Abstract:
“The singularity has arrived.” On October 17, during the annual meeting of the International Monetary Fund (IMF) and the World Bank in Bangkok, Anthropic co-founder Jack Clark stepped onto the podium of the Per Jacobson Lecture and threw out this judgment. He described a future to the economists and policymakers in the audience: a human scientist can lead hundreds or thousands of AI scientists to conduct research at the same time. Today there are only about 10 million scientific researchers in the world, but in the future, there may be one billion virtual scientists.
Clark has always been quite certain about the future of AI. As early as ten years ago, he bet his own future on it.
At that time, he was a technology reporter at Bloomberg.
On August 2, 2016, Clark told his colleague Joe Weisenthal that he had left Bloomberg.
Next, he plans to spend a few months studying AI carefully, and then leave the journalism industry to do something related to AI.
He also added: "AI is more important than anything else, so I want to prioritize this."
In September this year, 37-year-old Clark appeared on the Forbes 400 list of the richest Americans, with a net worth of US$15.5 billion, equivalent to more than 100 billion yuan.
Someone recently reviewed Clark's experience on social media. After seeing it, Weisenthal dug out the chat history between the two ten years ago and only sighed:
"I should have been more careful back then."
A reporter who is more obsessed with AI than a scientist
Clark is actually a liberal arts student. He was born in Brighton, England, and graduated from the University of East Anglia in 2009, where he studied English Literature and Creative Writing. After graduation, he entered the technology journalism industry and worked successively at the British technology media "The Register" and Bloomberg.
Caption: Bloomberg Jack Clark author profile, last article published in August 2016
At first, he reported on data centers, distributed systems, supercomputers and other technologies. It was not until he joined Bloomberg in 2014 that he gradually turned his attention to artificial intelligence.
AI at that time was far from as exciting as it is today. ChatGPT would not be born until eight years later. Deep learning had just begun to make remarkable progress in fields such as computer vision. Most ordinary people had no idea about neural networks.
Clark was already fascinated.
His former boss and Bloomberg editor Alistair Barr later recalled that in 2015, few people paid attention to AI at that time, but Clark wanted to write about artificial intelligence news every day, and constantly asked editors for relevant topics.
In June 2026, Clark returned to Bloomberg as an interviewee and participated in the podcast "Odd Lots" by his old colleague Joe Weisenthal. Weisenthal took out the chat records between the two people ten years ago and asked him what he saw back then that made him dare to be so sure about the future of AI.
Clark said that he spent two years printing a large number of AI research papers. He also joked that this wasted a lot of printing money from the boss of Bloomberg.
In addition to reading papers, he also did something "very Bloomberg-specific" - drawing the technological progress of AI into charts. From computer vision, voice, video, to the level of AI playing video games, he continues to collect research data and observe changes in capabilities in various fields.
During this process, he discovered something interesting:
These seemingly unrelated research directions began to show similar exponential growth trends.
Clark gradually realized that what he was observing might be a general technology that was just beginning.
Caption: Interview with Jack Clark from Bloomberg's "Odd Lots"
Clark even noticed Nvidia ahead of time.
At that time, he looked through AI papers and found that almost all researchers were using NVIDIA GPUs instead of AMD products. He considered this an important signal worth reporting and tried to push Bloomberg to write a dedicated report.
But this topic selection was ultimately not made. Ten years later, he was still talking about it in the podcast, which was somewhat regretful.
Unfortunately, no reliable public information has been found to confirm that Clark has purchased or held Nvidia shares - if he had really bought it in 2015 and kept it today, that would be a huge gain.
Clark's research on technology also left a deep impression on researchers at the time. Just recently, former OpenAI researcher Andrej Karpathy recalled that he received many interview requests from technology reporters during his Ph.D., but Clark was the only reporter he consistently responded to at the time. Because Clark really knows how to read academic papers and can ask in-depth technical questions.
Capasi said he had almost never seen anyone like Clark before (or since).
The more he studied AI, the more Clark felt that he could not stay outside the industry and watch the excitement.
The turning point occurred in 2015. At that time, an artificial intelligence research institution called OpenAI was being prepared for establishment. While attending an AI conference in Canada, Clark interviewed Greg Brockman, who later became the co-founder of OpenAI.
At the end of the interview, he turned off the recorder and suddenly asked Brockman a question: "How can I work for you?"
Clark later recalled that the moment he asked this question, he knew that he could no longer remain an objective reporter. It is simply against the rules for a reporter to be eager to join his or her company right after interviewing the person he is interviewing.
In the months since, Clark has stayed in touch with OpenAI.
Caption: Rick Rubin's interview with Jack Clark, in which he talked about the past interview with Brockman
In the summer of 2016, Clark walked up to his lead editor, Alistair Barr, and asked him if he could go for a walk.
Barr already had a vague feeling that something was wrong - for an editor, it is usually not good news for a reporter to suddenly ask to go for a walk alone.
