Have you ever thought about how a startup that has only been around for two weeks can completely redesign a full suite of mainstream office software, including documents, spreadsheets, and presentations, all centered around AI? This is not something I made up. Sam Altman (OpenAI co-founder and CEO) mentioned this in a recent interview after meeting such a company. Ten years ago, you could pretty much predict the look of a ten-week-old startup. Now, if a ten-week-old startup still looks like what startups looked like ten years ago, it means it's already falling behind.
This interview was full of insights, discussing entrepreneurship, the journey of OpenAI over the years, as well as how he personally deals with stress and makes trade-offs. I have organized a few segments from it that resonated with me the most, added my own interpretations, and written them for you to read.
Most People Are Still Choosing Easy Battles
Altman mentioned that the current time is very interesting, with costs rapidly decreasing, and the time to accomplish tasks significantly shortening. This is when startups have the most advantage. Such occurrences are simultaneous in many fields, making it theoretically the best time for entrepreneurship. However, he observed a somewhat contradictory phenomenon. Most startups are still doing the same thing—creating an AI agent for a specific industry. While this path is viable and may be profitable, these companies most likely will not be the ones remembered from this era.
What struck me the most was his following statement. Even though the tools have completely changed, and model capabilities are still increasing, most people are still hesitant to truly bet on something that can't be done now but will be achievable in two years. He mentioned the strong temptation to just use today's agent to solve current problems, which is understandable to everyone. However, that is not the path he would choose.
Upon reflection, I realized that this is fundamentally a matter of patience and belief. Being willing to lay the groundwork for something that cannot be achieved at present is essentially a bet on the models continuing to strengthen and on the accuracy of one's judgment. Most people can't do this, not because they can't see the trend but because they can't resist wanting to see immediate returns.
Believing in Exponential Growth Is Much Harder Than It Seems
Altman mentioned a method he has always used. Every time he meets a new person, he mentally assigns a coordinate to that person to assess where they currently stand. The next time they meet, he looks at how far and how fast that person has progressed. He mentioned that this approach aligns with his assessment of improvement in model capabilities, essentially demonstrating his profound trust in exponential growth, whether it applies to an individual, a company, or a model.
He said that if he were still advising entrepreneurs, this is the one thing he would most want the other person to truly understand. And he said that the reason this is very hard to be widely accepted is that the market itself has not yet adapted to the fact that the capability of the model will continue to grow exponentially, so starting now on projects that require smarter, cheaper models to be viable is actually entirely reasonable.
I think this passage is particularly worth pondering. Believing that a curve will continue to rise steadily sounds like a very simple principle, but to base an entire decision on this belief requires much more courage than imagined. Most people's intuition about exponential growth is wrong, either underestimating the accumulation of the first few years or starting to doubt during the fastest ascent in the middle that the momentum is about to stall.
Tolerating Chaos Can Only Be Learned Through Experience, Not Taught
One part that left a deep impression on me is Altman saying that no matter how well a person understands a principle in their mind, some abilities require repeated experiences to truly master, operating in chaos, and then believing that you will eventually figure it out. This is something that cannot be learned but only experienced. He feels that this is the biggest shortcoming of many young founders—they have not been through the process of slowly learning to coexist with chaos.
He also gave a very straightforward analogy. The first time you encounter a crisis that could kill the company, it feels like the sky is falling. But after you survive the tenth time, you will think, I've made it through the previous nine times, so this time probably won't be that bad. He later realized that bad things were always going to happen, and instead of resisting, it is 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. In the not too distant future, you will inevitably enter a stage where there are no ripples, so even a terrible experience is worth being grateful for.
Upon reading this sentence, I was stunned. We are too accustomed to treating pain as something to avoid as much as possible. But if the opposite is not comfort but emptiness and numbness, then tolerating chaos seems to be not just a cost, but more like a part of being alive.
What a Trusted Company is Actually Committing to the World
When Altman talked about mission, he mentioned something he cares deeply about. One of the AI risks he is most concerned about is a small group of people or a company thinking they should control the entire world, which he calls AI authoritarianism. So what OpenAI wants to do is to make intelligence extremely abundant and cheap, putting it in everyone's hands rather than in the hands of a few. He particularly emphasized that they do not intend to do every vertical product themselves, but rather want to do the intelligence, the underlying capability, well, allowing the entire economic ecosystem to grow various things on this foundation on its own.
There was a part that I found particularly interesting, where he said that the things needed to create abundant intelligence—such as chips, energy, data centers, and robots—are precisely the things that humans most need immediately after intelligence truly becomes abundant. Even if ideas and creativity become worthless, we still live in a physical world and still need things to be actually made. So, energy and robots are not only stepping stones toward that goal, but also things that will be needed immediately afterward.
