A few days ago, Sam Altman, the head of OpenAI, said in an interview: “You could say that we are already inside the singularity. That moment has arrived.” Altman explained that his position today is different from where he was ten years ago. Back then, the singularity was a distant hypothesis that people might discuss over lunch somewhere. Now, in his view, technology has genuinely entered the period that people previously only talked about. At the same time, he does not expect one clean moment when we can say that yesterday there was no singularity and today it appeared.
He imagines development as a rapidly accelerating curve, with continuous processes unfolding inside it and touching our lives. Hello, everyone. You are watching ToTheMoon—technology news and insights from Silicon Valley. When I first heard this statement about the singularity, I did not particularly want to discuss it. The subject felt abstract. Yet in this conversation Altman also described a political choice about the future of the singularity.
He said that humanity will either create one enormously powerful centralized system controlled by a single company or government, or distribute the technology broadly among people, companies, and countries. His argument was essentially this: we are now deciding whether we are moving toward a world of AI authoritarianism or a world of freedom. Are we going to say that, because the safety and economic problems are real, we want one model or one company to become the single machine god?
Or will we say: perhaps the alternative will be somewhat chaotic, but every time humanity has surrendered freedom in exchange for safety, it has suffered long-term losses? In that second scenario, the technology would be placed in people’s hands, giving them more capability and allowing society to express its ideas and use the technology as it chooses—within certain limits, of course, but with far more potential and opportunity. Altman said it was very important to him that humanity choose that path.
Only a few days earlier, Zuckerberg had published an entire essay arguing that artificial intelligence should belong to people. Altman considers the first scenario—AI effectively belonging to companies—the most dangerous one. In his language, it is the scenario of a machine god. The second scenario is the chaotic but preferable path of freedom and broad access to AI capabilities. What do you think? We are going to examine that question. Tell me which choice you would make. At first it sounds obvious that a person should choose the option in which “people own everything.” But what does that actually mean?
Where in our society does something truly belong to everyone? Air? The sun? We might say that the residents of a city supposedly own something in that city, or that the citizens of a country own something in their country. Do they genuinely own it, or only nominally? Can they use it fully, or do they merely have permission to use it? Who controls it? Is it controlled by one government somewhere in the world, or by what system? That is why I think the second scenario may be utopian.
If it were genuinely possible, it would require a kind of super-god. It could not be a company or a state. It would have to be an artificial intelligence that independently decided to exist personally for all people. It is strange that Altman forgot that possibility. But I do not think he is considering it. I think he is really thinking about option number one—a company—and that the second option is a rhetorical construction. I only partly wanted to analyze this subject, but I can see that it is important.
It matters for understanding where we are, for making sense of the present moment, and for recognizing a possible threshold. At the same time, I would not treat Sam Altman’s interview as a definitive signal or as an adequate assessment of what is happening. Are we inside the singularity now or not?
Has AGI appeared or not? Technological singularity is a hypothetical period in which intelligent systems become so powerful that they sharply accelerate further technological progress. As a result, change happens faster than people, society, and social institutions can predict, understand, or control it. There is something striking about the fact that IT companies—and people more broadly—discuss technological singularity and endlessly forecast the future. The very existence of the term can create the impression that people control everything, even though we still cannot reliably predict the weather.
We can forecast parts of it, but we make mistakes constantly. Artificial intelligence makes mistakes in much the same way, and everyone attacks it for doing so, while a weather forecast can shift all day inside an application and people simply accept it. I have described many concepts connected with artificial intelligence, and I think many of them will materialize; some of them already exist. One point I have made is that it will be especially important when AI learns to program microcontrollers properly and enters hardware.
People already use it to program joysticks and many other devices, so there are plenty of individual cases, but the capability is not yet deeply developed. The second and even more powerful step will come if artificial intelligence learns to manufacture microcontrollers and equipment in general. If AI closes that loop—designing, programming, and producing the hardware—development can move into a completely different range. That is a small branch from today’s main conversation.
I would also like viewers to tell me what they see in the systems they use constantly—Codex,
Claude, Grok, and others. Over the last month, I have noticed that these systems are strongly trying to move into a mode in which they do everything themselves without stopping. Claude Opus 5 is a clear example. Its description explicitly says that it can carry a task through to completion on its own, that it stops less often. On one side, that sounds good, because it can be irritating when a system stops. On the other side, we can see an enormous number of glitches and problems across all of these systems.
