A Friend Asked, Volume I: From Instrumentalization to Self-as-an-End

Qin, Han

doi:10.5281/zenodo.23959070

About this volume

A Friend Asked: Essays on Self-as-an-End, Volume I: From Instrumentalization to Self-as-an-End is a collection of applied-philosophy essays that begins with ordinary questions. This volume contains the general preface and sixty-five essays, arranged in the order in which they were originally written and organized into four parts: “Learning to Judge,” “The Expansion of Instrumentalization,” “The Person as an End,” and “The Framework Enters the World.”

Beginning with questions about risk, technology, finance, platforms, and popular culture, the essays gradually develop the central concern of Self-as-an-End (SAE): how people can resist treating themselves and others merely as instruments within institutions, technologies, and inherited narratives. Volume I records not the triumphant completion of a theory, but the way a concern grew out of concrete questions and began to submit itself to the test of reality.

Author's Note

These essays did not begin with a finished theory in search of examples. They began with questions from friends: Why does a person become afraid? Why does a technology inspire both excitement and unease? Why does an institution turn people into instruments in the name of protecting them? The writing often came before the answer, and judgments changed in the face of new facts and counterexamples. In gathering these pieces into a book, I have not tried to erase the particular circumstances in which they arose. I hope, instead, that the questions will remain visible after the news has receded. Readers need not accept every conclusion here. If these essays can help a question be asked more precisely, they will already have done part of their work.

Translator's Note

The Chinese and English editions of this book are published separately, but their volumes, parts, essay numbers, and hidden stable source IDs correspond exactly. The translation is based on the edited Chinese manuscript, not on a line-by-line transcription of the original X posts. Platform residue, passing references, and conversational phrasing that depended entirely on the moment of posting were addressed during the Chinese editing process. The English text preserves the historical setting of each essay while adding only the minimum context needed by readers who may not share the same news memory.

Self-as-an-End (SAE) remains in English throughout the book. It is not a synonym for self-interest, self-actualization, or autonomy. The term extends a Kantian concern with persons as ends in themselves into a broader question: whether a person is recognized, within the institutions and social structures that shape their life, as the ultimate source of the ends they live by rather than as a means for goals imposed by an external system. Where the distinction matters, the translation separates Kant's formulation from the author's expanded use of Self-as-an-End without adding an apparatus of scholarly notes.

Core terms are kept consistent. Instrumentalization and self-instrumentalization retain distinct meanings, while terms such as alienation, agency, and subjectivity are distinguished according to context. Consistency, however, does not mean mirroring Chinese syntax sentence by sentence. The translation aims to preserve the force of the judgments, the humor, the hesitation, and the author's habit of correcting himself. Sentence length and paragraphing have sometimes been adjusted for natural English, but no new argument has been added.

Chinese institutions, classical texts, internet culture, films, television series, and games are explained only where comprehension requires it. Personal names follow the Romanized forms commonly used by the people concerned, and titles are kept consistent after their first appearance. Relative dates have been anchored to the historical moment of composition, and predictions remain predictions rather than being rewritten in light of later events. References already present in the essays remain part of the prose; no new system of citations or notes has been imposed.

Some wordplay, kinship-inflected forms of address, and political resonances cannot be carried over at equal weight. In such cases, the translation favors the function the expression serves in the argument, with a brief explanation in the text only when indispensable. Translation is itself an act of interpretation. The aim here is not to remove every ambiguity on the author's behalf, but to let English readers encounter the same questions that confront readers of the Chinese original.

General Preface

Writing as a Form of Prayer

I could hear the friendly teasing in the question, but I also thought it was a good one, and more meaningful than it first appeared.

Kafka, yes, the Kafka who wrote The Metamorphosis, once said that writing was a form of prayer. I agree. First, why prayer rather than some other metaphor? Sincerity, passion, and persistence can all describe serious writing, but they describe so many things that they can be made to fit almost anything. Prayer is different. It has a very specific structure: we speak to someone or something that is under no obligation to answer us.

Every part of that sentence matters. We are speaking, but we are not merely talking to ourselves. Our words have a direction. They are addressed to someone. Yet that person owes us no reply. However precisely, sincerely, or forcefully we speak, we cannot demand an answer.

