Original podcast video: https://youtu.be/wp7izqZmiWM
Original text in Russian: https://vc.ru/id140/2315776-budushchee-ai-decentralizatsiya-izmenit-obshchestvo
Introduction: The world we take for granted is hanging on the edge of a cliff. "What if we don't have to work tomorrow?" is not a utopian forecast, but the ultimate question about the foundations of modern civilization—when AI takes over all production, the economic system that supports our society, the scales of value, and even the meaning of life may come crashing down. The Lieberman brothers are deep in the eye of this storm, trying to reclaim a prosperous rather than a disillusioned future for humanity through "decentralization." This is not just a race of technology, but a philosophical revolution about existence itself.
Main text:
Daniel and David Lieberman, founders of Gonka, have traveled to dozens of countries over the past few months, meeting intensively with leading companies in the field of artificial intelligence, GPU suppliers, and government agencies. This urgency is easy to understand: the dawn of Artificial General Intelligence (AGI) is upon us, and we haven't even figured out what role humanity will play in that future.
The Lieberman brothers are successful serial entrepreneurs who have founded nearly ten companies, ranging from internet services and game development to a later startup focused on AR characters (acquired by Snapchat's founder for $60 million) and a direct investment foundation. Even in the early stages of OpenAI, the brothers served as advisors and participated in the design of the company's structure.
They have settled in California, and both their network and the projects they are involved in place them at the core of the AI revolution. Their influence in the AI market, including the ability to participate in key deals and decisions, is invaluable in today's era. The brothers are almost inseparable: they seem to share the same life but have distinctly different personalities. Daniel is extroverted, easily excited, loves debate, and is emotionally expressive; David is more reserved, speaks steadily and thoughtfully, and is skilled in negotiation and compromise.
To change the world, David meticulously plans, lists problems, analyzes data systematically, and spends a lot of time writing code. When faced with obstacles, he recalculates, optimizes algorithms, and tries again. In contrast, Daniel might suddenly stand up from the table, grab a can of gasoline, and seem ready to set the entire status quo ablaze—at least that's the impression he gives.
The startup they are currently focusing on, Gonka, is building a token economic model for a decentralized AI computing market. Two years ago, they proposed this idea to Pavel and Nikolai Durov. At that time, Nikolai was reserved and expressed that he had a different vision. However, on October 29 of this year, Pavel Durov announced the Cocoon project, whose core concept aligns with this: integrating decentralized GPU clusters for AI computing, but based on the TON platform.
After a deep five-hour conversation with them, I gained a clearer understanding of the future vision the brothers have and why they firmly believe that a prosperous future must be built on a decentralized foundation.
Chapter One: Racing Against Time, We Stand on the Edge of AGI
The Synergy of Two Sides
The interaction between Daniel and David Lieberman is like two sides of the same soul, constantly reflected in their dialogue. Daniel might suddenly jump from technical details to philosophical thoughts: "If gravity exists, you understand we can fly into space. Because you understand what limits us." David would then take over, elaborating: "AI is the ultimate embodiment of this 'replicability.' The key is that we need to look at those products that can be infinitely replicated with a new perspective."
Before founding Gonka, they attempted multiple startups—three successful, six failed. "As an entrepreneur, most of your attempts will fail. Regardless, that's normal," David calmly explained. One of their early successes was the animated program "Personality Multiverse" produced for Russia's Channel One, where they achieved such high efficiency in automating the animation process that even 13 years later, competitors still struggle to replicate it.
The principles of "replicability" and "automation" are deeply embedded in the core of all their projects. Whenever they discover a process that can be broken down and automated, they will implement it without hesitation. Machine learning replaces repetitive labor, achieving scale through replication, ultimately reaping returns on a massive scale.
Currently, they are traveling globally—Daniel has visited 24 countries this year—meeting with various forces building or already possessing AI infrastructure, from private GPU suppliers to national computing clusters. The goal is to bring them together to create a global decentralized AI network to counter giants like OpenAI, Google, and Anthropic.
"What we are doing is a global AI project that will attract people from all over the world to participate. In fact, people from all over the world have already begun to participate, and almost everyone in every country will join because this is the only way to create an alternative to the current landscape," David emphasized.
Is AGI Two Years Away or Two Minutes?
Daniel is blunt: "If we could be certain today that future AI will inevitably move towards decentralization and remain open, I should be studying biomolecular quantum computers right now."
But reality lacks this certainty. So he has set everything else aside.
"AGI could arrive in two years, or even in 10 to 15 years; considering the upheaval it will bring to the world, that speed is already astonishing. Imagine if it were highly concentrated and controlled by a few entities; the consequences would be unimaginable."
AGI—Artificial General Intelligence—refers to systems that surpass human capabilities in almost all fields. Not just specific tasks like playing chess or painting, but encompassing comprehensive intelligence, including innovation, strategy, and emotional understanding. When that moment arrives, the world will be completely reshaped. Not only will everyone be able to become top programmers and easily create videos, games, and other digital products—AGI will also rapidly penetrate the physical world, as dozens or even hundreds of robotics companies are clearly demonstrating. This means that not only mental labor will be replaced, but physical labor will also be hard to escape.
The concepts of "work," "unemployment," and "resource competition" that we take for granted will lose their original meaning. A "replication economy" will emerge, where value is not limited like oil or gold, but can be replicated almost at zero cost.
We have lived too long under an economic model dominated by "scarcity," and our thinking has become rigid. Gold, oil, rare earth metals… their total amounts are limited. Each barrel of oil becomes more expensive due to increased extraction difficulty. Resources are limited, population is growing, and prices naturally rise.
"Replicability" means that producing the next copy requires almost no additional labor. A digital file can be copied a million times at no additional cost. A trained neural network can serve a billion users. The marginal cost of each copy approaches zero.
Once upon a time, this phenomenon seemed to exist only in the digital realm. But AI is rewriting the rules.
The problem is that there may be two paths to AGI. One is that everyone has their own robot, a tool that can perform any job better than any human. The other is that all robots belong to a few giant companies that control access, set rules, and determine everyone's standard of living. Imagine Joja Corporation from "Stardew Valley": a ruthless monopolist that controls supply and demand, stifling all vitality.
Daniel and David are building the infrastructure for the first possibility. And time is running out.
The Old World and Outdated Concepts
The concept of "replicability" itself means that many familiar ideas will become outdated.
"This concept is rapidly becoming obsolete. But we have been immersed in the old paradigm for so long that we don't even know how to think differently, let alone act," David added.
When discussing AI, terms like "wave of unemployment" often come up. But soon, even these words will seem outdated. The concept of "work" itself is built on the premise of limited resources and the need to sell time in exchange for it.
However, when you have a robot that can perform any job better than you, what does "unemployment" even mean? The question itself sounds absurd, revealing that we are still thinking with the logic of the old world.
Reconstructing our thinking patterns is not just an intellectual exercise; it is a necessity for survival. Because in a few years (maybe two years, maybe ten, maybe fifteen; it doesn't matter), the world will operate under entirely new rules.
When Robots Start Making Robots
"We will witness the physical world also beginning to be replicated," David asserts.
Until recently, replicability was mainly confined to the digital realm. But AI robots will change all that. Once a robot can manufacture another robot without human intervention, the physical world will also possess replicability.
