Sam Altman's Raj Shamani Interview: Key Points and Timestamps

A summary of Sam Altman's interview on Raj Shamani's Figuring Out podcast: India, the GPT-4 compute bet, the early ChatGPT launch, two ways AI could go wrong, and his advice to founders.

▶ Watch the full interview on YouTubeRaj Shamani · 3 November 2026

Sam Altman, the CEO of OpenAI, was the guest on episode 583 of Figuring Out, the podcast hosted by the Indian entrepreneur Raj Shamani. The episode was published on YouTube on 3 November 2026 and runs 68 minutes^[1]. It had more than 4 million views by 10 November 2026. Shamani asked mostly about how Altman thinks and makes decisions, not about new products. This page summarises what Altman said, grouped by topic, with timestamps.

Altman is also on our list of the most powerful AI influencers.

Topics and timestamps

The chapter names are shortened from the episode description^[1].

India is OpenAI's biggest market

Shamani said people he talks to in the US are slightly pessimistic about AI, while people in India are very optimistic. Altman said India is OpenAI's biggest market in the world, bigger than the US, and is growing very quickly^[1]. As an example, he described an Indian student who had never raised money before and used OpenAI's tools to run a school fundraiser from start to finish.

Altman gave one reason for the gap. Countries whose economies grow faster, and where products and services are in short supply, tend to welcome AI more than developed economies. He said the US is "unusually negative on AI" compared with the rest of the world, and India is perhaps unusually positive.

Equity, criticism and pushback

Shamani asked why Altman has become "the bad guy" in the US although he does not own equity in OpenAI. Altman said Americans are sceptical of people in power and of the power of the tech industry. He thinks that scepticism is partly healthy but has gone "a little far".

He called his decision not to take equity in OpenAI a mistake. He made it for reasons he thought were good, but it has confused people, because working without equity is far from the usual reasons people start companies. He compared it to an artist or an athlete who works hard without earning much from it. He said he does the job because it is the most interesting and useful work he can imagine.

Altman said people do sometimes tell him he is delusional, and that he should probably hear it more. Many people in roles like his have gone wrong because they were surrounded by yes-men. His defences are old friends, people he has worked with for more than ten years, and a culture that rewards people for pushing back when he is wrong. He said avoiding yes-men gets harder the more often you have been right and the more power you have.

He also said he rarely acts against everyone's advice. If smart people around you all disagree and you go ahead anyway, he thinks you should treat it as an expensive, high-stakes bet that is unlikely to be right.

Scale: the idea behind OpenAI

Altman said most people get only a few great insights in life and apply them again and again. One of his is that scale has surprising effects. He saw it in companies, cities, careers and networks such as Y Combinator, where he worked before OpenAI. When something that works at small scale is scaled up, the returns often grow faster than the size.

When OpenAI started in 2015, few people took it seriously. Altman said the common view was that artificial general intelligence (AGI) was perhaps 100 years away, or would never come. OpenAI's team saw that neural networks got better as they got bigger, years before scaling laws made this precise. He said he still finds it confusing that from 2015 into 2021 other people looked at the same data and did not invest much more in neural networks.

He described one argument against OpenAI's approach of scaling up models that sounded convincing at the time. A model learns to predict what comes next from its training data, so the argument said it could never produce new knowledge. Altman said this missed the main point: the models learn a general ability to reason, and that ability can extend to problems they have never seen.

The compute bet after GPT-4

Shamani asked for one big decision Altman made without certainty. Altman chose the period after OpenAI finished training GPT-4, which he placed around August 2022, seven or eight months before its public release. The team saw that the model would open new research directions, including the chain-of-thought work that led to reasoning models. OpenAI then decided to "bet the company" on buying a large amount of compute, before the rest of the world had seen GPT-4.

The alternative was a smaller compute purchase and slower scaling. Altman said that if GPT-4 had been even slightly less capable, at "GPT-3.8" level, OpenAI would not have had the conviction for a bet that large.

How he judges a contrarian bet

Altman said this is the best time he has seen to start a company, and probably the best time in history. On picking ideas, he compared OpenAI with photo-sharing apps. In 2015 OpenAI may have been the only AGI lab founded that year, while there were perhaps 10,000 photo-sharing startups. Most people thought photo sharing was the better idea, but those founders competed with 10,000 others and with Instagram's network effect.

His advice is to look for ideas that are more valuable than they seem. If such an idea would be very valuable when it works, the bet is worth taking even when success is unlikely. He said understanding a technology before other people do is a real advantage, and founders should value those moments.

On how to find a non-consensus bet, he said school taught him to answer other people's questions, not to choose his own. Learning which questions to ask is how he found his non-consensus bets, and he thinks young people should be taught this skill.

To check that a bet is not wishful thinking, he explains it to a friend who is optimistic and smart but mentally disciplined, and not an expert. In OpenAI's early days, these people often said the evidence changed their view. If everyone had agreed, the idea would have been too obvious. If no one had agreed, Altman would have doubted it more.

Why ChatGPT launched early

Altman named the ChatGPT launch as a time when he disagreed with the rest of OpenAI's leadership. Most of them wanted a chatbot but thought this one was too early and not good enough. An internal version had poor retention: about 5% of the people who tried it kept using it.

He gave two lessons from Y Combinator as his reasons. First, if you are not a little embarrassed by your first version, you launched too late, because you need feedback. Second, a small number of people who love a product matters more than many people who like it a little. The 5% who kept using the internal version used it heavily.