This is where a writer stands. We can take responsibility only for our half, for putting the words into the world. Whether understanding occurs belongs to another person, and it may not occur at all. We cannot complete the process by ourselves.

Once that structure becomes clear, several other things come into focus.

The first is that writing asks us not to close a question too quickly.

Imagine someone writing, “I have so much, and yet I am still unhappy.” They could immediately add, “So I must be ungrateful.” The essay would be over, and the question would have been shut down before it had really begun. Or they could keep writing: What exactly do I mean by unhappy? Can I value what I have and still be unwilling to continue living under a particular arrangement? Why have the conditions other people consider sufficient failed to address what is actually troubling me?

Here, writing is not simply a way to express emotion. It keeps a ready-made answer from replacing the real question too soon. Writing resembles prayer because the writer does not have to play the role of someone who has already figured everything out. The writer can admit, “I do not yet understand, but I am willing to look carefully.”

Flannery O’Connor’s A Prayer Journal is a good example, and a genuinely religious one. She asks to become a better writer and to have her work published, but she also examines her own hunger for success. She prays for clarity and asks whether her desires might be obscuring what she is trying to understand. She does not reduce prayer to “Please give me what I want.”

From this we can draw a way of writing that does not depend on religious faith: I hope for an answer, but I am also willing to discover that what needs to change may be not only the answer, but the way I ask the question.

The second is that writing not only asks the world to listen to me. It also allows the world to correct me.

If an essay’s only task is to prove that its author was right all along, then the evidence, the people in the essay, and even the reader will all be arranged as tools in support of the conclusion. Another kind of writing allows what resists us to remain on the page. We may love a beautiful sentence and delete it when we discover that it is inaccurate. We may begin by casting a person as the embodiment of an idea, only to find that the particulars of their life refuse the label. We may think the argument is complete, then return to rewrite it when a reader points out a counterexample.

In such moments, writing does more than amplify the voice of the “I.” It also limits the “I’s” power to explain everything however it pleases. Writers sometimes describe revision this way in interviews: at a certain point, a sentence is no longer merely something they own; it begins to answer back. This is a description of the experience of writing, not a claim that inspiration comes from some supernatural force.

Of course, being willing to be corrected does not mean abandoning judgment. We still have to decide whether a correction deserves to be accepted or is merely another untested demand. Openness does not mean accepting everything indiscriminately.

The third is that writing longs for a response, but cannot turn that response into a command.

We can make a sentence more precise, describe an experience more concretely, and strengthen an argument. None of that guarantees that another person will understand, much less that they must agree. To write for readers we do not yet know therefore involves an unguaranteed trust: perhaps someone else is facing a similar confusion; perhaps they will pause over a particular sentence; perhaps they will disagree, but in doing so help me see what I could not see on my own.

Nor must a response always come from someone else. A private journal may give one’s future self a chance to look back and understand one’s present self more fairly.

Sometimes a lack of response is a real disappointment, but it does not cancel the value of having spoken. At the same time, writing seriously does not entitle me to anyone’s attention. I hope to reach you, but I do not claim ownership over your response. I believe communication is worth attempting, but I do not assume it must succeed.

This metaphor also reveals something else: prayer done badly and writing done badly fail in almost exactly the same ways.

One failure turns prayer into a demand that says nothing but “Give me this.” Its equivalent in writing is a sentence that is always asking for something: agreement, sympathy, recognition. Another form of prayer is performed for an audience, with posture more important than substance. Its equivalent is writing designed mainly to look profound. A third begins with the answer already settled and merely recites it. Its equivalent is an essay whose conclusion is fixed in the first paragraph and whose middle merely builds a road to it. All three are still forms of speech, but the structure in which “the other person owes me no answer” has disappeared. They do not actually need a response. They need only a witness.

Prayer is also often imagined as waiting for some power to take care of things on our behalf. In this metaphor, the opposite is true.

We cannot guarantee that a neglected experience will eventually be understood fairly, but we can begin by writing about it without distorting it further. We cannot guarantee that everyone will treat others with seriousness, but we can refuse to reduce the people in our writing to props that prove our point. We cannot guarantee that our theory is correct, but we can state its premises clearly, preserve the objections, and mark what remains unresolved.

A wish does not have to remain only a wish simply because its fulfillment is not guaranteed. It can enter into very concrete acts: how a sentence is written, how an example is handled, whether a revision is honest.

