffed animals and were rewarded for arranging “the most romantic hookups between the different characters,” according to the couple’s wedding website.
阿莫代伊博士戴着熊猫头套出席婚礼,晚宴的着装要求是“滑稽正装”。《纽约时报》看到的一张照片显示,他身着装扮,与戴着长颈鹿头套的奥拉,以及阿莫代伊博士当时交往的女友卡米·克拉克站在一起。他与克拉克女士于2022年结婚。
Dr. Amodei wore a panda head to the wedding, where the dress code for the reception was “silly formal.”In a photo viewed by The Times, he appeared in costume alongside Mr. Olah, who wore a giraffe head, and Cami Clark, a woman Dr. Amodei was dating. He and Ms. Clark married in 2022.
2018年,丹尼拉·阿莫代伊加入OpenAI,最终成为安全与政策副总裁。一些员工很快对这对兄妹日益增长的影响力提出质疑。
In 2018, Daniela Amodei joined OpenAI, eventually becoming vice president of safety and policy. Some employees soon raised questions about the siblings’ growing influence.
据三位知情人士透露,那年12月,OpenAI研究员约书亚·阿奇姆告诉首席执行官萨姆·奥特曼,任命阿莫代伊博士为首席伦理官将是不明智之举。阿奇姆援引阿莫代伊兄妹权力的增大,表示AI监督应由持有不同观点的人共同承担。
That December, Joshua Achiam, an OpenAI researcher, told Sam Altman, the chief executive, that it would be unwise to name Dr. Amodei the company’s chief ethics officer, three people with knowledge of the discussions said. Mr. Achiam cited the Amodeis’ increasing power and said A.I. oversight should be shared by people with different perspectives.
当阿奇姆坚持应聘请外部专家担任首席伦理官时,阿莫代伊博士与他发生对峙。知情人士称,阿莫代伊博士提高音量,表示没有人比他更适合该职位。
When Mr. Achiam insisted that an outside expert be named chief ethics officer, Dr. Amodei confronted him. Dr. Amodei raised his voice and said no one was better suited for the position than him, the people said.
2018年,阿莫代伊博士和阿莫代伊女士接手了公司的一个名为GPT的项目,即生成式预训练转换器的简称。据六名曾在该公司工作的人士透露,他们探索了一种安全技术,旨在教会人工智能在人类指令下执行任务。
In 2018, Dr. Amodei and Ms. Amodei took over a company project called GPT, short for generative pre-trained transformer. They explored a safety technique that aimed to teach A.I. to perform tasks with human instruction, six people who worked at the company said.
这种方法——将人工智能推向潜在危险的边缘,然后用安全技术对其进行测试——后来定义了阿莫代伊博士的职业生涯。
This approach — pushing A.I. closer to potential dangers and then testing it with safety techniques — would come to define Dr. Amodei’s career.
OpenAI总裁格雷格·布罗克曼曾试图协助GPT项目。但据两名知情人士透露,阿莫代伊博士和阿莫代伊女士予以抵制。知情人士称,阿莫代伊博士曾多次告诉奥尔特曼,如果不让布罗克曼远离该项工作,他将离职。
Greg Brockman, OpenAI’s president, tried to help with GPT. But Dr. Amodei and Ms. Amodei resisted, two people with knowledge of the company said. Dr. Amodei repeatedly told Mr. Altman that if he did not keep Mr. Brockman away from the work, he would leave, the people said. For Mr. Altman, the ultimatums went too far, a person with knowledge of his thinking said.
据一名了解奥尔特曼想法的人士透露,对奥尔特曼而言,这种最后通牒走得太远了。他拒绝了阿莫代伊夫妇的要求。
He refused the Amodeis’ demands. In December 2020, Dr. Amodei, Ms. Amodei and other colleagues quit OpenAI. Four weeks later, they started Anthropic.
2020年12月,阿莫代伊博士、阿莫代伊女士及其他同事离开了OpenAI。四周后,他们创立了Anthropic。
Ryan Mac is a Times reporter who covers corporate accountability across the global technology industry. Kalley Huang is a Times reporter in San Francisco, covering Apple and the technology industry.
凯利·黄是驻旧金山的《纽约时报》记者,负责报道苹果和科技行业。
Cade Metz is a Times reporter who writes about artificial intelligence, driverless cars, robotics, virtual reality and other emerging areas of technology.
乔治亚·吉是新闻研究员,负责管理《纽约时报》新闻线索热线。
Georgia Gee is a news researcher and runs The New York Times tips line.