Asked for an example of his current big problem, OpenAI chief executive Sam Altman pointed to “the pace of progress as we enter this RSI era”. He spoke on Raj Shamani’s Figuring Out podcast, in an episode posted on YouTube on 3 November. The questions he listed included “how smart an AI system should be allowed to be relative to a person” and “How quickly can that improve?”
In the same conversation Shamani named Mira Murati, Ilya Sutskever, Dario Amodei and Elon Musk, who now lead rival labs, and asked what the disagreement was about. Altman answered that “there is something that happens in many people’s brains where they only trust themselves to build this because the stakes feel so high”.
The four labs those names lead made their first public announcements within 1,362 days of each other, by our arithmetic: Anthropic on 28 May 2021, Thinking Machines Lab on 18 February 2025.
The full episode · Raj Shamani
Sam Altman on Figuring Out with Raj Shamani, episode FO583, posted on YouTube on 3 November 2026. The pacing passage starts at about 59:08; the rival-labs answer at about 41:26.
The key insight: As we read it, Altman is arguing two things at once: that AI is safer with several competing labs, and that the people who built those labs left because each trusted only their own approach. Both describe a race in which control over the pace of improvement is the prize.
Shamani asked: “What’s the most uncomfortable thing that you do these days?” Altman said that “this current debate about pacing progress feels like a very difficult thing to get right and pretty important”.
Asked for an example, he described how to avoid “a loss of control incident or a concentration of power incident”. He listed “what is the right governance mechanism”, how smart a system should be allowed to be relative to a person, how quickly that can improve, questions around “security and monitoring and alignment”, and “where can we afford to take risks”.
RSI is short for recursive self-improvement. In the passage, Altman gave no threshold, mechanism or date for any of these questions.
On 21 October, in an interview with Matthew Berman, he had said the world will need “speed limits or whatever on how quickly we let RSI go”. Separately, Anthropic says in its amended S-1 that it “may imminently meet” a threshold in its Responsible Scaling Policy.
First public announcement of each lab whose leader Raj Shamani named in his question to Sam Altman: Anthropic (Series A, 28 May 2021), xAI (company announcement, 12 July 2023), Safe Superintelligence Inc. (19 June 2024) and Thinking Machines Lab (18 February 2025), with OpenAI’s own announcement of 11 December 2015. Data: company announcements; departures as reported by Fortune and Bloomberg. The 1,362 days are our arithmetic.
What Was Posted on X
๐จBREAKING: Sam Altman just gave a mind-blowing podcast. He said in a conversation released with Raj Shamani today that the hardest problem he is working on right now is pacing progress ‘as we enter this RSI era’, including deciding ‘how smart an AI system should be allowed to be relative to a person’ and how fast it can improve.
Asked why Dario Amodei, Ilya Sutskever, Mira Murati, and Elon Musk all left OpenAI to build rival labs, Altman said ‘there is something that happens in many people’s brains where they only trust themselves to build this, because the stakes feel so high.’ He summed up the mindset as: ‘His approach is bad, and so I can’t let power get concentrated in his hands.
He said: ‘One of the most confusing things I have ever observed in my career’ is why from 2015 through 2021, other researchers looked at the same scaling data and did not say, ‘This is crazy, like we have to invest much more in this.’
He said it was so obvious inside OpenAI that ‘almost we felt delusional’, while outsiders assumed ‘there’s no way that something so dumb can just fix all of its problems by scaling it up.’
Sam Altman says the last time he felt ‘heart-pounding’ nervousness was right before OpenAI announced its Navier-Stokes math result. He said OpenAI’s recent math results convinced him ‘that we are closer to this milestone than I expected us to be’ on building an AI to discover new science.
‘It’s an extraordinary time in terms of just all of the model progress.’
And on what he ultimately wants AI to become:
‘I am extremely excited to see how far we can push AI to discover new science. Math – the math results were quite amazing to me, and it really made me think that we are closer to this milestone than I expected us to be. But a machine that can help us understand the world better, discover, you know, the deepest insights in physics, figure out new medical treatments, new sources of energy, things that I think we can’t even imagine, but that better scientific understanding will enable. I am extremely excited for us to go build that machine.’
@choblin29 posted this clip of the Raj Shamani episode at 17:42 UTC on 3 November, with quotes from Altman. This piece quotes YouTube’s automatic captions of the full episode, embedded above; the post’s ‘hardest problem’ is its own summary of Altman’s words.
Why He Says Rivals Left
Shamani said OpenAI had attracted the talent and then people started leaving. He named Murati, Sutskever, Amodei and Musk, said they are all building something of their own, and asked: “What is the disagreement about?”
Altman gave the answer about people who “only trust themselves to build this”. Asked whether it was about approach or about power in someone else’s hands, he said “I think they go together”.
He put the mindset this way: “his approach is bad and so I can’t let power get concentrated in his hands. Something like that.” He named no one in the answer. The names came from Shamani’s question.