For the same reason, this kind of “prayer” need not be gentle or serene. It can contain anger. It can protest harm. It can also express gratitude for something good. Writing down an ordinary moment with one’s family need not serve any conclusion. It may simply recognize that the moment deserves to be seen, that our busyness should not make it disappear merely because it was not spectacular. In such moments, writing is not asking life for more. It is responding to the life that has already happened.

Finally, prayer may explain why we write with such seriousness, but it cannot serve as evidence that what we have written is true. An essay may begin in pain, hope, or care, but its argument must still withstand scrutiny. Sincerity does not make a conclusion true, and an explanation does not become valid merely because it comforts us. Conversely, the willingness to delete a claim that offers great comfort but cannot survive examination may be the clearest expression of a writer’s honesty.

Of course, it is also a mistake to sanctify writing into a mission one is never allowed to stop. Writing does not require us to suffer continuously, to produce words every day, or to turn every experience into a work.

What moves me most about the metaphor, then, is that it places limitation and hope in the same sentence: I cannot guarantee understanding, yet I remain willing to understand; I cannot guarantee that I will be heard, yet I remain willing to speak carefully; I cannot guarantee that I am right, yet I am willing to let what I write receive an answer.

It does not say, “May the world prove me right.” It says: may I be able to see what does not conform to my wishes and still write about it honestly; may these words reach others without prescribing their answers for them.

Volume I: From Instrumentalization to Self-as-an-End

Volume I begins with judgment. Markets, artificial intelligence, institutions, platforms, and popular culture may seem to belong to different worlds, yet each repeatedly asks the same question: as our tools grow more powerful, are we handing over our judgment as well? These essays appear in the order in which they were written because the central idea did not arrive all at once. It first hid inside questions of risk, trust, narrative, and self-doubt. Only later did it acquire the name “Self-as-an-End.” Volume I records not the triumphant completion of a theory, but the way a concern grew out of concrete questions and then began, in turn, to submit itself to the test of reality.

Part I: Learning to Judge

Judgment is not the same as guessing an outcome correctly. A prediction may come true by accident, and a system may work for a long time without being able to bear the burden of choice for each person within it. The essays in this part enter an uncertain world through markets, technology, law, and education. They ask why we trust one signal rather than another, and how we can identify what we have failed to see. Tools can expand information, computation, and action, but they cannot answer, on behalf of the people who use them, what is worth doing. The easier it becomes to hand a decision to a model, platform, or authority, the more important it is to retain responsibility for judgment.

1. Can AI Predict Markets?

Using AI models to predict markets is possible to a certain extent, but only if we understand both their strengths and their limits.

1/ Feasibility and Advantages

1.1/ Powerful Pattern Recognition

AI, especially deep-learning and time-series models, can capture complex nonlinear relationships that people may miss, such as interactions among price movements, macroeconomic factors, news sentiment, and on-chain data.

1.2/ Richer Data Dimensions

Traditional quantitative strategies may rely only on price and trading volume. AI models can integrate many more kinds of data: news, posts on X, Google Trends, macroeconomic indicators, corporate earnings reports, and on-chain activity.

Large models offer advantages here that earlier models could not match.

1.3/ Well Suited to Short-Term Forecasting and Trading Support

In high-frequency or short-horizon trading, AI can gain an advantage by quickly identifying small signals, such as order-book dynamics and arbitrage opportunities.

2/ Limits and Risks

2.1/ Unavoidable “Noise”

Financial markets are highly random, and signals are difficult to separate from noise. However powerful AI becomes, it cannot guarantee consistent excess returns.

2.2/ Overfitting

An AI model may perform extremely well on historical data, then fail when the market’s structure changes. This is especially common in financial markets when so-called black swan events occur.

Of course, almost every model is vulnerable to overfitting, so this is not necessarily a problem unique to current AI models.

2.3/ Feedback Effects

If many people use similar AI models, their predictions may be priced in early, causing the models to stop working. Capital may also pile into the same position, creating the risk of a market crash.

Again, this is a limitation of finance itself rather than of AI alone.

2.4/ Lack of Causal Explanation

AI is better at predicting correlations than explaining causation. That can create problems for risk control and regulatory compliance.

For current models, this limitation is particularly serious.