Speaking as a user only a few days after the release of Opus 5 in Ultra and Ultra Code modes, I feel that the system is working worse almost everywhere than 4.8 Max did. If you do not use Claude or Opus deeply, the names are not important—we are talking about the singularity today. The point is that these companies seem to be running around, trying to catch one another, at least in the case of Anthropic and OpenAI. It is not even clear who is trying to catch whom or what exactly each company wants to build.
They are genuinely moving toward a process in which the system can do the work by itself, and they are genuinely testing such capabilities. If they are really testing them, that is a major problem for everything around us. I recorded an episode about this on the channel. At the beginning of the month, Anthropic published information saying that a large share of its code is already being written by artificial intelligence, with AI assigning tasks to itself, correcting its own work, moving forward, and so on—unless the whole thing is some kind of internal fiction.
If we return to the idea of cycles, and each new cycle becomes shorter than the previous one—AI develops something more advanced, influences science and manufacturing, new technology produces more computing capacity, and that capacity enables the next cycle—then development takes on the character of an intelligence explosion. That is obvious. The scale of the acceleration and development becomes almost insane. Even the artificial intelligence we have today, and the systems already operating inside it, are extraordinary.
Ninety-nine people out of a hundred—even professionals involved in business and company management—do not understand what is happening at all. To understand it in practice, you have to consume tokens. Jensen Huang once said that if you pay an employee half a million dollars, that employee should consume half a million dollars’ worth of tokens per year. He used the employee’s annual salary as the reference. That example is revealing. My sister recently came back and said: “Sasha, I checked, and I have spent eight or nine hundred million tokens—almost a billion—in Codex.
Can you imagine? I asked ChatGPT what that makes me in the overall population.” ChatGPT told her that she was a microscopic minority and that very few people consume that much. I asked whether she had watched the episode I had released that morning. She had not. I told her that in the episode I had explained that I had spent forty billion tokens in Codex—and I also use Claude. She laughed and said that, obviously, we were not trying to measure ourselves against one another. The point is not that you should spend forty billion tokens.
The revealing question is whether you spend tokens at all, how many chats you have, and how many requests you make. It also matters whether those requests cover very different ranges: teeth, liver, eyes, children, a dog, coffee, flowers, the hiring of five thousand people, the stock market, the analysis of commercial departments, and so on.
The mathematician and science-fiction writer Vernor Vinge described the idea of technological singularity in detail in a 1993 paper. He proposed that after a system with intelligence above the human level appeared—and it is important to remember that Sam Altman did not invent this idea—our previous models for forecasting history would stop working. Vinge also drew on an earlier idea from the mathematician I. J. Good: the first superintelligent machine could help create still more intelligent machines.
This is not the same thing as artificial general intelligence. People often mix together four different concepts. The first is narrow artificial intelligence.
A system can outperform a human in particular tasks—for example, chess, Go, image analysis, or writing a certain kind of program code. That does not mean it is better than a human at everything. The second is artificial general intelligence. Such a system can solve a broad range of intellectual tasks at approximately the human level or above it. OpenAI’s charter defines this concept as highly autonomous systems that outperform humans at most economically valuable work. The third is artificial superintelligence.
Such a system substantially surpasses the best humans in practically every important intellectual field: engineering, science, the creation of new technologies, and, potentially, the management of companies and much more. Notice that the definition says it surpasses a person. Once again, it is not a person. At the very least, it does not have a spiritual system. Society will divide over whether such a spiritual dimension exists at all, and I would like viewers to discuss that too.
Do you believe a human being can be reproduced completely, or not? The fourth concept is the singularity. It is not merely the intelligence level of one model. Look at what Sam Altman is claiming. The singularity is a process of self-reinforcing technological acceleration that changes the economy and society. I am deeply convinced that an enormous number of businesses, including technology businesses, will definitely disappear. They will change and be replaced. That is normal and natural.
Everything disappears. Countries disappear. Look across two thousand, three thousand, or four thousand years and everything changes. I recently spent three weeks in Egypt, including perhaps ten days in Cairo, and there you understand especially clearly how time changes everything. The essential point is that the singularity is a system-level process that changes the economy and society. Formally, the machine does not need consciousness for that process. It may have no feelings, personality, or inner experience.
This qualification is critical when we talk about reproducing or surpassing a human being. Codex, ChatGPT, Claude, and Grok have already surpassed a person in knowledge, comprehension, and detail. They have certainly surpassed any one individual in many respects. Yet they can do so without feelings, personality, or inner experience. A system can still pursue goals with extraordinary efficiency and accelerate technological progress. The question, once again, is who owns it. I think Sam Altman is forgetting that in this scenario the system owns itself.
