Sam Altman's Latest Interview: What Mindset and Judgment Should We Have in the Face of Exponential Change?

Compiled by: Deep Thinking Circle

Have you ever thought about how a startup that was only established two weeks ago could completely redesign a set of mainstream office software—documents, spreadsheets, presentations—all centered around AI? This is not something I made up; it was recently stated by Sam Altman (co-founder and CEO of OpenAI) in an interview, where he mentioned he had just met such a company. Ten years ago, we could almost predict what a ten-week-old startup would look like. Now, if a ten-week-old startup still resembles those from a decade ago, it indicates that it has already fallen behind.

This interview was packed with information, discussing entrepreneurship, how OpenAI has navigated these years, and how he personally handles pressure and makes trade-offs. I’ve compiled some of the most striking segments, adding a lot of my own understanding, to share with you.

Most People Are Still Choosing Easy Battles

Altman mentioned that this moment in time is quite interesting; costs are rapidly decreasing, and the time required to get things done is shortening, which is precisely when startups have the most advantages. This phenomenon is happening across many fields, and theoretically, it should be the best era for entrepreneurship. However, he observed a rather contradictory phenomenon: most startups are still doing the same thing—creating AI agents for specific industries. This path is viable and can even be quite profitable, but it is unlikely to result in companies that will be truly remembered in this era.

What struck me as profound was his subsequent statement: despite the tools having completely changed and the capabilities of models continuing to rise, people are still hesitant to bet on something that cannot be achieved now but could be in two years. He said this temptation is particularly strong; it’s easy to use today’s agents to solve immediate problems, which is understandable, but that is not the path he would choose.

Upon reflection, this is fundamentally a matter of patience and belief. Being willing to lay the groundwork for something that cannot yet be realized is, in essence, betting that models will continue to improve and that one’s judgment of direction is correct. Most people fail to do this, not because they cannot see the trends, but because they cannot resist the urge to see returns immediately.

Believing in Exponential Growth Is Harder Than It Seems

Altman mentioned a method he has been using: every time he meets someone new, he mentally plots a coordinate for that person, assessing where they currently stand. The next time they meet, he checks how far and how quickly that person has moved forward. He said this belief is the same as his judgment of model capability improvement; fundamentally, he has a deep trust in exponential growth, whether it applies to a person, a company, or a model.

He said if he were still giving advice to entrepreneurs, this is the one thing he would want them to truly understand. Moreover, he noted that this concept is difficult to accept universally because the market itself has not yet adapted to the fact that model capabilities will continue to grow exponentially. Therefore, starting projects that require smarter and cheaper models is entirely reasonable.

I find this segment particularly worth pondering. Believing that a curve will continue to rise sounds like a simple principle, but truly basing all decisions on this belief requires more courage than one might imagine. Most people’s intuition about exponential growth is flawed; they either underestimate the accumulation of the previous years or begin to doubt whether the rapid ascent will continue during the fastest growth phase.

Enduring Chaos Is Something That Can Only Be Learned Through Experience

There was a part that left a deep impression on me; Altman said that no matter how well someone understands the reasoning in their mind, some abilities can only be truly developed through repeated experiences—operating in chaos and then believing that they can ultimately resolve it. This won’t cost you your life; even if you don’t know how to solve it now, you will find a solution. He said this is something that can only be learned and cannot be taught, and he believes this is the biggest shortcoming of many young founders—they have not gone through the process of gradually coexisting with chaos.

He also shared a very straightforward analogy: the first time you encounter something that could potentially lead to the company’s demise, it feels like the sky is falling. But after you survive the tenth time, you’ll think, “I’ve survived the previous nine times; this time probably won’t be that bad.” He later realized that bad things will always happen; rather than resisting them, it’s better to learn to accept this uncomfortable process. He said most people think the opposite of a bad experience is a good experience, but in reality, the opposite of a bad experience is no experience at all. In the not-so-distant future, you will inevitably enter a phase where nothing happens, so even a terrible experience is worth being grateful for.

This statement left me stunned. We are too accustomed to viewing pain as something to avoid, but if the opposite is not comfort but rather emptiness and numbness, then enduring chaos seems not just a cost but a part of being alive.

What Promise Does a Trustworthy Company Make to the World?

When Altman discussed mission, he mentioned something he cares about deeply: one of his biggest concerns regarding AI risks is that a small group of people or a company might feel they should control the entire world, which he refers to as AI authoritarianism. Therefore, what OpenAI aims to do is make intelligence extremely abundant and cheap, putting it in everyone’s hands rather than in the hands of a few. He emphasized that they do not intend to create products in every vertical field themselves but want to build the foundational capability of intelligence, allowing the entire economic ecosystem to grow various things on that foundation.