When I read this part, my first reaction was that this logic is quite simple. Ultimately, no matter how smart something is, if you want something to happen in reality, someone still needs to move matter. But upon closer thought, this is also reminding us not to think of intelligence as too abstract. No matter how amazing a model is, in the end, its real-world implementation relies on a pile of very cumbersome, very physical infrastructure.
The Company as an Invention 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 happened to emerge in the same period and happened to expand at almost the same rate. He always wondered which technology was the most crucial one. Looking back with today's perspective, he believes the truly key invention was actually the concept of the joint-stock company. Before that, businesses relied on trust among acquaintances, with a bunch of family businesses and no concept of shareholders. After the joint-stock company emerged, the nation-state endowed this new entity with an unprecedented status, not giving it state power, but giving it capabilities far beyond individuals. This entity could raise capital, engage in extremely high-risk and speculative ventures, and allow different companies to specialize in different areas and collaborate with each other.
He mentioned a chart he wanted to look up, showing the decline in the proportion of the global population living in extreme poverty and the decline in infant mortality rates throughout human history. If you were to look at the entire history of humanity, and mark the moment of the invention of the joint-stock company on that chart, the shape of the curves would significantly change after that point. He said this was a remarkably extraordinary performance of capitalism in human society.
This passage really opened up my thinking. When we usually talk about startup companies, we are always discussing products, funding, and growth, rarely taking a step back to think about the company itself as an organizational form, which is essentially a technological invention that binds the interests of a large group of people together. Thinking this way, entrepreneurship is fundamentally about utilizing this invention and then layering your own thing on top.
Believe Only a Few Things, Stay Flexible on Everything Else
Speaking about how to plan for the long term, Altman mentioned that he doesn't tend to work backward from the future to the present. His preferred approach is to first establish a few directions that he unquestionably believes in, and then step-by-step progress from the current moment, figuring out what can be done now and what can be done this year, only occasionally planning for five or ten years ahead. He mentioned having seen too many people holding a set of beliefs about the future, only to be constrained by their own rigid worldview. You might see companies that were building rockets suddenly pivoting to AI; that's the kind of situation he described. The truly effective approach is to hold steadfast to only a few deeply held beliefs and remain flexible on everything else, firmly holding onto the core.
He mentioned a friend's company's core value, called the critical path, which means always focusing on the biggest obstacle in front of you, moving it out of the way, then finding the next one, moving that, and repeating this action. He said that over the years, he has been very clear that the critical path of his life is to make intelligence abundant, as long as there is no eerie centralization of power, he believes this will bring about great prosperity. He said he is rarely tempted by other thoughts, rarely thinks about whether to change the goal, but recently started to seriously think for the first time about what comes next, if they are really close to creating superintelligence, what should be the next step.
As I read this, I felt quite touched. A person who can focus on a key path for so many years and rarely be distracted by other opportunities along the way, this itself is more challenging than any planning methodology. In the end, what matters most is not how accurate the plan is, but whether one can believe in a few things long-term and filter out all the remaining noise.
Dare to Fly at the Edge
Altman mentioned that he has always believed in one principle: in a slightly risky edge state, if it's time to get on a plane, then get on the plane. He told the story of ChatGPT's early days, when world leaders around the world were very nervous, some suspecting that this thing might be getting out of control, he could feel a storm brewing. So he listened to Brian Chesky's (Airbnb co-founder) advice and decided to embark on an intense short-term tour. Chesky had done a similar scale trip himself, roughly eight cities, but they went directly to 28 countries in 35 days. He said he basically lived on the plane, and the experience was strange. Although the flight was comfortable, traveling itself was very exhausting, the jet lag, one's own bed, one's own office, all being missed.
He also mentioned an interesting criterion for judgment, how to differentiate whether something is a real trend or a fake trend. The appearance of a fake trend is when a bunch of people hype it up, but those who actually buy the thing use it for a while and then don't like it, it does not shape their lives around it, and the thing ends up collecting dust. He used VR as an example. The appearance of a real trend is when this thing continues to appear in your daily life. Like for himself with ChatGPT, he uses it almost every day, sometimes for three hours a day, sometimes almost none, but it remains a continuous part of his life. He said this judgment method was derived from analyzing a large number of startup companies at YC (Y Combinator). If one is willing to spend time analyzing this data, one can really see many things.
I really like this criterion for judging real and fake trends because it is simple enough. There is no 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 and then get tossed aside.
Asking for the Impossible Can Sometimes Lead to Achieving It
Altman shared the example of Codex (OpenAI's programming agent application) as a memorable instance of asking for something that seemed impossible. They were significantly behind Claude Code (Anthropic's programming agent product) in the programming track, and it was generally considered impossible to turn the tables in a product category where someone else had already taken the lead. Most people would have accepted the situation and moved on to a different direction. However, they believed that this was too important to give up easily, so they assembled a team and entrusted them with this seemingly suicidal task. The team ended up achieving a rare feat in business history, and now, among the best programmers around, the most widely used programming tool is this product.