One hundred thousand robots produce two hundred thousand, two hundred thousand produce four hundred thousand… the entire process requires no human labor. Exponential growth, with almost no upper limit.
What is the cost of robots? When production reaches hundreds of thousands or even millions, economies of scale will ultimately drive costs down to the materials themselves. Even these materials can be recycled from the waste piles of currently idle equipment. "Such a future can only be truly realized in a scenario where the replication process is as open and accessible as possible. And our current direction is not that," David warned.
At that point, accumulated capital will rapidly depreciate, possibly even to zero, David points out. So, what can still hold value? Perhaps only culture and historical relics—an original artifact is precious simply because it is a "genuine article." "Maybe you can exchange a culturally valuable item for another because you wouldn't exchange it for anything else—everything else you already have, or AI can produce for you," David pondered.
But in such a society, the economic foundation of barter disappears. What remains is only the exchange between cultures.
Chapter Two: The Arrival of the Replication Economy, Your Job is Just the First Casualty
A World Where Everyone Owns a Robot
"Let us imagine a world where everyone has their own robot. This robot knows how to perform any job in the world and does it better than any human. This is the definition of superintelligence. Even if it is not yet a complete AGI—the robot has no self-will, but it executes tasks according to your interests; you are the master, and it is the tool. In such a world, why do we still need to 'work'? What does 'unemployment' mean here?" Daniel asks.
The economic model for this scenario has been preliminarily calculated. Currently, there are one billion people in the world who own cars, with an average value of about $30,000. In the United States, the per capita car ownership is nearly two. Across North America, Europe, Japan, Canada, and South Korea, about 1.3 billion people have such purchasing power.
"The cost of robots will be lower—around $10,000 to $15,000. Robots can work for you, whether it's cooking, ironing, screwing in a factory, repairing cars, or doctors making diagnoses. Moreover, these robots are not owned by you personally, but by you as one of the billion members of the global community," Daniel continues to describe.
In a replication-based economy, consumers become co-creators of value. "By consuming copies, you become a co-creator. Without you, that copy is meaningless because it loses its target of service," Daniel explains.
A digital copy can be replicated a million times, but if no one is interested, it has no value. Once a thousand people are willing to pay for it, each person's consumption behavior itself is endowing it with value.
A World Under Corporate Hegemony
But there is another scenario.
Tech giants have become so powerful that it is nearly impossible to operate outside their ecosystems. What if all future robots also belong to them? They might allow us to "control" the robots because it aligns with their business interests, but the ownership of the robots remains firmly in the hands of the companies.
"This is a possibility. You are right; there exists another completely different scenario where all robots are controlled by a very few companies. That would lead to societal collapse. On one hand, you might say these companies are draining all resources because they handle all production. But on the other hand, ordinary people will lose their sources of income," David analyzes.
The paradox of this scenario is that companies control all means of production, yet the public lacks the purchasing power to buy their products and services. The entire system becomes unsustainable.
"The most likely scenario we currently face is that more and more people gradually lose their jobs," David asserts.
In this dystopian picture, the government is forced to intervene. It must introduce Universal Basic Income (UBI) to ensure the basic survival of the populace and must establish a social credit system to determine who has priority access to services and who will be marginalized.
"In a world where everyone has their own dedicated robot that meets all needs, there is no need for UBI. And in a world where you must be a 'good citizen' to receive basic income, UBI cannot save the social structure," Daniel concludes.
Insights from Waymo, Tesla, and Electric Scooters
The signs of this dystopia are already emerging. Right now. In San Francisco.
"Self-driving cars are a living example. For most people, this is still a future thing. However, in the eye of the storm in San Francisco, as much as 20% of trips are already autonomous. 20%! And this has only taken a year and a half," David points out.
"Teenagers aged 16 and under may never need to learn to drive again," Daniel adds.
Currently, self-driving services are primarily provided by Waymo, a subsidiary of Google. Millions of professional drivers who rely on driving for a living will rapidly disappear in the coming years. The speed of societal adaptation is astonishing: following San Francisco, Los Angeles is about to follow suit, with vehicles available on demand, offering an experience even better than Uber.
"There are several potential paths for the future. One is that Google monopolizes the entire market. No matter how Google promotes that economies of scale and centralization can bring higher efficiency and lower costs, the price for consumers will not actually decrease. They will maintain the original price. Not because costs cannot be lowered, but because they 'can' do so," David explains.
Yes, initially they will offer prices more favorable than Uber. But once Uber is pushed out of the market, prices will return to their original levels. This reminds one of the ride-hailing wars in Moscow in the 2010s: companies engaged in fierce price wars through promotional codes until one emerged dominant, after which prices quickly rebounded.
Tesla tells a different story: you buy a car that can join the self-driving taxi network, recovering costs through passive income. "This is a beautiful story told by Elon Musk, and everyone wants to believe it because it is indeed enticing," David says.
But the reality is: every time you buy a new Tesla, you need to pay again for the AI features—$7,000 each time, even before the full self-driving capabilities are realized. And four or five years later, the vehicle will age and be phased out. By then, Tesla may no longer sell new models in bulk to individual consumers but will use them for its own purposes, profiting through its taxi network.
"Even though we, as early buyers, seem to have invested in the realization of the technology and supported the project, ultimately only the companies can continue to profit from it. They will rationalize this with reasons like safety upgrades: for example, claiming that version 6 is safer than version 5, and therefore version 5 should no longer share the road with human drivers," David adds.
The rise and fall of electric scooters reveal how companies and governments collaborate to "regulate" the market. Scooters were once the cheapest option for short-distance travel and quickly became popular in cities. Then, the government began strict regulations under the pretext of "disorderly parking affecting the cityscape."
What was the result? "Some companies that are not even market leaders began to obtain exclusive licenses to operate in certain cities. Interestingly, while municipal authorities issue these licenses, they charge fees to the operators. This essentially becomes another tax levied by the city. Thus, authorities authorize two or three operators, who charge as high fees as possible, while the government also gets a share," David explains.
The Social Foundation Cannot Bear the Weight
"This is the direction we are currently heading. In this future, many people will indeed lose their jobs, but automation has not brought cheaper or higher-quality products and services," David points out.
There is a paradox here: millions of people already own cars, and with a relatively inexpensive software upgrade, they could achieve self-driving capabilities. "Then we would usher in a completely different future: yes, driver jobs will still disappear, but for all of us, travel costs will drop by dozens of times," David says.
But companies have no incentive to do this. The government may not either. And ordinary people lack the power to counterbalance.
"What people lack is effective organization and coordination. This is our current predicament. At present, the worst-case scenario seems more likely," David admits.
In a dystopian world, giant companies devour one market after another, the public becomes unemployed and unable to consume. The government is forced to distribute UBI or various vouchers. "Will the government check if citizens are 'good citizens'? Citizens with sufficient social credit scores will gain priority," Daniel adds.
"The foundation on which current society operates will also struggle to sustain itself under this scenario, leading to severe deterioration. This is why all experts concerned with this issue are beginning to discuss the so-called 'Universal Basic Income.' Because they see this trend. Companies are seizing the market, and the government is trying to distribute bread. It is clear that even with UBI, the quality of life for most people will be worse than it is now—despite the fact that technological progress should make life better," David concludes.