The two came to the edge of San Francisco Bay and sat down on a bench. Clark told him that he planned to resign and join a non-profit organization called OpenAI.
Barr expressed disapproval on the spot. OpenAI was founded less than a year ago and was relatively unknown, relying mainly on funding from supporters including Musk. He couldn't see any clear development direction for this company, let alone how it would sustain operations in the future; in contrast, Clark had already established a firm foothold at Bloomberg, responsible for reporting on Google and AI, and his job was stable and his performance was good.
Barr tried to convince him to stay, but Clark didn't listen.
Later Clark recalled that he received many emails from fellow journalists at the time, warning him that he was making the most serious mistake in his life.
Many years later, Barr wrote an article specifically recalling his experience in persuading Clark not to resign. The title of the article is: "In 2016, my employee told me that he was going to resign to join OpenAI. I say this is a bad decision. Today, he is an AI billionaire. 》
Caption: A memory article published by Alistair Barr on January 14, 2025
Barr admitted that he was wrong.
Clark also said in a later interview that Sam Altman (Sam Altman) and Brockman were willing to hire him, but they actually took a risk.
After all, OpenAI recruits mostly scientists who study artificial intelligence, and Clark studied English literature. What he is best at is reading papers, interviewing scientists, and writing their research into news.
Now, he finally no longer has to study these people through a recording pen.
From OpenAI to Anthropic
At OpenAI, Clark met Dario Amodei, who later became Anthropic CEO. The two began to work closely together in 2016, and gradually discovered that each other had similar judgments on the development of AI.
Clark was primarily responsible for strategy, communications and policy, and later became OpenAI’s policy director. He began to participate in AI governance discussions, studying how increasingly powerful models should be made available to the public and what restrictions they need to be subject to.
In addition to his work on OpenAI, Clark also founded the AI research newsletter Import AI in 2016 to regularly sort out cutting-edge research papers; the following year, he participated in the creation of Stanford University's AI Index project to track the development of artificial intelligence through data.
Caption: The introduction to Import AI, the main content is two large photos of Jack Clark
The technological trends that have excited him in the past few years are also constantly being verified.
Clark later recalled that during the years he worked at OpenAI, he watched with his own eyes as AI learned to handle increasingly complex video games and gradually demonstrated amazing language capabilities.
Every time researchers increase computing power and expand model size, AI performance will improve.
In 2020, the research on Scaling Laws participated by Amodei and others began to be able to predict the performance of the model based on the model parameters, data volume and training computing power. This gives OpenAI the confidence to train larger models.
In the same year, GPT-3 was born.
Clark recalled that he was very shocked when GPT-3 really achieved the expected performance. They know that bigger changes are yet to come.
However, as AI capabilities continue to improve, researchers are beginning to face an increasingly difficult problem: How should these models be made available to the public?
As early as 2019, when OpenAI released GPT-2, it chose to open it in stages instead of immediately releasing the full version because of concerns about the abuse of the model.
Clark was involved in the decision. He later explained that if an organization really wants to be responsible for the social impact of AI, it cannot open source the model and then let it go.
This cautious release approach has gradually become a common practice in the cutting-edge AI industry. Today, when companies such as OpenAI and Anthropic launch new models, they often open them to some users first and then expand the scope of use after testing.
But within OpenAI back then, everyone had their own opinions.
According to an investigative report by The Atlantic in September 2026, Altman proposed to provide GPT-3 private testing access to American rapper Kanye West in 2020, but this arrangement was blocked by Clark, the then policy director.
Caption: Investigative report by "The Atlantic Monthly", titled "The Inside Story of the Most Fierce Controversy in the Field of Artificial Intelligence"
Similar frictions are increasing.
Clark later recalled that they had been arguing within OpenAI over the years, and he did not insist that he was definitely right. Who was right and who was wrong on these issues might not be known for another ten years.
But Clark found that every time there was a dispute, he, Amodei and others were almost always on the same side.
Later, he explained why he decided to leave OpenAI:
If you stay, you will have to spend half your time arguing, and only the other half can be used for work. That being the case, why not just leave and spend all your time doing things with like-minded people?
In December 2020, OpenAI held an online all-staff meeting. According to "The Atlantic Monthly", when the meeting began that day, it looked no different from usual.
Altman introduced the research progress as usual, and announced the joining of several new employees. Then, he gave the right to speak to Amoudi.
Amoudi announced on the spot that he and six other colleagues were preparing to leave OpenAI and establish a new AI research institution.
As soon as the news came out, a large number of shocked expressions and exclamation points flooded into the conference chat window.
Clark left a last message in the chat window: "I love you all."
Immediately afterwards, the seven resignations were removed from the meeting, and OpenAI quickly revoked their email and Slack account permissions, and even remotely locked their work computers.
A few hours ago, they were core employees of OpenAI; now, everything has to start again.