Pluralism and OpenAI’s Edge
Earlier in the episode, Altman said “I believe in pluralism”. He described two ways AI goes wrong: a loss of control to AI, or “one person or company or country gets way too powerful relative to everybody else”.
He said it would be “much better for the world for there to be multiple AI efforts with slightly different approaches that are competing with each other”.
Asked about OpenAI’s advantage among four or five large companies, he said OpenAI has “the most users and the best stack from chips to data centers to models”. He added: “We don’t want to be like the one company to rule everybody.”
He called it “one of the most confusing things I have ever observed in my career” that others looked at the same scaling data from 2015 to 2021 and did not invest more.
On India, he said “India is like our biggest market in the world” and “growing very quickly”. In his longer answer the captions render the next words as “by touch users”, which we cannot resolve, so we do not say which measure he meant.
The Four Labs, by Date
OpenAI announced itself on 11 December 2015. Its post named Sam Altman and Elon Musk as co-chairs and Ilya Sutskever as research director.
Anthropic announced a $124 million Series A on 28 May 2021, led by Dario Amodei as chief executive; it said its team had “previously conducted research into GPT-3”. xAI’s company page lists its company announcement on 12 July 2023. Bloomberg reported that day that Musk had left OpenAI’s board in 2018.
Sutskever posted “I am starting a new company” on 19 June 2024, for Safe Superintelligence Inc. Thinking Machines Lab announced itself on 18 February 2025; Fortune reported that Murati had left OpenAI in September 2024.
From Anthropic’s Series A to Thinking Machines Lab’s launch is 1,362 days, by our arithmetic. Anthropic’s Series H announcement in May 2026 valued it at $965 billion post-money.
The Business Lever
As we read it, the business lever is concentration: of talent, of compute and of control over how fast models improve. Altman argues for several competing labs while describing rivals who left because they trust only their own approach.
As we read the dates, the four labs Shamani named opened within 1,362 days of each other, by our arithmetic, and Anthropic, the oldest, put its post-money valuation at $965 billion in its May 2026 Series H announcement.
As we read Altman’s pacing answer, he framed the open questions as governance ones: how smart a system should be allowed to be relative to a person, how quickly it can improve, and where risks can be taken. He did not give OpenAI’s answer to any of them.
As we read his advantage answer, Altman ties OpenAI’s edge to scale: the most users and a stack from chips to data centers to models, which is the result of the compute bet he described making after GPT-4.
Sam Altman, Figuring Out with Raj Shamani, posted 3 November 2026 (automatic captions)
“there is something that happens in many people’s brains where they only trust themselves to build this because the stakes feel so high”
Three Implications
FOR AI LABS Altman says several competing labs with different approaches are better for the world than one, while describing the rival founders’ motive as trusting only themselves.
FOR GOVERNANCE Altman lists the open questions as how smart a system should be allowed to be relative to a person, how quickly it can improve, and security, monitoring and alignment; he gave no thresholds.
FOR CAPITAL The four labs Shamani named launched within 1,362 days, by our arithmetic; Anthropic’s Series H valued it at $965 billion post-money.
The framework argues: “The transformer is a commodity now; what changes between paradigms is where intelligence accumulates and who can capture it.”
As we read it, Altman’s answer about rival labs is a dispute over the second half of that sentence: who should capture it. The framework is about data pipelines and moats; it does not discuss lab founders or AI governance.
What Is Not Established
Altman did not say what pacing mechanism OpenAI will use, or how smart a system should be allowed to be. He did not attribute the “only trust themselves” mindset to any named person, and we did not ask Murati, Sutskever, Amodei or Musk for comment.
The X post that spread the clip says Altman called pacing the hardest problem he is working on. His words in the captions are “most uncomfortable” and “a very difficult thing to get right”; “current big problem” was Shamani’s phrase.
The quotations are from YouTube’s automatic captions of the full episode. The captions misspell two of the names in Shamani’s question, and we did not transcribe the clip’s audio separately; the post’s quotes on pacing, the rivals’ mindset and scaling match the captions.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
In a Raj Shamani episode posted on 3 November, Sam Altman said that pacing AI progress “as we enter this RSI era” is a very difficult thing to get right, and that rivals left OpenAI because they “only trust themselves to build this”. The four labs Shamani named launched within 1,362 days of each other, by our arithmetic.
A note on sourcing. This piece is based on the full episode on Raj Shamani’s YouTube channel, posted on 3 November, read through YouTube’s automatic captions and compared with the wording of the @choblin29 post that spread the clip; the founding announcements of OpenAI, Anthropic, xAI, Safe Superintelligence Inc. and Thinking Machines Lab; and reports by Fortune and Bloomberg on the departures. The 1,362 days are our arithmetic. We did not contact OpenAI, Altman or the four labs.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