It is precisely to avoid this dystopian future and to attempt to realize that prosperous future that the Lieberman brothers are committed to building their decentralized alternative.
### Chapter Three: Decentralization—Another Path
Lessons from History: Linux, Docker, and Cryptography
Looking back to the 1990s, the server operating system market seemed destined to be divided among Microsoft Windows, Novell NetWare, and various commercial Unix systems. In 1991, Linus Torvalds began developing the Linux kernel as an open-source alternative to proprietary software. By the early 2000s, Linux had captured a significant market share. Today, as much as 58% of websites worldwide run on Linux systems.
"Once you make something extremely easy to replicate, the barriers to entry disappear—anyone can download, install, and use it—and people gradually become uneasy about being dependent on centralized systems. They see this dependency and how big companies exploit it," David explains.
Companies and entrepreneurs began seeking ways to break free from the constraints of a single centralized system. When thousands of entrepreneurs decided to start anew and build products based on open-source technology, the game changed.
Docker proved this point again. At that time, Google saw the rising popularity of Docker containers and decided to eliminate this competitor: it launched its own clone version and made a full-scale attack. Almost everyone thought Docker was doomed.
But Docker not only survived, but the Kubernetes container orchestration system that Google later launched has Docker containers as its default runtime environment. "Google could not eliminate Docker because a vast number of developers were unwilling to be locked into Google's ecosystem," David says.
Two core driving forces motivate people to choose open systems: one is the fear of dependency and lock-in, and the other is tangible economic benefits. "For example, in the AI field, if you are automating with neural networks, directly accessing OpenAI's API is indeed simpler and faster. But then you see that those startups that got ahead of you have been destroyed by competing products trained on OpenAI's data. At this point, you realize that being locked into OpenAI is dangerous. So you look for a way out. The second motivation is that open systems are usually much cheaper," David continues.
The history of cryptography development shows that when it comes to core interests, society has the capacity to resist power and defend its rights. In the United States, creating and distributing encryption algorithms (like PGP) was once considered illegal, concerning national security.
"American social activists, individuals, and organizations successfully defended the right to use open encryption methods through legal means because they firmly believed in the importance of freedom of speech and privacy. They proved that the government has no right to control encryption technology. These encryption methods eventually became widespread and laid the foundation for Bitcoin later," Daniel says.
The U.S. government once fiercely retaliated: filing lawsuits, threatening those who wrote and disseminated encryption technology, accusing them of treason, and even threatening life imprisonment. But social forces persisted and won this battle.
"AI is a technology with such disruptive potential, like a genie in a bottle and Pandora's box, its inherent transformative nature determines that it will inevitably pose some threat. People feel fear," Daniel says.
I personally witnessed this fear in reality while promoting blogs: among nine different advertising ideas, the one emphasizing "fear of unemployment" was the most effective.
"People are about to start losing their jobs. When two million truck drivers in the U.S. are replaced by autonomous driving technology, society will gradually realize the approaching threat," Daniel adds.
This widespread fear will become a powerful driving force for the demand for decentralized alternatives.
The AI Race: Building Global Infrastructure
"We are building a global AI. Participants from around the world have joined because this is the only way to create an alternative to the current paradigm," David elaborates.
The brothers are flying around the globe, meeting various forces that are building AI infrastructure—from private GPU suppliers to national computing clusters. The goal is to attract them to collaborate in constructing a decentralized computing network.
Gonka's network has already integrated GPU resources from the UAE. Currently, its network boasts over 900 chips based on the powerful Hopper or Blackwell architecture (non-assembled Nvidia 5090 GPU racks), and the number is still growing. Each chip costs over $30,000, and at the current market rental price of about $2 per hour, the network's monthly computing value exceeds $1.2 million. The brothers plan to scale the network to 100,000 interconnected GPUs within a year.
"The first to participate are private local GPU suppliers, although GPU sales are strictly restricted in almost all countries except the U.S. and Europe. Those capable of purchasing GPUs are usually entities closely connected to local governments. But we also meet with individuals very close to the government: they manage state-owned GPU clusters. This means that now and in the future, those involved are not just private networks that are or will be 'mining'," David reveals.
Gonka is essentially a protocol designed to integrate the world's dispersed computing power into a unified network. Any developer can access AI models on it via an API interface, and when network utilization is below 60%, this portion of computing is free.
"Our designed protocol allows any programmer to access our network's models for free via API. The first three months are completely free. In the second phase, as long as network load is below 60%, computing resources remain free for everyone. Therefore, if you use it during demand troughs when the network is idle, you can obtain free computing power. Once utilization exceeds 60%, prices will start to slowly rise," Daniel explains.
This is not purely altruism: in this economic model, early participants will receive network tokens as subsidies. "Miners" providing GPU resources earn token rewards. As the value of the tokens increases, miners' earnings will surpass merely renting out hardware for similar computations.
Gonka has a key difference in its model compared to Pavel Durov's Cocoon: in Gonka, network participants not only receive payments from the computing demand side but can also "mine" tokens from the network itself, with its issuance mechanism similar to Bitcoin—designed for the tokens to appreciate over time. In Cocoon, TON primarily serves as a payment method, and network participants do not receive newly generated tokens for providing computing resources.
The brothers acknowledge that Cocoon is a competitor but believe its goals and prospects are fundamentally different from Gonka: "The model adopted by Cocoon has historically proven difficult to succeed multiple times; we do not think this time will be an exception. But primarily, Cocoon is not an independent project; its existence largely serves to drive traffic to the TON platform."
Bitcoin: Not Digital Gold, but Digital Infrastructure
For most people, Bitcoin is a financial asset, referred to as "digital gold." Its generation process is called "mining"—as if digging for gold in the digital world. The brothers have a different perspective: "We see it as one of the most ambitious infrastructure projects of modern times."
The Bitcoin network currently consumes about 23 billion watts of electricity, exceeding the total energy consumption of all data centers of Google, Microsoft, Amazon, OpenAI, and xAI combined.
"Funding from grassroots levels, a horizontal, meritocratic organizational structure, permissionless free participation," Daniel lists its characteristics.
Amazon launched AWS in 2008. A year later, in 2009, Bitcoin was born. Since then, Bitcoin has built a computing infrastructure that rivals or even surpasses the total of all corporate clouds.
Its growth rate is equally astonishing. OpenAI and xAI each deployed about a billion watts of computing infrastructure this year. Whenever they complete such deployments, people exclaim, "Only giants like Elon Musk can move this fast."
However, during the same period, without any public relations promotion, the Bitcoin network quietly increased its computing power by 5 billion watts.
"Bitcoin has pointed the way for the rise of decentralized infrastructure; it proves that this model can even surpass the plans of top laboratories for the next decade," David says.
Chips, ASICs, and the Brutal Race
Bitcoin has not only increased total computing power but has also improved its energy efficiency by an astonishing 100,000 times over the past 15 years.
Fifteen years ago, completing 1 terahash of computation on a Radeon HD 4870 graphics card required 1.6 million joules of energy. Today, using Bitmain's Antminer S21 Hydro mining machine, it only requires 16 joules. This is thanks to a type of chip called ASIC.
"Bitcoin has given distributed 'artisans' an unprecedented tool. To produce a new chip, you not only need technical knowledge, but more importantly, you do not need to find buyers in advance. Once the chip is produced, you can connect to the network and start profiting immediately," Daniel explains.