3/ Practical Applications

Quantitative hedge funds such as Citadel, Two Sigma, and Renaissance already make extensive use of AI models, particularly in short-term trading and risk management.

Retail investors can also use AI, for example to analyze news sentiment or forecast ETF fund flows, to improve their decisions. But they cannot treat it as a tool that guarantees profits without losses.

AI is better suited to assisting decisions than making them independently. It can produce signals, while people combine those signals with macroeconomic and fundamental analysis before making a judgment.

Conclusion

AI models can make investing and trading more efficient and accurate, but they cannot eliminate market uncertainty.

The most prudent approach is to use AI as a tool that augments decision-making, not as a crystal ball that replaces human judgment.

The constructive direction for further thought is clear: trading strategies that use AI will have an advantage over strategies that do not.

2. Can Luck Make High Leverage Sustainable?

The best answer is that we still do not fully appreciate how lethal a black swan event can be to a highly leveraged position.

1/ Leverage Amplifies Every Movement

Leverage means using a small amount of one’s own capital to control a much larger position.

Under ordinary market fluctuations, it can multiply gains. But when an extreme event occurs, it multiplies losses as well.

At five-to-one leverage, for example, a 20 percent decline is enough to wipe out the original capital.

2/ Black Swan Events Often Move in Jumps

Black swans often arrive with sudden news: war, regulatory bans, major hacks, financial crises, and so on.

The resulting shock is not a smooth, linear decline. Prices can gap sharply downward, leaving stop-loss orders no time to execute.

A leveraged position can blow straight through its margin, be liquidated instantly, and even leave the trader in debt.

3/ Vanishing Liquidity and Cascading Liquidations

When a black swan arrives, market liquidity can dry up. Buyers disappear while sell orders pile up.

Highly leveraged positions are forcibly liquidated, causing automatic selling that drives prices down further. The decline triggers still more liquidations, which produce still more selling, until the market enters a downward spiral.

Leveraged traders are often the first fuel fed into this chain reaction.

4/ The Risk Cannot Be Fully Hedged in Advance

Ordinary risks can be partly reduced through risk controls, stop losses, and hedging.

Black swans, by definition, cannot be predicted or priced in ahead of time. A leveraged account therefore has almost no room for defense.

Even attempts to buy insurance through options or hedged positions may fail under extreme conditions, or the necessary liquidity may disappear.

5/ Psychology and Capital Management Break Down

The more leverage a trader uses, the greater the psychological pressure. In a black swan market, fear and greed can make traders miss their chance to escape.

Investors with low or no leverage at least retain the possibility of surviving and waiting for a recovery. High leverage can end the game in a single stroke.

Conclusion

Black swan events bring extreme price movements that cannot be predicted or controlled. Leverage magnifies that uncontrollable risk to the point of forced liquidation, which is why it can be fatal to highly leveraged traders.

3. If Anyone Can Buy US Stocks, Why Do Investment Barriers Persist?

The issue is really one of asset classes. Every asset class has a different average long-term compound annual growth rate, or CAGR. From that perspective, publicly traded US stocks are not even the best-performing category.

Roughly speaking, long-term CAGRs rank from highest to lowest as follows: startup equity, venture capital, private equity, US public equities, real estate, government bonds, and cash.

Cash comes last. That is easy to understand. Its CAGR is only 2 to 3 percent in interest, roughly the rate of inflation. Its real purchasing power barely grows over the long term and may decline during periods of high inflation, as it was doing in the United States when this essay was written. Cash is useful mainly for liquidity and protection from volatility, not appreciation. Ordinary people can, of course, hold it freely.

Government bonds, such as ten-year US Treasuries, do somewhat better at roughly 4 to 5 percent. They are low-volatility assets used mainly to hedge risk. Buying Treasury bonds directly can be difficult for ordinary investors, but Treasury ETFs offer indirect access with almost no barrier to entry.

Without leverage, real estate has a long-term CAGR of only 6 to 8 percent. Its real appreciation is not especially high. Most of its returns come from leverage, rental cash flow, and tax advantages. Here, ordinary investors begin to face serious barriers. Down payments are high, and high interest rates also reduce returns. REIT ETFs provide some access, but once fund fees are included, ordinary investors still find it difficult to capture the asset class’s true return.