There was a part I found particularly interesting; he said that the things needed to create abundant intelligence—chips, energy, data centers, robots—are precisely what humanity will need immediately after intelligence becomes abundant. Even if ideas and creativity become worthless, we still live in a physical world and still need things to be genuinely created. Therefore, energy and robots are not just stepping stones to that goal; they are also what will be needed right after.

When I read this part, my first reaction was that this logic is quite simple: ultimately, no matter how intelligent something is, for it to have any impact in reality, someone must move the material. But upon further reflection, it also reminds us not to think of intelligence as something too abstract; no matter how powerful the model, it ultimately relies on a lot of heavy, physical infrastructure to be realized.

The Invention of the Company Is More Important Than Many Technologies Themselves

Altman shared a thought he had as a child; he was always curious about the Industrial Revolution, where a bunch of technologies coincidentally emerged around the same time and expanded at roughly the same speed. He always wondered which technology was the most critical one. He said, looking back from today’s perspective, he believes the truly key invention is the joint-stock company. Before that, businesses relied on trust among acquaintances, with many family businesses and no concept of shareholders. After the emergence of joint-stock companies, sovereign states granted this new entity an unprecedented status—not granting it the power of a state but giving it capabilities far beyond individuals, allowing it to pool capital, engage in high-risk and highly speculative ventures, and enable different companies to specialize in different areas while cooperating with each other.

He mentioned a graph I’m eager to look up, showing the decline in the proportion of extremely impoverished people in human history and the decline in infant mortality rates. If we extend the timeline of human history and mark the point when joint-stock companies were invented, the shape of the curves will look significantly different afterward. He said this is an extraordinary manifestation of capitalism in human society.

I found this segment particularly enlightening. When we usually discuss startups, we focus on products, financing, and growth, rarely stepping back to consider that the organizational form of a company itself is a technological invention that binds the interests of a large group of people together. Thinking this way, entrepreneurship is essentially utilizing this invention and layering your own elements on top of it.

Believe in a Few Things, Keep Everything Else Flexible

When discussing how to make long-term plans, Altman said he doesn’t usually work backward from the future to the present. His more habitual approach is to first identify a few directions he firmly believes in and then move step by step forward from the current point in time, clarifying what can be done now and what can be done this year. Only in rare cases does he plan for five or ten years ahead. He mentioned having seen too many people holding onto a multitude of beliefs about the future, only to be constrained by their rigid worldview. You might see some rocket companies suddenly pivoting to AI; this is such a situation. The truly useful approach is to hold onto a few deeply believed things while keeping everything else flexible, firmly maintaining the core.

He mentioned a friend’s company core value, called “critical path,” meaning to always focus on the biggest stumbling block in front of you, move it out of the way, then find the next one, and keep repeating this action. He said he has been very clear over the years that his critical path in life is to make intelligence abundant; as long as there is no strange concentration of power, he believes this will bring immense prosperity. He said he is rarely tempted by other ideas and seldom thinks about whether to change goals, but recently he has started to seriously consider what comes next if superintelligence is indeed on the horizon.

I found this quite moving; for a person to focus on a single critical path for so many years without being distracted by other opportunities is more challenging than any planning methodology. Ultimately, planning is not about how accurately you can calculate but whether you can consistently believe in a few things and filter out the remaining noise.

Are You Willing to Fly on the Edge of Risk?

Altman mentioned that he has always adhered to a principle: when in a somewhat risky state, one should get on the plane. He shared a story from the time shortly after ChatGPT was released; world leaders were quite anxious, with some questioning whether this thing was going to spiral out of control. He could sense a storm gathering. So, he took Brian Chesky’s (co-founder of Airbnb) advice and decided to embark on a series of intensive visits, originally planned for about eight cities, but they ended up covering 28 countries in 35 days. He said he basically lived on the plane during that time; it was a strange experience—though comfortable, traveling itself is still quite exhausting, with jet lag and missing one’s own bed and office.

He also mentioned an interesting criterion for distinguishing between a real trend and a false trend. A false trend appears as something that many people are excited about, but those who actually buy it lose interest after a while and do not design their lives around it; ultimately, it gathers dust. He used VR as an example. A real trend, on the other hand, is something that continues to appear in your daily life; for him, ChatGPT is almost used every day—sometimes for three hours, sometimes hardly at all—but it remains a constant part of his life. He said this judgment method is something he summarized after observing numerous startups at YC (Y Combinator); as long as one is willing to spend time analyzing this data, many insights can be gained.

I really like this method of judging the authenticity of trends because it is simple enough; you don't need to look at complex growth curves. Just ask one question: has this thing quietly embedded itself into your daily rhythm, or did it just excite you for a while before being cast aside?