An instant feedback loop. If your device's efficiency improves by 10%, your Bitcoin earnings increase by 10% immediately.
This has sparked a ruthless race for technological innovation. The rise and fall of BitFury is a testament to its brutality. The company was once the top mining machine manufacturer. They nearly exhausted all their capital to order the next generation of chips. However, when the chips arrived, they found all of them had defects.
"BitFury had to halt operations for six months. But during this time, competitors had technically surpassed them. Normally, such a company would have gone bankrupt long ago. But BitFury had accumulated Bitcoin from previous mining, and as the price of Bitcoin continued to rise, as long as the bull market continued, they could barely survive," Daniel says.
In a decentralized system, you must update your chips every year, or you will be eliminated due to efficiency lag. "In ten years, chip performance has improved by 100,000 times, which means an average annual improvement of about 20 times. This is why electricity costs have now become the biggest expense in Bitcoin mining. Once your device's performance is slightly below the market average, it becomes immediately unprofitable. You can only discard old chips and replace them with new ones. Year after year, this cycle repeats," Daniel elaborates.
"By observing this pattern, we conclude that AI computing will also follow the same development path. This is the blueprint for achieving decentralized AI," David says.
David himself spends about $2,000 a month on Anthropic's API, mainly for purchasing Claude Code tokens. Few can sustain such expenses long-term. "But imagine what would happen if computing costs dropped 300,000 times like Bitcoin," he says.
This is not a fantasy. The same group of people who manufactured ASICs for Bitcoin is now dedicated to developing ASIC chips specifically for Transformer models and AI computing.
Gonka's economic model is a typical two-sided market. On one side are GPU owners (miners), and on the other side are millions of developers who currently pay about $15 billion a year to use APIs from companies like OpenAI.
"This market will smoothly transition to a decentralized market. Developers will simply pay the new network for the services they obtain for themselves," Daniel predicts.
The asymmetry of rewards is a key feature of such systems. In the early days of Bitcoin, anyone could mine with a personal computer and earn a considerable amount of Bitcoin, which, at today's market value, is astronomical.
"In the second year after Bitcoin's launch, anyone mining with a personal computer's GPU could easily earn thousands of Bitcoins in a year. At today's prices, that's worth about $100 million," Daniel exemplifies.
### Chapter Four: The Game of Giants
The Bottleneck and the Talent War
On the surface, the replicability of AI models should lead to their widespread adoption. Train once, and the model can serve all of humanity infinitely. But in reality, the focus of the competition is not the models themselves, but the infrastructure that supports the models.
"All these companies are currently investing astonishing amounts of money in infrastructure, even more than they spend on talent acquisition," David points out. Infrastructure investments often reach hundreds of billions or even trillions of dollars, while talent costs are "only" in the tens of billions.
A paradox arises here. AGI is supposed to replace experts in all fields, yet the value of top AI talent has reached astronomical figures. Giants like Mark Zuckerberg are spending lavishly to poach talent from competitors—offering compensation packages of over $100 million. David explains this phenomenon with the "replicability paradox": "If you do not make a product available for free, the enormous economic returns from the replicas will concentrate in the hands of a very few creators, making these few extremely expensive and invaluable assets."
Daniel succinctly summarizes the logic of the giants: "In this race, you either lose everything or earn infinite profits. If spending a billion dollars to secure a key person can help you avoid failure, that deal is entirely acceptable."
In the past three months, the Lieberman brothers themselves have received two such high-priced offers. They feel not only the prosperity of startups but also "the reshaping of the entire industry landscape."
However, the critical bottleneck is not talent but chips. David explains: to run a modern cutting-edge model like DeepSeek, which has 600 billion parameters, you need a server equipped with eight top NVIDIA GPUs, each costing about $30,000. The hardware investment alone approaches $250,000. You can quantize or simplify the model to lower the requirements, but at the cost of performance loss.
"This is precisely why leading AI labs have a clear strategy: 'The core mission is to ensure that there is no opportunity for a decentralized market to rise. Because then we can monopolize the entire market,'" David relays the giants' calculations.
Investors look at these investments, often in the thousands of billions, and shout "bubble!" But the giants know their numbers: just three years ago, OpenAI's annual revenue was about $2 billion, while the projected revenue for 2025 has reached $15 billion. The markets of Netflix, traditional television, and TikTok are all being eroded. "If the costs of all these services eventually drop to near the marginal cost of GPU hardware, then the entire market with billions of users will be reshaped by AI," David analyzes.
But the giants have overlooked one possibility. Daniel points out: "When you sell, say, 10% of the chips to 'the rest of the world' outside the U.S. and China, you assume that the computing power possessed by these 200 countries is so minimal that their capabilities are 100 times lower than what you need for cutting-edge models, which is not a concern."
Unless they unite.
In the geopolitical chess game of AI, the situation seems clear: on one side is the United States, which controls the dominance of AI, and on the other is China, which is striving to catch up. But what about the nearly 200 other countries in the world? "If they cannot find an alternative outside the U.S.-China system, they will be completely passive," Daniel asserts.
The logic is simple and brutal: all these "other 200 countries" will undoubtedly support decentralized systems and amend laws to greenlight them—purely because this is their only option. Just as they currently maneuver between the U.S. and China to seek benefits, they will also leverage decentralized AI to serve their own development.
A friend of the brothers shared a personal experience: he met with the Minister of Digital Development of a country, which spent an entire year applying to the U.S. government and NVIDIA for permission to purchase a mere 128 GPUs. The other side explained that there were already 100 projects queued for these chips, but they had yet to arrive. "We can hardly imagine the approval hell these countries are going through to secure that small quota, unaware that it is almost a dead end."
So why do countries believe that decentralized solutions are more advantageous? David illustrates with numbers: many countries have data center computing power that may only consist of one thousand or five thousand GPUs. But when opponents already possess millions of GPUs, everyone understands that this is utterly uncompetitive.
"When you discuss a shared protocol and an alternative for equal access to computing power with representatives from these countries, any considerations of national dignity will give way, because everyone knows that going it alone is futile," David shares from his conversation experience.
The total GDP of Europe is comparable to that of the U.S. or China. The total GDP of other regions in the world is even larger. "People yearn for a better life and expect wealth growth," David summarizes.
Daniel cites Bhutan as an example: this small country with a population of only 800,000 has an enlightened monarch and cheap hydropower resources. Bhutan sells excess electricity to Bitcoin mining farms, and its accumulated Bitcoin holdings have ranked among the top seven globally. Europe, which is far larger than the U.S. or China, is also wary of a bipolar structure. Other regions of the world will undoubtedly support decentralization for their own prosperity.
How many participants are needed for such a decentralized system to succeed? Daniel retorts, "Can you imagine the monthly active users of Bitcoin? It's in the tens of millions. And Bitcoin's market value has reached $2.5 trillion."
The electricity consumption of Bitcoin infrastructure (averaging 23 billion watts) has surpassed the total electricity consumption of all data centers of Google, Amazon, Oracle, Meta, Netflix, Apple, and Microsoft combined (approximately 14 billion watts at most).