He said he is happy for OpenAI to release "some early sketches" to learn from, but changes that reach every ChatGPT user need a high bar. Enterprise products need a very high bar, because companies now rely on OpenAI almost as critical infrastructure.

Two ways AI could go wrong

Asked about his biggest worry, Altman said OpenAI has built something that will be an extraordinary force in society. He believes it will be a force for good overall, but the stakes and the responsibility are high.

Shamani then asked why OpenAI building superintelligence would make it less of a threat. Altman described two "cartoonish" ways AI could go very wrong:

• Loss of control. People hand control of the future, on purpose or by accident, to AI systems far more capable than they are, and cannot get it back.
• Concentration of power. One person, company or country becomes far more powerful than everyone else.

Altman said he believes in pluralism and that OpenAI alone having this technology would not be good either. Because he dislikes intense concentration of power, he wants several AI efforts with different approaches competing with each other. Later in the episode he said one hard, new problem is how to think about the pace of progress "as we enter this RSI era". RSI is a common abbreviation for recursive self-improvement, and he did not expand it. The problem includes how smart an AI system should be allowed to be relative to a person, how fast it may improve, and what governance, security, monitoring and alignment it needs.

Asked about OpenAI's advantage over four or five well-funded rivals, he named the most users, its stack from chips and data centres to models, its brand, and values he thinks users want. Of his own advantage, he said he has watched AI develop for a long time and has connections across the industry. He noted that Demis Hassabis, who leads Google DeepMind, has been working on it longer.

Departures and OpenAI's culture

Shamani listed people who left OpenAI or started rival efforts, including Mira Murati, Ilya Sutskever, Dario Amodei and Elon Musk. Altman said that because the stakes feel so high, many people come to trust only themselves to build this technology. He described their thinking as: this person's approach is bad, so power must not be concentrated in this person's hands.

On why about 700 employees threatened to resign when he was fired, Altman said it was not mostly about him. He credited a group of people motivated by the same mission and a culture that holds high conviction where needed and stays open-minded elsewhere. He hopes history will remember OpenAI for rediscovering the culture that makes a great research lab.

His advice for founders who want that kind of culture is to state a clear, ambitious goal and not hedge it. OpenAI said openly that it was trying to build AGI when few others would, and he thinks DeepMind had stopped saying so at the time. That attracted the talented people who shared the belief, even though most people thought OpenAI was crazy.

What Altman looks for in founders

Shamani mentioned companies such as Reddit and Stripe from Altman's investing career. Asked how he judges a founder in a short meeting, Altman named the founder traits he thinks get talked about less:

• clear communication that is easy to follow
• leaving the meeting wanting to work with them, because an early founder must win over customers, hires and investors
• coming up with new ideas, and improving ideas, quickly
• well-reasoned views on the world he has not heard before

He said there is no single pattern for success. If you swapped the CEOs of the top ten Y Combinator companies, it would be a disaster. Failure is easier to predict: lack of drive, focus, clarity or vision, and an inability to solve hard problems again and again.

Altman said ambition can be taught, which was one of his most pleasant surprises at Y Combinator. What worked best was a peer group going through the same programme together, together with challenges of increasing difficulty. He said he became much more ambitious himself this way.

His general advice to founders is to "listen to less advice". He thinks Y Combinator could reduce its experience to five to ten confident rules, led by its motto "make something people want". Beyond those, most advice is too specific to help, and founders learn by doing.

Compounding and questioning yourself

Shamani mentioned Altman's essay of 13 rules for success^[3]. Altman said he does not remember the 13 rules and could write a shorter version now. He said the first point, about compounding, is probably underestimated. He repeated a joke: someone may have 20 years of experience but "20 times the same one year of experience". He said this happens because many people they stop doing uncomfortable, new things early in their careers.

He said conviction does not mean never questioning yourself. He tries to question everything while still acting consistently until he finds a reason to change his mind. His method is to look at a question from every angle, throw out bad ideas, then ask other people and try a new version. He also needs a rhythm of intense meetings at the office and long periods of solo work at home.

His life now

Altman said his life is now precisely scheduled and he has not done anything spontaneous in a long time. The last time someone asked what he misses, he realised it was sitting in a coffee shop on a Saturday afternoon with a coffee and a laptop. He could still do it, but two security guards would sit at the next tables.

He said he gets nervous rarely now. The last time was before OpenAI announced its Navier-Stokes result. OpenAI announced on 8 September 2026 that it had solved the Navier-Stokes Millennium Prize problem, and mathematicians have since questioned how the result was produced and who reached it first^[2]. Altman used to meditate for about two hours a day and can still reach a calm state quickly.

Asked whether he is comfortable with his children growing up in a world shaped by AI, he said he does not know what the future will look like. He has "enormous trust in people" to adapt, and he thinks the world will most likely keep getting much better.

What he wants to build next

Altman said he is most excited to see how far AI can go in discovering new science. The recent math results made him think AI discovering new science is closer than he expected. He wants a machine that helps people understand the world better, with new insights in physics, new medical treatments and new energy sources.

On the cost of his role, he said the work is tiring and stressful, and many people project their anxiety about AI onto him. He said he still feels very lucky, and this life is now all he knows.

Sources

1. Sam Altman On AI Revolution, Superintelligence & Future Of Humanity | FO583 Raj Shamani — Raj Shamani, YouTube (3 November 2026; views checked on 10 November 2026)
2. A Modest Proposal — Peter Woit, Not Even Wrong (9 September 2026)
3. How To Be Successful — Sam Altman (2019)