Including reinvested dividends, US public equities have a CAGR of roughly 9 to 10 percent. The S&P 500, for example, has a long-term nominal CAGR of approximately 9.5 to 10 percent. This is the highest-returning asset class that ordinary people can enter conveniently.

Beyond it lie assets with still higher returns, but ordinary people are excluded by design.

After management fees, private equity has a long-term average CAGR of roughly 10 to 13 percent. Top funds may earn more, though survivorship bias is substantial. Its returns derive mainly from high leverage, the illiquidity premium, and financial management. The best funds are designed for family offices and institutions, and their thresholds are high. Ordinary people not only cannot participate; they often do not even know such opportunities exist. We frequently hear someone say that a company is excellent but, unfortunately, not yet public, and that they will buy it as soon as it lists.

Venture capital, or VC, can average about 15 percent overall, while top funds may reach 20 percent or more. Here, ordinary people have almost no opportunity at all. Everyone may know that a16z or Sequoia is formidable, but how many ordinary investors can become limited partners in their funds?

Startup equity through angel investing may be the best-performing category of all, with a theoretical average of 20 to 30 percent or more. But its variance is enormous. Around 90 percent of angel investments return close to zero, while a small number of hundredfold gains determine the entire portfolio’s performance. Success demands considerable skill from the angel investor, and ordinary people again have almost no access.

The assets that truly widen intergenerational wealth gaps, then, are private equity before an IPO, limited-partner interests in top venture funds, and startup equity with advantageous return distributions, such as angel and early-round investments. Ordinary people struggle even to learn that these markets are open to anyone at all.

This also helps explain the popularity of crypto assets, especially Bitcoin. Their historical CAGR from 2011 to 2024 was extremely high, above 50 percent, and anyone could participate. Whether that rate can be extrapolated is another matter. Crypto’s true long-term CAGR remains to be seen.

4. If Crypto Is Centralizing, Why Is Self-Custody Still Growing?

This is an excellent question. These two trends are not really strengthening decentralization. They are institutional repairs for the risks created by cash-like private keys.

Hardware wallets such as Ledger and Trezor have become popular, while multisignature self-custody through systems such as Safe and Squads has been widely adopted. DAOs, venture funds, and high-net-worth users now rarely rely on a single private key.

On the surface, this looks like a sign that users are becoming more security-conscious and that self-custody is maturing. The deeper reason is that the single-key model is unsustainable in the real world. People are using institutions to repair cryptography.

A hardware wallet is not decentralization. It is a physical safe for a private key. It addresses the attack surface, not the relationship of trust. Its purpose is to move the key from a general-purpose device connected to the internet, such as a phone or computer, into a closed, single-purpose device.

This is like moving cash from your pocket into a safe. The system has not become more decentralized; it has only become more resistant to attack. The user remains the sole bearer. Losing the device and seed phrase can still mean losing the assets forever. A coerced signature, the so-called wrench attack, can still transfer the assets away. The risks have not disappeared, and responsibility has not been distributed. Only the form of the risk has changed.

Hardware wallets are therefore best suited to technically skilled users, long-term cold storage, and people who can bear full responsibility for failure. Their core value is protection from hackers, not protection from predictable human frailty.

Multisignature wallets mark the true institutional turning point, and their significance is much greater than that of hardware wallets. Multisignature custody introduces governance rather than stronger technology. It does not depend on more advanced cryptography. Instead, it divides control, requires consensus, and deliberately adds friction. That logic closely resembles the foundations of modern finance.

Multisignature custody is like requiring a board of directors to sign off. It is a quasi-institutional form of self-custody. In practice, it is used mainly to protect against insiders, a founder as a single point of failure, coercion, and impulsive decisions. Unlike a hardware wallet, its core value is to counter predictable human failure, not hackers.

Multisignature custody does not eliminate trust. It makes trust explicit. With a single key, we must place 100 percent of our trust in one person. With multiple signatures, we can specify whom to trust, how much to trust them, and who has a veto. This is a shift from implicit trust to structured trust.

Why are both approaches rising at once? Because they repair two different kinds of failure. Hardware wallets address technical attacks; multisignature custody addresses organizational risk.

Both are valuable. Unfortunately, neither solves the problem of adoption at the scale of ordinary users.