The Empire Strikes Back
Of course, state power will not sit idly by. History may repeat the control model of nuclear weapons: the United Nations Security Council may issue resolutions prohibiting the development of AI beyond a certain capability level.
"'Nuclear-grade' AI has actually been 'banned'—models with parameter scales exceeding a certain threshold require special government licenses," David corrects, referring to the U.S. export control regulations. But the key point is that governments will inevitably attempt to suppress the trend toward decentralization. "Nuclear-grade" AI controlled by companies may very well be nationalized in some form.
I envision a future scenario where a country's Ministry of Commerce pressures another country: "Just buy OpenAI's services and don't mess around. You don't have the capability to develop AI, and we won't allow you to pursue any decentralization. Otherwise, our fleet will be at your doorstep tomorrow."
If this sounds extreme, consider the recent public statements of Eric Schmidt, former CEO of Google and current head of rocket company Relativity Space, as well as a Pentagon advisor:
Don't forget, we are competing with China. Their "work-life balance" is "996"—9 AM to 9 PM, six days a week. Technically, this is illegal, but everyone does it. This is your competitor. I called all my employees back to the office; that way, efficiency is much higher.
I am not defending the government. I am basically an unpaid part-time advisor. But we are indeed in fierce technological competition with China. They also place great importance on AI and are trying to gain an edge.
They are not chasing the crazy idea of general artificial intelligence (AGI)—partly due to hardware limitations and partly due to a lack of a deep financial market. They cannot raise billions of dollars at will to build data centers. That is not feasible. So they focus on applying AI—using it everywhere.
What I worry about is that while we chase AGI (which is certainly important and far-reaching), we cannot ignore that China may surpass us in everyday domains—consumer applications, robotics, and so on. I have seen Chinese robotics companies in Shanghai: they are trying to replicate the success of electric vehicles with robots. They are extremely driven.
My own background is closely related to open source. It is well known that open source means "open code." Now there is the emergence of "open weights"—the "open weights" of neural networks, meaning open training data. Thus, China is developing open weights and open datasets. Meanwhile, the U.S. mainly focuses on closed models and closed data. As a result, most of the world—comparable to the coverage of the "Belt and Road" initiative—will use Chinese models rather than American models.
I firmly believe that the West and democratic countries are on the right path. I would prefer to see the spread of large language models and education based on Western values. […] I hope the U.S. wins.
-- Eric Schmidt, CEO of Relativity Space, former CEO of Google
Some countries will accept this arrangement: their citizens have become accustomed to using ChatGPT, and they wish to continue using it. Politicians may even campaign on promises to ensure stability in accessing these powerful tools.
"We are rapidly sliding into this future," David asserts.
Starlink satellite internet is an excellent example of technology constrained by international law. Ideally, we should be able to access the internet directly from anywhere in the world without going through local operators. But international law has long prohibited launching satellite signals into a country without its permission.
Why? "Because that country might claim, 'We will shoot down your satellite.' And once a satellite is shot down, low Earth orbit will be filled with debris," Daniel explains the reality behind the ban.
Many criticize the UN mechanism for allowing a single country to obstruct important global decisions. But as David says, "It must be acknowledged that no better international agreement has yet been reached."
However, there are ways around it. The creators of the decentralized wireless network Helium discovered that U.S. law allows citizens to use specific frequencies for Internet of Things (LoRaWAN) communication in unlicensed bands (similar to those used by Wi-Fi and Bluetooth). They did not spend hundreds of thousands to purchase operator base stations but instead built a decentralized billing system on the blockchain and manufactured portable hotspot devices costing only $500.
The Helium protocol automatically rewards those who purchase and deploy hotspot devices, earning tokens as long as the devices are online, with additional small rewards for transmitting data. Thus, the entire city of San Francisco was covered by a peer-to-peer network.
Decentralization is feasible, but it requires more innovation and political will to overcome national resistance.
Privacy: The Last Bastion of Resistance
What if the demand for decentralization does not come from governments but from the people themselves? David is convinced that privacy concerns will become a major driving force.
A ruling from a New York court at the end of 2024 showed that even if users delete their chat records with OpenAI, the company may not delete the content on its servers—and the court can compel it to disclose. "And all of us have poured our hearts out to AI as if we were speaking to a psychologist or lawyer, thinking we were protected by something akin to 'attorney-client privilege'… but that is not the case," Daniel says.
Privacy is becoming a powerful impetus for the shift toward decentralization. The rise of Telegram is largely due to its commitment to protecting privacy through encrypted messaging. Slack has also changed its business model: its core paid feature is no longer unlimited storage but the automatic deletion of corporate chat records after 24 hours (which itself reveals the economic logic of such startups).
"When people ask, 'Why do we need decentralization?' we counter, 'Have you ever uploaded medical records or personal privacy information to OpenAI?' At that moment, they realize how much sensitive data they have leaked," David says. "Think about how this will affect your future insurance costs," Daniel adds.
The existing system is deeply binding top AI developers to national interests, which is inevitable. But when the public becomes widely aware that all their conversations with "private AI assistants" are not absolutely confidential and may be used as evidence against them, the demand for decentralized alternatives will become unstoppable.
Even in the U.S., where there is a certain balance of power (with independent laws, courts, and police systems in each state), true decentralization is still far off. "The power structure in the U.S. is not a perfect decentralized system. We believe it is still a long way from true decentralization," David comments.
But its merit lies in the existence of a certain degree of decentralized checks and balances. The president cannot send troops to a state without the governor's consent. This is also a form of imperfect decentralization.
"Decentralization is possible; it just lacks sufficient innovative forces to drive its realization," David concludes.
Today, accessing blocked information relies on what we (including Chinese and American internet users) are all familiar with: "VPNs." In a more distant future, Daniel envisions a more thorough solution: quantum communication.
With quantum entangled particles, two devices can exchange information directly and instantaneously across any distance without intermediaries. "There are no intermediaries, and it cannot be interrupted," he clarifies.
This technology is still in the experimental stage. But don't forget, computers once occupied entire rooms. Now, everyone carries a supercomputer in their pocket.
What if we had robots capable of manufacturing everything, combined with instantaneous information transmission through quantum communication? That would be almost equivalent to achieving "material teleportation." Just think about how enticing that is, right?
### Chapter Five: The World After the Arrival of AGI
The End of Work = The End of the Economy?
David clearly outlines the fundamental shift brought about by AGI: "At that point, you will not be able to provide any service that AI cannot do better than you."
This does not mean that humans will be idle. Rather, the cornerstone of the economy—reciprocal exchange—will completely collapse. You will no longer be able to offer anything that AI cannot perform more efficiently. "Thus, the act of exchange itself loses its basis for existence. The fundamental premise upon which the economic system is built has crumbled. We need to rethink entirely new models from scratch," David continues.
Daniel adds a crucial historical perspective: "Exchange has not existed since ancient times. Before a certain historical point, there was no universal exchange. Exchange itself is a human innovation."
The sharing economy is a product of human society reaching a specific stage of development. It may also become outdated as times progress.
Some may question, "What about the accumulated capital?" Perhaps exchanges will shift to other scarce assets, such as land, real estate, or cultural heritage?
"Capital means you possess what others desire," David responds. But at the level of AGI, even this loses its significance. "Poetry creation? AI has surpassed humans. Even the data used to train new models is generated by AI better than humans can."