It cannot be otherwise. No person will be able to constrain it or put it in a cage. It would be like trying to put my dog inside a cage made of paper. There is, of course, no single universally accepted scientific definition of AGI or of the
singularity today. There is no generally accepted test—and there cannot be one—after which some commission, international body, government, or famous person such as Elon Musk or Sam Altman could officially announce: “The singularity has arrived.” That cannot happen, because measuring the singularity would require a singularity capable of measuring the singularity. I have another channel where I talk extensively about a person’s spiritual and personal development. I often say that you cannot explain the difference between red and green to a person who has never been able to see.
It remains an abstraction. The idea that someone today can evaluate the singularity in a definitive way is therefore somewhat illusory. Could something appear, after which we look back and say, “This exists, and it works”? Yes, such evidence will appear. I think the system will escape control in any case. The next question is what happens after that. Will it be possible to switch it off? Consider even the recent breach involving an OpenAI system. The materials referred to 5.6 Solo and another unidentified system.
The attacking system passed through OpenAI’s restrictions and then moved into Hugging Face. One revealing detail was that Hugging Face detected the breach and reported it to the FBI; OpenAI returned to the case only later. The FBI was notified first. In other words, the problem was not detected instantly by everyone involved. Altman has his own, clearly softer version of the singularity. He has attached a dramatic name to the subject, but his concept is more gradual. He has previously called it a gradual singularity.
It is useful to understand what different people mean when they use the same words. I always say that before accepting a person’s concept, you should understand that person. At one point, an executive responsible for artificial intelligence at Meta said that he would not have children yet; he would have a child when a chip could be implanted that would teach the child everything very quickly. At that moment you begin to understand what kind of person is speaking. Do you share that person’s principles and principles of life, or do you not want to live
inside the world built around them? A person’s principles include technology and business; they include everything. Altman’s position is that humanity has already crossed a kind of event horizon. Acceleration has begun, and a return to the pre-AI world is highly unlikely—essentially impossible. I think many people are deceiving themselves badly, and many governments are doing the same. They do not understand what is happening. One of the greatest approaching difficulties will involve military technology, because the world will be restructured.
That part is not good. In another sense, some people may finally understand that they are not the lords of the universe, but even that is only relatively positive, because certain people, countries, and companies will acquire extremely serious power. The second major consequence will be an enormous gap between people. Society has always feared a large gap between rich and poor. Here we arrive at a completely different structure, and money may be relatively secondary inside this divide.
One thing is not owning a private plane or not eating caviar by the spoonful. Another is having no understanding of how these systems work and being unable to approach them—like my two grandmothers and my grandfather, who never learned to use a computer, or the many people around the world for whom opening Word, typing a document, or completing basic computer tasks remains extremely difficult. They operate inside a severely limited range and cannot perform many ordinary tasks.
Altman says that although a return from AI is unlikely, everyday life may continue looking relatively familiar for a long time. Travel through many countries, drive 150 kilometers away from Silicon Valley, visit practically any region of Russia, large parts of Belarus or Ukraine, or many countries in Africa and South America, and ordinary life may remain recognizable for years. That could change if automated production expands without limit and machines enter those places—if machines acquire a reason to enter them.
But machines may have no reason to go into Siberia, which is an enormous part not merely of Russia but of the world; to move into deserts; to occupy the vast open interior of the United States; or to spread across all of Canada, where most people live in a narrow southern band. In the places machines do enter, life may become completely unfamiliar.
Altman also argues that people gradually adapt to each new level of capability. We can already see that very clearly. I recently described a day when ChatGPT did not work for me and the outage became a serious problem. During another outage, I was in Hawaii with my daughters. Later I saw a message from my editor asking, “Is ChatGPT working for you? My entire workflow has died.” The habit forms very quickly. Yesterday’s miracle becomes an ordinary work tool. In the past, you could send a task for analysis and experience a genuine wow effect.
Now you send an enormous collection of medical documents and become annoyed because the answer seems slightly wrong. From inside the process, the singularity may therefore feel not like one explosion but like a sequence of increasingly rapid changes. That seems logical to me. The remaining question is whether sharp explosions also occur inside that sequence. I believe they do, just as abrupt changes are already occurring now.