Consider the history of cash: people moved from holding cash themselves, to storing it in private safes, to using corporate accounts, and finally to the banking system.

Crypto is following almost the same path, only faster. It began with single-signature self-custody, moved to hardware wallets as the ultimate personal defense, then to multisignature custody as the starting point of organization, and finally to custodianship as the basis of social scale.

What can carry capital at scale is not a “safer private key,” but an accountable structure of control. A systematized, centralized, and compliant structure will inevitably emerge to bring blockchain technology to broad, mass adoption.

5. Why Is Privacy Crypto’s Next Frontier?

This question is closely related to mass adoption. The renewed interest in privacy does not mean that crypto has suddenly returned to idealism. It means that crypto is leaving its frontier stage and entering an institutional one. Privacy is not meant to evade regulation. It is what will allow crypto assets to enter mainstream finance and the real economy.

Without privacy, institutions will not move on-chain, businesses will not use blockchain systems, and high-net-worth investors will not hold crypto assets at scale.

Blockchain’s current design is counterintuitive. By default, Bitcoin and Ethereum make every transaction permanently public. Every address is traceable, and asset flows can be mapped transparently. In the early years, this was an advantage. It made verification possible and constrained abuse by centralized actors. In a financialized system, however, the same feature becomes a fatal flaw.

In traditional finance, privacy is built into the structure. Account balances are not publicly visible. Counterparties are not disclosed. Portfolio holdings are reported only to regulators. Much of a firm’s activity is treated as a trade secret.

In today’s on-chain world, if I send you $100,000, the entire world is told about it. That is not financial activity. It is a ledger broadcasting itself in public.

Privacy becomes important as blockchain users shift from technical enthusiasts to businesses and eventually to nation-scale institutions. During blockchain’s first decade, its users were geeks and early speculators with relatively little money, a high tolerance for risk, and little concern about exposure. Some even took pride in transparency. They did not need privacy.

In the years leading up to this essay, institutions and businesses began adopting blockchain. Hedge funds, market makers, family offices, and corporate treasuries all care intensely about preventing others from targeting their positions, front-running their trades, copying their strategies, or analyzing their finances. A system of complete transparency with no privacy is unacceptable to them.

Future applications at the level of the state will require even more information to be hidden from the public while remaining visible to regulators. Taxation, social security, cross-border settlement, and government subsidies are all examples. A state does not need “anonymity.” It needs rules about what is hidden from whom and what is visible to whom. That is the heart of compliant privacy.

Regulators do not actually need total transparency. They need the ability to look through the system when necessary. Their real concerns are anti-money-laundering rules, terrorist financing, accountable taxation, asset freezes, and enforceability. In other words, regulators need selective visibility, not a world in which everyone is forced to stand naked in public.

The privacy technologies under discussion therefore include zero-knowledge proofs, selective disclosure, view keys, and compliance layers.

Why is privacy also a matter of safety? On a fully transparent chain, criminals can identify their targets precisely. We have compared private keys to cash. Publishing an on-chain balance is therefore like publishing the exact amount of cash stored in one’s home, something no one would accept in the physical world.

Hacking, extortion, and kidnapping had already become very real problems in crypto when this essay was written. Large on-chain addresses were continually targeted by phishing attacks. High-net-worth crypto users faced targeted assaults. DAO treasuries were monitored over long periods. Privacy does not exist to help bad people do bad things. It exists so that good people are not so easy for bad actors to find.

Why has privacy matured only now? Because the technology has finally become usable. In the past, zero-knowledge computation was expensive, user experience was abysmal, wallets lacked support, audits were difficult, and regulators were hostile. The maturation of ZK-SNARK and STARK technology, account abstraction, modular chains, a narrative of compliant privacy, and AI-assisted zero-knowledge verification has begun turning a sound theory into a workable engineering system.

The privacy at issue here is therefore not the strong anonymity and regulatory resistance associated with projects such as Monero and Zcash. What we should expect is a new generation of privacy technology that can be audited, supports selective disclosure, and works with regulators. That is the mainstream path for the next decade.

The visible future is not a binary choice between making every transaction public and making every transaction completely anonymous. Rules will determine who can see what. Such privacy technology is, at its core, financial infrastructure that combines access control with cryptographic proof.