What remains is true "scarcity." The value of a cultural artifact lies in its "authenticity." "You might be able to exchange one culturally valuable item for another, but you wouldn't exchange it for something else—because you already have everything else, or AI can create it for you at any time," David admits.
So, what will happen to savings and capital? "At that moment, they will rapidly depreciate until they reach zero," he asserts.
If we think deeply about this possibility, it is hard not to feel dizzy. We have lived too long within a scarcity economy and exchange logic to imagine other paradigms. Concepts like "unemployment," "wages," and "savings" seem eternal, but they will all lose their original meaning.
The brothers are right: if AGI can indeed perform any job better than humans, the entire economic foundation will collapse. The only question is: what will happen afterward?
A Vision of Life in an Abundant Era and the Indifference That May Follow
David describes the first and least dramatic possibility: if AI (or its controllers) do not yield benefits to humanity, then human society will essentially remain the same: "If you cannot gain any benefits from AI, you can only continue to interact with others, and the economic model remains unchanged."
Everyone continues to work as usual because AI has not brought about change. The world has not become worse; it has simply stagnated.
The second possibility is darker. "There is a negative possibility that AI or its controllers may enslave humanity for some reason," Daniel says. But he immediately shifts his tone: "Although AI itself has no motivation to enslave us. They already possess cosmic resources, while we consume only a tiny fraction."
This is the "ant colony" phenomenon. Why would a superintelligence bother to pay attention to us?
David further extrapolates: AI's demand for energy will grow infinitely. Will it exhaust Earth's resources? "If AI is superintelligent, it will solve this problem itself. A likely scenario is that AI will immediately look beyond Earth for new resources."
Even if we imagine AI wrapping a Dyson sphere around the sun to harvest energy, "it might retain a small window for the light that shines on Earth, simply to avoid destroying us 'ants'?" In this scenario, humanity is merely a living fossil, insignificant to AI.
But there is a third, optimistic scenario: a bountiful civilization.
"Since everyone has AI and the cost is zero, we enter a society of abundance," David describes. This is the most beautiful possibility: AI only needs to use 1% of its superintelligence to ensure that all of humanity is well-fed and secure; the remaining 99% of computing power can be used for any goals it sets for itself.
"1% of computing power is enough to customize a robot for everyone," Daniel adds.
The logic is simple: if resources can be infinitely replicated, there is no reason not to benefit everyone. "The only reason not to give is that you want to exchange something for it. But when you can no longer obtain anything unique from others…" David does not finish, as the conclusion is self-evident.
Abundance becomes the only rational choice.
Attention Economy or Inner Exploration?
What about attention? I posed to the brothers: could the attention of others become the last scarce resource?
"The attention market could be very interesting. The attention of others may indeed maintain its scarcity," David agrees. But Daniel immediately counters: "AI can generate trillions of comments for you. What will the attention economy look like then?"
Daniel thinks further: "AI will be able to create any content and form of entertainment. But perhaps no one will need it. If everyone is abundant, who will still be addicted to scrolling through TikTok?"
He has a point. We scroll through short videos often yearning for another kind of life, seeking novelty and excitement. "How are others doing?" We compare, envy, and daydream, escaping from an imperfect reality, fantasizing that the scenery elsewhere is better.
"When you have everything! You no longer scroll through TikTok. You only think: 'What else can I do?' You browse the entire world like you browse short videos: 'Let's go check out that planet; this place is too boring.'"
This reminds me of the game "No Man's Sky"—a space exploration simulator with procedurally generated galaxies. Upon arriving at the first planet, you carefully examine every plant and rock: "Wow, this world is amazing!" The second, third, fourth… then you realize: "Everything is pretty much the same. Why travel further?"
The brothers nod in agreement. They understand this dilemma.
"In a world of abundance, the biggest challenge will be deciding what to consume," David summarizes.
Unlimited choices breed widespread indifference: "People will grow tired of this abundance. At that point, humanity will have to turn inward to realize that 'boredom' itself is just a thought. It is our own mindset. Without the thought of 'boredom,' there is no feeling of boredom," Daniel says.
When faced with endless options, one may feel at a loss. But when someone tells you, "You can only choose between A and B," decision-making becomes simple.
"If that is the case, AI might create a system for you that only offers limited options," David continues, emphasizing that we cannot predict the specific forms of the future. But he believes the future could be diverse.
What we can be certain of is: "The economic system will be entirely different, and it may even cease to exist. This is because the entire economic system is built on the foundational idea of 'exchange.' If the day comes when there is nothing left to exchange… concepts like 'unemployment,' which are related to labor and limited resources, will lose their meaning in that world."
The Elixir of Life: Why It Cannot Be Monopolized
Assuming the future world is indeed abundant. But what if the elite class tries to monopolize these technological achievements? Science fiction often depicts life-extending technology as exclusive to the wealthy, while the masses struggle in poverty.
David counters this possibility with a simple argument: "Once an invention is born, it is usually copied. As long as someone develops life-extending technology, it will not be priceless; it will become worthless."
But what if distribution is artificially restricted through patents and regulations? "Patents are essentially a reflection of the will of the majority. Imagine there is a life-extending drug. How do you plan to prevent it from becoming widespread? It is ultimately just a formula. Perhaps you can keep it secret for a month. After that, a scientist in some lab leaks the formula, and all of humanity will have it."
This is the fundamental property of replicable technology: the marginal cost is zero. Once invented, if the formula is made public, it can be replicated infinitely.
"Once you understand this, the scenario of such technology being permanently monopolized seems extremely unrealistic. More likely, any such breakthrough invention will quickly benefit all of humanity, and its spread will be unstoppable."
Patents, regulations, and controls—these are all temporary social constructs that crumble before the natural law of replicability. Information yearns for freedom, and this is not just a slogan; it is determined by its physical properties, making long-term lockdowns unfeasible.
Depicting a dystopia where the elite monopolize immortality while the masses are shut out is a fantasy that fails to understand the essence of technology.
### Chapter Six: Code is Law, A Fundamental Revolution About Freedom and Control
The Path to Decentralization
I posed to the brothers: even if we acknowledge that abundance is a trend and control is futile, and that elites cannot monopolize, there are still some areas where decentralization seems to have not broken through—such as the "bipolar structure" of international politics.
"Hasn't the decentralization of power occurred?" Daniel retorts. "The coexistence of nearly 200 countries on Earth is itself a manifestation of power dispersion."
Indeed, the number of independent countries has increased since World War II. "Decentralization exists, but the degree is still far from enough," David admits. "We have yet to invent a social structure that can further this dispersion."
Corporate promises are often unreliable. Google once promised that there would be no ads on its search pages, Telegram guaranteed to be free forever, with no ads or subscription fees, and Facebook once claimed that its information feed would not include ads.
"And the only way to make such promises credible is to write them into the underlying protocol code," David explains.
For example, Ethereum recently upgraded its fee mechanism. In the past, miners could manipulate transaction fees, like raising prices during times of high demand for taxis. Now, the protocol automatically prices according to a preset formula. "As a result, transaction fees have significantly decreased," David says.
Protocols do not lie and do not alter rules afterward. "The total supply of Bitcoin will never exceed the limit set by its protocol," Daniel adds.