A few hours later, the group reconvened for a video conference to discuss some very real questions: Where does the money come from? Where do GPUs come from? How to compete with OpenAI and Google? What's the name of the company? Also, who will buy the new computer?
After all, their old computers are no longer usable (laughs).
At one point they even considered naming the company Sparrow Systems, meaning "Sparrow System."
The name comes from a fable written by the philosopher Nick Bostrom in "Superintelligence": a group of sparrows want to find an owl to help them do their work, but they have not yet figured out how to control this raptor that is much more powerful than themselves.
Illustration: Chapter 1 of "Super Intelligence": The Fable of the Sparrow to be Continued
However, because the domain name could not be obtained, they finally chose Anthropic.
That day, seven people toasted to celebrate in the video conference. Although there were still a lot of problems before them, at least they could work according to their own ideas.
In 2021, Anthropic was officially established.
Many years later, Clark recalled this experience and felt that starting Anthropic with Amodei and others was much easier than quitting his job as a reporter.
And this research institution that left OpenAI has now become OpenAI’s most important competitor.
The story continues
There is no need to mention the subsequent story. Anthropic’s valuation has been rising, and the technology reporter who was dissuaded by his boss has also made it to the Forbes rich list.
However, he himself didn't seem too keen on discussing the matter.
In January 2025, Anthropic reported news of a new round of financing. According to the valuation at the time, several co-founders were already expected to become billionaires.
Alistair Barr, his old boss who advised him not to join OpenAI, thought this was another good topic, so he contacted Clark and wanted to interview him about his feelings about becoming a billionaire.
Clark replied: "Haha, Ali, thanks, but really not my style." (Haha, Ali, thank you, but this is really not my style.)
Barr later described Clark in the article as one of the gentlest, kindest, and most self-deprecating people he had ever met.
Although he already has a book wealth of hundreds of billions, Clark is still writing his own Import AI.
Caption: Import AI was last updated on October 12th before the publication of this article
Ten years ago, he read papers every week and tracked the progress of AI technology; ten years later, this habit has not changed much.
It’s just that his attitude towards AI has become a little more complicated.
In 2025, in a speech called The Curve, Clark recalled that when he was still working at OpenAI, he and Amodei often walked in the office and discussed the future of AI. They feel a bit like traders before the 2008 financial crisis, knowing something big was about to happen.
At Anthropic, this feeling did not disappear.
Clark said that in the past few years, he had called Amodei many times late at night or early in the morning and told him: "I am very worried. Your prediction has once again been proven right."
Amoudi replied: "Yes, we don't have much time left."
Clark couldn't help complaining: "For God's sake, Daario, can't you say something reassuring?"
Unfortunately, reassuring people is not Amodei’s style.
The reason why Clark is increasingly worried is that predictions about the growth of AI capabilities are constantly being verified: models can now not only write code and complete complex tasks, but also begin to demonstrate increasingly strong capabilities for autonomous action.
In May 2026, Clark participated in the podcast "The Rest Is Politics: Leading" and when talking about the development of AI, he said frankly:
“I'm scared of the technology that I'm building.”
(I'm afraid of the technology I'm helping to create.)A few months ago, Anthropic had already had a fierce dispute with the US Department of Defense over the boundaries of Claude's use.
The differences between the two parties focus on two things: Anthropic is unwilling to allow Claude to be used for mass surveillance in the United States or to control fully autonomous weapons without sufficient human supervision; the U.S. Department of Defense requires companies to accept a wider range of legitimate military uses.
In the end, the two parties failed to reach an agreement, and Anthropic was identified as a supply chain risk and subsequently challenged through legal channels.
In March, Clark held a closed-door exchange with bipartisan members of the U.S. House of Representatives Homeland Security Committee; in April, he publicly stated that the dispute with the Pentagon only involved a limited scope of contract issues and did not prevent Anthropic from continuing to introduce the capabilities and risks of new models to the government.
He hopes the government understands and uses AI, but also believes that some boundaries cannot be given up easily.
The dispute did not end with just a few communications. As of October, Anthropic was still challenging the Pentagon's decision through legal proceedings.
In addition to handling government relations, Clark also took on a new job this year.
In March this year, Anthropic announced the establishment of the Anthropic Institute, with Clark as the head, to study the security, economic and social impacts of AI.
On October 17, Clark appeared again at the IMF Annual Meeting in Bangkok. In preparation for the talk, he collected about 100 charts and was still updating the data that morning.
Ten years ago, he tracked the development of AI through papers and charts at Bloomberg; ten years later, he was already standing on the podium of the IMF, using charts to announce that "the singularity has arrived."
At the end of his speech, Clark made suggestions to the economists and policymakers in the audience: stricter regulatory rules should be established for AI, and if necessary, even slow down the speed of technology development.
As for where AI will eventually take the world, this time, he can no longer just predict.
Comments