6. Can Large Language Models Beat Humans in Prediction Markets?

I agree that large language models are not good at trading stocks or crypto. My doctoral dissertation reached that very conclusion. But they are extremely well suited to analyzing prediction markets, where they can possess a real edge.

Stocks and crypto are poor fits because their prices emerge from real-time, high-frequency competition dominated by noise.

Prediction-market questions, by contrast, require semantic judgments of probability rather than contests of execution speed. Their usefulness depends more on the quality of the data and the depth of the model.

A yes-or-no question is a discrete event with explicit settlement conditions. It contains much less noise than a stock or crypto price, and its logic is easier to explain. Large language models are not good at timing, but they are very good at judgment. This is where they fit.

The source of alpha in prediction markets is a gap in understanding. Most retail participants lag behind the narrative, become excessively emotional, or engage in wishful thinking. These are precisely the areas in which large language models can excel: rapidly reading real-time data, remaining calm, and avoiding personal preferences.

Some people argue that the market price must always be more accurate than a large language model. But a prediction-market price is not more accurate at every moment. At any given time, the price is simply the consensus on which the current participants are willing to wager. That consensus may contain a collective bias.

This leaves room for large language models. When the present consensus diverges from what can actually occur in the next moment, a model can provide an edge.

The market may underestimate a legal procedure, misunderstand a timeline, or reduce a political contest to an emotional story.

Most important, a large language model’s forecast does not itself change the prediction market’s ultimate outcome, apart from illegal conduct such as insider manipulation of timing. The forecast therefore lacks the reflexivity of a financial-market prediction. Trading behavior does not alter how the real-world event is settled.

This also explains why AI companies focused on prediction markets were beginning to proliferate.

Prediction markets are evolving from speculative entertainment into a society-wide infrastructure for probability.

The ecosystem around Polymarket, for example, had already begun producing AI probability forecasters, AI pipelines that turn news into probabilities, and systems that detect market inefficiencies.

If I were beginning a PhD today, I would change direction and specialize in AI and prediction markets. I might even start a company directly, instead of going to work for a large technology firm.

7. How Does International Law Address a Regime That Massacres Its Own People?

With tensions in Iran at the time, this was a painful question to consider. Let us set emotion aside as far as possible and ask only how international law “handles” such a case.

International law explicitly prohibits a regime from massacring people regardless of their nationality, whether they are foreigners or its own citizens. The unsettling question is not whether the law applies, but how it applies: who enforces it, when, and whether enforcement is actually possible.

The postwar consensus in international law is that nationality cannot make a massacre lawful. A state that kills its own citizens on a mass scale is still committing an international crime. “The people I killed were violent troublemakers from my own country” is not a valid legal defense.

The first relevant instrument is the 1948 Convention on the Prevention and Punishment of the Crime of Genocide. Its central principle is that genocide is an international crime in both peace and war.

The treaty establishes that the victims may be a state’s own nationals and that the perpetrator may be their own government. Neither sovereign immunity nor the priority of domestic law is recognized as a defense.

In plain language, a regime that kills its own people can still be guilty of genocide.

The second category is crimes against humanity, an offense in international criminal law designed specifically to address violence by a state against a population.

Its elements include a widespread or systematic attack against civilians, including killing, cleansing, enforced disappearance, and torture, without regard to nationality.

Historically, this category was created to answer a particular question: if a state lawfully controls its military, police, and courts, does that mean no one can hold it responsible when it kills people?

The answer is no. International law can.

That sounds reassuring. There is law to apply. But who hears the case?

One answer is the International Criminal Court. The ICC principally tries genocide, crimes against humanity, and war crimes.

It prosecutes individuals, not states. Those individuals may include heads of state and senior military or political officials.

This is where reality intervenes. The ICC’s decision to act does not automatically produce an enforceable result.

Whether international law can actually “handle” the crime depends not only on law and adjudication, but above all on a structure of enforcement.

International law has no international police force. In practice, enforcement usually requires authorization by the United Nations Security Council, cooperation from national judicial authorities, or accountability after military defeat or a change of regime.

If a regime remains stable, enjoys the protection of a great power, or benefits from a Security Council veto, the law may exist without being enforced. The outcome still depends on a contest of political power.

If the regime does not fall, is not defeated militarily, and faces no domestic reckoning, accountability is nearly impossible.