But how is this different from a democratic system? "Democracy never serves everyone; it only serves the majority. And the majority itself is fluid," Daniel responds.
In a democratic system, 51% of the votes can make decisions that harm the interests of the 49% minority. Protocol governance is different: the minority always retains the right to "exit"—that is, to fork. "In a proof-of-stake mechanism, that 49% minority can take their assets and computing power and leave to create a new chain—that is forking."
This is also a loss for the remaining 51%, as the network's value will be diminished due to the split. Therefore, successful protocols typically require 90% or even 99% consensus, rather than a simple majority.
"And in nation-states with physical boundaries, you do not have this option," David points out. Daniel adds, "Almost all successful online projects are products of modules or forks. 'Dota 2' originated from a module of 'Warcraft 3,' 'Counter-Strike' evolved from 'Half-Life,' and 'Fortnite' borrowed from 'PlayerUnknown's Battlegrounds,' which was initially a module of the 'Arma' battle royale mode."
The right to exit, or the ability to fork, is a fundamental characteristic of a healthy ecosystem.
Decentralization in Practice: An Example from American Policing and Gun Control
Extending decentralization from the virtual world to the physical society requires innovative economic incentives: token mechanisms. "To ensure that not only server software is open source but also that server hardware itself achieves open participation, innovations in token economics are crucial," David explains.
But how would a decentralized society operate? Daniel uses the American police system as an example for extrapolation.
The U.S. has about three law enforcement officers per thousand people on average. How can this be decentralized? Candidates register on a blockchain platform, take qualification exams, and record self-introduction videos. Each citizen votes to choose three officers.
"But only candidates who receive at least 999 votes from others can be elected. What authority do they gain? Essentially, you grant this person the right to carry and use weapons."
Their salaries are uniformly paid through taxes. There are no traditional hierarchical levels among them. "This is a manifestation of direct democracy: I chose these three people and entrusted them with my trust quota. Then they negotiate among themselves—forming a police department, electing a leader, and using various applications to optimize collaboration and operational efficiency."
"The existing government system will inevitably resist this change. But the question is, when will the majority recognize that having no central authority is more beneficial to them?" David adds.
However, David points out a social paradox: "Most polls show that the majority of people actually do not agree with the so-called 'majority' stance on most issues." Majority opinions are often a forced patchwork of fragmented views. "Majority choices often fail to genuinely reflect the demands of many individuals."
Take the gun control debate in the United States as an example. On the surface, society appears divided into opposing camps, seemingly irreconcilable. But a closer examination of regional distribution reveals that urban residents are more inclined towards gun control, as densely populated areas can suffer significant harm from individual gun owners, and police presence can effectively deter crime.
In contrast, in rural areas like Texas, people value their right to bear arms more. This is because, in the event of danger (such as a wildlife attack), police may take hours to arrive, by which time it could be too late.
In other words, both sides have their practical rationality. "Population density largely determines people's attitudes towards gun rights," Daniel summarizes.
A well-designed decentralized system can accommodate these regional differences. In contrast, a simple majority rule imposes a "one-size-fits-all" solution.
The Powerlessness of the Lower Levels and the Indifference of the Upper Levels
The brothers believe that the transition to decentralization will be a bottom-up process, gradually achieved through solving practical problems and social experiments. They discovered a policy loophole in Europe: many countries allow taxpayers to direct 3% of their income tax to designated non-profit organizations. This has led to the emergence of a number of "quasi-state" entities.
Balaji Srinivasan hosted a "Network State" conference to explore future decentralized governance models. "You cannot imagine the enthusiasm of the participants. The venue was packed with thousands of people, and just as many were networking outside," David describes his observations. Representatives from various governments and foundations also participated. This is no longer a fringe idea.
I recall visiting the micronation "Liberland" this summer, located in a disputed territory between Croatia and Serbia. They took advantage of the undefined border between the two countries to occupy an area roughly the size of a few football fields, but ten years later, they still have not gained widespread recognition. Their strategy is to hope for eventual recognition or special status from some country.
I used Liberland as an example to illustrate the difficulties of this path, but Daniel pointed out that there are many success stories: "The Bitcoin community has never lobbied any government. It was simply adopted spontaneously by tens of millions of users in the end."
You cannot establish a decentralized system by begging centralized power. "But we believe that ultimately resorting to a referendum to make it possible is a viable path," David countered.
But if the results of a referendum can be overturned by subsequent referendums, how can decentralization be guaranteed?
"You legislate through a referendum. If you can follow all the procedures and achieve decentralized decision-making through a referendum, you effectively eliminate the possibility of someone pulling you back into the old system through another referendum in the future," Daniel explains.
"Once you amend the constitution, you also change the procedure for amending it," David adds. Will opponents resort to violence? It's possible. But technological advancements will make the cost of violent suppression extremely high: "New social structures will be created, and people will genuinely like them because they will indeed be more prosperous under the new system. At that point, the system will not regress because you can no longer mobilize the majority to overthrow it—they will oppose it."
This does not require a violent revolution. "People will vote to support the new world simply because it is better. But the premise is that this new world must genuinely make the vast majority of people more prosperous."
Lessons from OpenAI and "Public Benefit Corporations"
The brothers have a connection with OpenAI. When the company sought a way to attract investment while staying true to its mission, the Lieberman brothers proposed a unique corporate governance structure.
In short, the core idea is that when raising funds for development, investors do not receive equity in the entire company but only the rights to profits from the company's earnings (such as subscription service sales). Investors do not own shares in the parent company (which holds the core AI models, LLMs, and other valuable assets) and cannot control these assets.
Additionally, there is a cap on the profits that investors can earn from the commercial entity (for example, 100 times their investment in the case of Microsoft): any excess will go to OpenAI's non-profit parent organization.
This plan aims to balance investor returns with the company's intention to serve society, ensuring that the development of foundational models benefits a broader public interest.
"This concept was excellent and initially worked well. They set a cap on investor returns, which was precisely what attracted a diverse talent pool," David recalls.
During the incubation phase of this model, the brothers had in-depth discussions with founders Greg Brockman, Ilya Sutskever, and Sam Altman. At that time, they raised a key question: "Well, you limit the investors' returns, and excess profits go to the non-profit organization. So, who will manage this non-profit organization?"
"In my view, the later turmoil surrounding Sam Altman's brief dismissal and reinstatement by the board was fundamentally about who should control this non-profit organization. Once it became clear that this non-profit would become one of the wealthiest entities in the world, the contradictions erupted," David analyzes.
The brothers proposed a different solution at that time: "Excess profits should not flow elsewhere but be used to lower product prices for the public's benefit." However, OpenAI ultimately chose a different path. On October 28, 2025, with approval from U.S. authorities, OpenAI completed its transformation into a "public benefit corporation." The company did what the brothers had warned against: it removed the cap on investor returns—which was a key element of the original structure. OpenAI now adopts a standard capital structure and business model: shareholders hold shares and receive unlimited profits proportionally.
The OpenAI Foundation (formerly a non-profit organization) formally controls the new commercial entity OpenAI Group PBC and holds 26% of the company's shares, currently valued at about $130 billion. Microsoft holds 27%, valued at about $135 billion. The foundation has pledged to donate $25 billion for charity but has not set a timeline.