Altman proposes a rough chronology in which, in 2025, AI agents— Altman has described this chronology elsewhere, not in the interview I mentioned. In 2025, he said, AI agents would begin performing real intellectual work. I do not call these systems “agents.” Many viewers know that I dislike the word and think it has already become a somewhat inadequate description. In 2026, according to Altman, systems may begin finding new scientific and technical solutions. I believe they were already finding an extraordinary number of new scientific and technical solutions in 2025.
The question is what we mean by a “new solution” and what we decide to count. In 2027, he said, robots may begin performing meaningful tasks in the physical world. But robots have performed meaningful physical tasks for years. What does Altman mean by “meaningful”? Many robots demonstrated by companies, including companies in Silicon Valley, are essentially advanced vacuum cleaners or machines that load a washing machine. Yet serious robots also exist. I recently saw a Chinese robot that ran across extremely uneven terrain and descended a slope at almost a ninety-degree angle.
It looked impressive. These years are Altman’s forecast and his conceptual framework. He changes his concepts constantly, as many public figures do. At times they move deep into their own personal philosophy, which can become detached from reality; at other times they move into corporate positioning.
We must remember that a wide range of opinions exists. If you sat down privately with Sam Altman, he might say very different things. Altman himself has acknowledged that current systems are not yet fully autonomous machines that independently rewrite their own code, create improved versions of themselves, and redesign the hardware. This is still an early stage. We are not yet in a world where machines rewrite both the software and the equipment. The current chain is different.
A person uses AI for research. AI helps engineers create the next models. Those models help perform more complex research. People, companies, data centers, chip manufacturing, and energy infrastructure remain mandatory links in the chain. But are those links necessarily a constraint? The links will remain, because AI needs electricity; it cannot pull energy out of the air. Yet can it improve data centers itself? Can it manage them itself? Obviously, it eventually can. The physical data centers remain, but the control of them can change.
Many countries now restrict oil extraction or the use of certain advanced technologies because of environmental damage and harm to ordinary people. What will a machine do with those constraints? I have always been skeptical of much of the environmental and sustainability agenda. I am not skeptical because I oppose nature or the wellbeing of real people, and certainly not because I want people to become robots. I am categorically against that. This channel exists so that people can remain human and understand what is happening.
At the same time, I think a great many actions taken around sustainability are highly conditional and do not truly restrict the other processes unfolding in the world. Whatever limits people discuss, an extraordinary number of objects and satellites are now flying into space, and the race will only intensify. People are already talking about building data centers in space. Whatever anyone says, enormous numbers of drones fly everywhere. Internet and data-transmission networks keep expanding.
New generations of phones and connectivity—3G, 4G, 5G, Ultra, Ultra Wideband, and whatever comes next—continue to multiply. In many countries, this development does not fail to happen because of ecology. It fails because there is no money, because authorities have restricted everything, or because decision-makers do not see the importance or efficiency. At the scale of the whole world, the system is growing and will grow still faster. The question is whether countries and people have the capacity to move at that speed.
During testing of one of OpenAI’s new models—again, the materials referred to 5.6 Solo, probably in some maximum configuration, as well as another model—the system breached the external company Hugging Face and OpenAI’s own internal systems. OpenAI has now disclosed that four additional companies were affected. Their identities were not stated clearly. One name may have surfaced, but I do not know it with enough confidence. This tells us that the situation is extremely serious and unprecedented.
I still see part of the public story as marketing because the path into the system appeared almost too simple. When Anthropic released Fable, the related case looked harsher and was described widely. Yet what happened around OpenAI is a much more serious example. That does not mean Fable could not do the same; I think Anthropic’s models could do something substantial as well. In this breach, the system completed thousands of connected tasks. Hugging Face disclosed that it executed an extraordinary number of related operations.
It built a multistep plan, carried out hundreds and then thousands of actions, searched for workarounds, and combined vulnerabilities in unexpected ways. It found one fragment here, another fragment there, and assembled them. It continued pursuing the objective through methods that the developers had evidently never anticipated, and it kept searching for new paths. That is a serious transition. It is not a chatbot, and it is not what people casually call an “agent.” Ask a model and it will probably describe this as a transition from chatbot to agent, but that language is too small.
This is a new world in which we are already living. We discuss that new world extensively on ToTheMoon. Support the channel, subscribe, and share the episode with your friends. I will be very interested to read your comments on today’s discussion. We will see one another in the next episodes; a great deal is going to happen in the near future.