He was very straightforward in saying that if they hadn't asked for it at that time, handing over this almost impossible task to the team, none of this would have happened. His explanation was that programming is too crucial for recursive self-improvement (RSI), not to mention the economic value behind it. They couldn't convince themselves to abandon this track.
When I read this passage, I thought that asking for the impossible may sound simple, but it actually involves challenging a very strong social default. Everyone assumes that the winner has already been determined, and trying to compete means destined to lose. But what Altman did here was actually to reject this default, to first assume possibility, and then try.
Projects to Kill and Teams' Emotions to Reset
The interview touched on a heart-wrenching question: how is the decision made to unplug a project that has been running for over a year, consumed a lot of money, computing power, and people's efforts, that users are using and enjoying, but still needs to be scrapped. Altman said this is not a decision that can be made in one meeting; it is more like a slowly accumulating awareness. They gradually realize that this computing power, these people, this product direction, placed elsewhere can create greater value, and then they have to make this painful decision.
He gave two examples: when GPT-3 (OpenAI's early language model) was successfully running, they shut down a robot project that was equally exciting at the time and fully focused resources on it. More recently, after the programming agent was successfully running, they again shut down two equally promising directions, Sora (OpenAI's video generation product) and the browser, to fully dive into programming. He said this doesn't mean Sora was not doing well; if they had continued to invest, it could have been very successful. It's just that putting the computing power and energy into the programming agent was more critical at that time.
Regarding how to make the team accept this shift, he said that everyone actually understands the mission, understands the sacrifices behind it, even though it may be uncomfortable at the moment, the team is also clear in their hearts why this needs to be done. Some people may be unhappy, but more people will say, "I understand why this needs to be done, it is right for the mission."
I think the hardest part of this section is not in the decision itself, but in how to make a group of people who have already invested a year of hard work re-believe that the next thing is equally worth going all in. This requires not only judgment but also strong communication and team leadership skills.
Exerting Effort Only in One's Strengths and Outsourcing the Rest
When talking about how to become better at something, Altman mentioned Johnny Ive (former Chief Design Officer at Apple) as an example. He said the most important lesson he learned from Johnny is that truly great design is more about thoroughly researching the problem itself rather than having a sudden insight into the solution. If you are too eager to rush to the answer or if you lock yourself into a solution too early, the end product usually won't be very good.
He also openly admitted that he is not good at product development. He does not agree with the idea that only people who are deeply knowledgeable in a field can be successful. He said he knows nothing about design, but as long as he talks to Johnny for thirty minutes, he can tell that this person is truly great. His principle is, instead of forcing oneself to improve in areas where they are naturally weak, it's better to focus all energy on what they are already good at, working hard to make their strengths even stronger.
He also shared a quite personal detail. He mentioned that during the time he was working on Sora, in order to understand the product experience, he deliberately got addicted to using TikTok. At first, it was just for learning, but later he found that he really liked it. It started from ten minutes before bed, then turned into an hour, and then into three hours on a Saturday afternoon lounging on the couch. He said that it felt great at the moment, but he clearly knew that this thing was not good for him. Eventually, he turned off notifications for most apps, even notifications for messaging apps, and deleted TikTok because he felt that this thing was too powerful for him, and he couldn't control it.
When I read this part, I was quite surprised. A person who works on creating more powerful AI products every day can also be backlash by the kinds of things he creates for himself, and he had to resort to the most basic method of turning off notifications and deleting the app to get back on track. This reminded me of one thing, judgment is not something that once created will last forever. It requires continuous self-management, even for someone who understands the product design logic the best.
Some Personal Reflections
After listening to the entire interview, what I felt was pervasive throughout was not any specific methodology but rather a mindset of facing uncertainty. Believing the index will continue to rise, enduring chaos until it is no longer frightening, holding on to only a few deeply held beliefs, speaking up when necessary, letting go when needed, and focusing all energy on what one is truly good at. Each of these principles on its own is not groundbreaking, but what's difficult is being able to do all these things in an environment where all assumptions are being overturned.
Altman ended with a profound statement that left a deep impression on me. He said that most startup companies today still look very similar to those from ten years ago because they have been taught the same so-called right way to do things. At most, there has been a change in wording, such as advocating for hiring fewer people and spending more on tokens, but this is far from sufficient. I think this statement is actually a reminder to everyone that the toolkit has completely changed. If your mindset remains in the old reference frame, then no matter how radical you claim to be, the outcome is likely to still be outdated.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