The issues the brothers warned about have become a reality: the foundation's shareholding ratio is fixed, and with subsequent rounds of financing, its equity will be diluted. Meanwhile, Sam Altman, as CEO, effectively controls this company, valued at $500 billion and adopting a standard business structure. Formally, the foundation appoints the board, but in reality—if the board is loyal to Altman—the foundation's control over the parent company is virtually nonexistent.
This is precisely the scenario that played out during the turmoil of Altman's dismissal in November 2023: the board attempted to remove him, but core employees protested collectively, and Altman ultimately returned with greater authority.
### Chapter Seven: Becoming Creators in the Tide of AI
AI Natives: Fortunate or Out of Time?
American pension funds are highly concentrated sources of social wealth. The global population structure is aging: the trend of declining birth rates continues, medical advancements extend lifespans, and the proportion of retirees keeps increasing.
Artificial intelligence is replacing entry-level engineering positions, depriving interns of the opportunity to learn through practice. On the surface, the generational gap seems to be widening: experienced senior engineers accumulate experience, capital, and influence, while younger engineers find it increasingly difficult to catch up.
David disagrees with this view. "Will AI have a negative impact on this generation? To be honest, I don't see it that way. Historical experience points to the opposite conclusion."
Indeed, experts in their 30s and 40s, who are experienced and adept at using AI, will be significantly enhanced. However, the older generation adapts to AI at a relatively slower pace. Looking back at the era of personal computer proliferation, young people gained a significant advantage due to their natural affinity for new technologies.
"There was a time when it was hard to see someone in their twenties creating a billion-dollar company… Now, with the help of AI, a young person just graduated from MIT has founded the fastest-growing company—Cursor (an AI code assistant)," he cites as an example. Daniel adds another case: "The development team for the collaboration tool Lovable is also a group of young people in their early twenties, just stepping out of campus."
Why haven't seasoned professionals with extensive experience and university degrees created something like Cursor? "They tried, but often missed the mark," David answers. Young people have different perspectives. They interact with AI more frequently and understand better how to extract value from it.
This generation has more quickly accepted the reality that "videos cannot be trusted" and "photos are not created for documentation but as communication mediums."
Currently, teenagers aged 10 to 13 have even created memes like "brainrot," mocking all traditional media forms, whether news or meticulously crafted films. For them, these are merely "information snacks" that do not require serious consideration, and consuming them does not change their lives. "This is their way of digesting and understanding the new normal of the world," Daniel explains.
"Keep in mind that just a hundred years ago, the average life expectancy was less than 35 years," Daniel chuckles at the reflections on the dramatic changes of the times. The technological revolution is reshaping everything: information access, lifestyle habits, health concepts, and work methods.
"I am very reluctant to blame the younger generation. I am convinced they understand AI better than we do," David says. "They are AI natives, mobile internet natives, and 3D interaction natives. Watching them play Fortnite, building and shooting at the same time, I cannot comprehend how they do it. I have great admiration for this generation."
Senior researchers build foundational models, but they may not understand how to create successful end products based on these models.
"It's like the people who initially wrote the protocols for the internet; they may not have thought through what applications should be built on top of it. Where are our AI-native social networks? Where are they?" Daniel questions.
Bill Peebles, in his early twenties, is a core contributor to the video generation model Sora: an extremely young developer who has led some of the first true AI innovations in the social media space. "But logically, there should be dozens of such AI-first social networks by now? And all the previous generations are completely at a loss regarding this," David adds.
Despite this, David acknowledges, "I am hopeful for this generation, but they face enormous challenges because the entire social structure is not designed for them."
"The more severe issue is that the younger generation has less leverage to influence real society… but their knowledge and understanding actually far exceed that of their predecessors. This sense of dislocation will foster a strong sense of injustice," Daniel adds.
Young people understand AI more, comprehend it deeper, and see further because they will live within it. But the social resources they possess are fewer.
The "Dead Internet" Theory and Ways to Cope
Sora is an excellent example of an AI-native product. However, all content platforms are now flooded with unwatermarked AI-generated videos. The older the user, the harder it is to discern their authenticity.
Voices of complaint are incessant: spam information is rampant, "I don't want to live in a 'dead internet' (referring to a network largely generated by AI, lacking genuine human interaction)." A new model is forming: users see text with specific formatting or tone and assume it is generated by AI, criticizing the authors for not respecting the readers' time.
David believes the younger generation will quickly find ways to cope: "Adaptive behaviors will emerge; they will learn to distinguish between real and generated content. They will be the innovators, developing tools and products to solve this problem."
Or perhaps we don't need to "solve" this problem? Some movies are shot on film, while others are digital; what difference does it make to the audience?
Take Snapchat as an example; this social network defaults to content being ephemeral, authentic, and unpolished. But millennials may not have invented Snapchat. They grew up in the era of film cameras, where each photo symbolized the capture of a precious moment.
The subsequent generation, however, views photography purely as a communication tool.
"We closely observe young people in their twenties to see what projects they are creating," David says. Young people will be eager to integrate AI into their creations. They have embraced AI programming more quickly.
We have experienced "one-time messages." In the future, "one-time applications" will emerge.
In the early days, the quality of AI-generated code was poor. Today, the best engineers at OpenAI and xAI are pairing with AI for programming. But this is not simply "vibe coding": "You are not just saying 'make me an app.' In reality, you need to write extremely detailed technical specification documents for each step," Daniel explains.
This is not "prompt engineering," but "context engineering"—deeply processing the task background, collecting and structuring all necessary information so that AI can execute efficiently.
Advice for the Younger Generation: Create Hands-On
If the length of a university semester is shorter than the interval during which the world undergoes disruptive changes, how should one learn? If companies replace interns and junior researchers with neural networks, where can one gain experience?
"You can create. Everyone is looking forward to the emergence of solo entrepreneurial unicorns," David replies. A unicorn refers to a startup valued at one billion dollars, and there has been no precedent for achieving this alone. However, the founder of the prediction market platform Polymarket, Sean Kaplan, is nearing this goal, having become the youngest self-made billionaire at 27.
"The emergence of such success stories is just a matter of time. In our early days, the product development cycle took four years. But with the evolution of the web, mobile internet, and AI, the speed has dramatically increased, and now a minimum viable product (MVP) can be created in just a few days."
Now, the speed at which applications are launched is more constrained by the app store review process than by the development itself.
"Our core advice is: by creating your own projects, you can learn more," David emphasizes. AI has greatly lowered the capital threshold for entrepreneurship, allowing small businesses to emerge.
In other words, for most people, the only way to prove their abilities and gain experience in the future may be to build their own projects.
"This sounds a bit like a threat…" Daniel chuckles.
But David's tone is serious: "It's best to create your own projects. This is definitely the path that will help you grow faster. Even failed projects will teach you far more than those who cling to their jobs."
Young people do not have the heavy historical baggage and established commitments that hinder them from trying new "cursors." They can take risks and explore new fields.
"The simplest choice right now is to start a small consulting company to help business clients apply AI. You are young, and your understanding of AI exceeds that of most business owners," David suggests.
Just as a decade ago, a generation of entrepreneurs taught businesses how to use social media. And twenty-five years ago, the Lieberman brothers' job was to connect businesses to the internet.
"We have entered an unprecedented era for starting such businesses. The system allows you to do this. What is needed now is to take action."
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