Sometimes Asking Out Loud Can Get You the Impossible

Altman talked about Codex (OpenAI's programming agent application) as an example. He said it was a memorable experience of asking for something seemingly impossible. At that time, they were clearly lagging behind Claude Code (Anthropic's programming agent product) in the programming space. Logically, in such a situation, trying to turn the tide in a category where someone else has already taken the lead is generally considered impossible. Most people would accept this and move on to the next direction. However, they felt this matter was too important to give up easily, so they assembled a team and assigned them this almost suicidal task. As a result, this team achieved a rare success in business history; among the best programmers around them, the most used programming tool is this product.

He stated directly that if they hadn't asked out loud and assigned this nearly impossible task to the team, none of this would have happened. His own explanation is that programming is too important for RSI (Recursive Self-Improvement), not to mention the economic value behind it; they couldn't convince themselves to give up this track.

When I read this part, I thought about how asking out loud seems simple but actually goes against a very strong social default. Everyone assumes the winner has already been determined, and trying to compete is destined to fail. But what Altman did was refuse to accept this default, first assuming that it is possible, and then trying.

Killing Projects and Reassessing Team Morale

In the interview, a very poignant question was raised: how do you make the decision to pull the plug on a project that has already been invested in for over a year, spent a lot of money, computing power, and human effort, and that users are enjoying? Altman said this is not something that can be decided in one meeting; it is more like a slowly accumulating awareness. You gradually realize that these resources, these people, and this product direction could create greater value elsewhere, and then you have to make this painful decision.

He gave two examples: when GPT-3 (OpenAI's early language model) was truly operational, they shut down a similarly exciting robotics project and consolidated all resources onto this. Recently, after the programming agent was successfully operational, they also shut down Sora (OpenAI's video generation product) and the browser, both of which were promising directions, and focused entirely on programming. He said this does not mean Sora was not doing well; if they continued to invest, it could have been very successful. It was just that investing computing power and energy in the programming agent was more important at that moment.

Regarding how to get the team to accept this shift, he said everyone actually understands the mission and the trade-offs behind it. Even if it is difficult at the moment, the team knows why they have to do this. Some people may be unhappy, but more will say, "I understand why we have to do this; it is right for the mission."

I think the hardest part of this segment is not the decision itself but how to make a group of people who have already invested a year of effort believe that the next thing is also worth going all in on. This requires not only judgment but also strong communication and team leadership skills.

Focus on Your Strengths and Find the Right People for the Rest

When discussing how to become better at something, Altman mentioned Johnny Ive (former Chief Design Officer of Apple) as an example. He said the most important lesson he learned from Johnny is that truly great design comes from thoroughly studying the problem itself rather than having a sudden flash of inspiration for a solution. If you rush to the answer or lock yourself into a solution too early, the result is usually not very good.

He also candidly admitted that he is not good at product design. He does not agree with the idea that you can only hire people who deeply understand your field. He said he doesn't understand design at all, but just by talking to Johnny for thirty minutes, you can see that this person is truly exceptional. His principle is that instead of forcing himself to fill in those inherent weaknesses, he should spend all his energy on his strengths and make them even stronger.

There’s also a personal detail he mentioned: during the time he was working on Sora, to understand the product experience, he intentionally made himself addicted to TikTok. Initially, he just wanted to learn, but later he found that he really enjoyed it. From ten minutes before bed, it turned into an hour, and then three hours on Saturday afternoons on the couch. He said that feeling was great at the moment, but he clearly knew it was not good for him. Later, he turned off most app notifications, including messaging apps, and deleted TikTok because he felt it was too powerful for him to control.

When I read this part, I was quite surprised that someone who is constantly creating more powerful AI products could also be affected by the very things they create. He had to rely on the most primitive methods, like turning off notifications and deleting apps, to regain control. This reminds me that judgment is not something you create once and for all; it requires continuous self-management, even for those who understand product design logic the best.

My Own Reflections

Listening to the entire interview, I felt that what runs through it is not a specific methodology but a mindset in the face of uncertainty. Believe that the index will continue to rise, endure the chaos until it is no longer frightening, hold on to a few deeply believed principles, ask when you need to, let go when you must, and focus your energy on what you truly excel at. These insights may not seem fresh when viewed in isolation, but the challenge lies in being able to do all these things simultaneously in an environment where all assumptions are being overturned.

Altman concluded with a statement that left a deep impression on me. He said that most startups today still look quite similar to those from ten years ago because that so-called correct way of doing things is still being taught. At most, they change the wording, saying they hire fewer people and spend more on tokens, but that is far from enough. I think this statement serves as a reminder for everyone: the tools have completely changed, and if the way of thinking remains in the old coordinate system, then no matter how radical the words are, the things produced are likely still old.