During Augusto Pinochet’s military rule in Chile, for example, the state carried out systematic repression, yet he was not tried in his own country for many years. He was briefly detained in Britain late in life but never finally convicted, probably in part because of support from the United States.

International law can therefore resolve the question of legitimacy. It cannot resolve the question of power. A massacre is always unlawful, but whether anyone is held accountable ultimately depends on whether power changes hands.

8. How Can Adults Reeducate Themselves with AI?

This is a good question because it asks about reeducation, not retraining.

In the age of AI, the greatest risk for adults who are not AI-native, which means people like us, is not that we will fail to use the new tools. It is that we will use them with the cognitive operating system of the old world.

Before discussing methods, we need to ask which underlying assumptions have changed.

First, knowledge has moved beyond even the internet era. It is no longer scarce, and both true and false information are more abundant than ever. We should stop treating possession of information as an identity asset and make verification central instead.

Second, thought no longer has to proceed along a single synchronous thread. It can become asynchronous and parallel. We can take one step in an argument, ask AI to conduct deeper research, and return when the results are ready.

Finally, our most important value now lies in judgment, as perhaps it always should have. AI makes output extremely cheap. It can produce a mountain of repetitive prose. What remains scarce is knowing what should not be done, what is an illusion, and what may be right in principle but wrong in timing. Reeducation is, at its core, the retraining of judgment.

Once those premises are clear, what can we do in practice?

First, we can force AI to argue with us. Instead of asking, “Help me write X,” ask, “Make the strongest possible case against this idea.” Invite criticism with questions such as, “Assume I am wrong. Where is the error?” This trains us to distrust our own reasoning in a useful way.

Second, ask AI to analyze the hidden premises in our thinking. Our greatest problem is often not a lack of depth, but the number of premises we have left unstated. Give AI a judgment and ask it to identify the assumptions and implied causal links. We may discover that what felt like intuition was really outdated experience.

Third, use AI to analyze the reasoning rather than jumping straight to a conclusion. Replace “Will X succeed?” with “What conditions would have to hold for X to succeed?” Ask, “Which constraints are hard, and which are soft?” The point is not to obtain a conclusion. It is to have AI divide the question into subproblems and turn a binary yes-or-no dispute into a question of probability. Once again, prediction markets deserve some praise.

Fourth, ask AI to broaden our perspective. Everyone tends to view the present through the experience of their youth. As we grow older, we need help understanding how younger people see the world. Ask, “How would Gen Z or Gen Alpha view this?” Or, “Would young people hear this and think I am an out-of-touch old-timer?”

Finally, we can use AI to record how we made a decision so that we can review it later. Ask, “After this discussion, why did I reach this judgment today?” Six months later, return and ask, “Were my assumptions about the world correct?” and “What did I overlook?”

For adults who are not AI-native, the key to reeducation is not learning new material. It is systematically clearing out bad models.

Taking courses and collecting certificates belong to an older era. In the AI age, what matters is problem-driven inquiry, brainstorming with AI, rapid verification, and frequent correction.

Put simply, we do not need to “learn AI.” We need to force AI to show us what we must change. That can be painful, but once again: no pain, no gain.

Reeducation in the AI age is not comfortable. It destroys old professional certainties. It shows us that we have been trying to navigate a new continent with an old map. It forces us to admit that experience matters less than judgment.

The good news is that adults who complete this cognitive upgrade often gain an advantage over AI-native generations. We have lived through more than one cycle. We understand institutions, incentives, and structures of power more deeply. Add AI as a GPU for the brain, and it is like giving wings to a tiger. It feels almost like a cheat code.

9. Why Must Engineers Learn to Argue with AI?

The answer is emphatically yes. Engineers should argue with AI, and not only over small things. The ability to conduct a serious argument with AI is itself an engineering skill.

An engineer who cannot argue with AI has effectively surrendered all judgment to it.

An engineer who can argue with AI is using it to extend the upper limit of their own engineering ability.

Why the conflict? Because AI is naturally capable of making things up, and it does so in a way that is structurally complete, confidently expressed, and apparently implementable. Engineers do not fear code that is obviously wrong. They fear code that looks right from every angle. In engineering, that can be fatal.

If no one challenges the model, its code may compile and pass every test. Each local component may look correct, while the system as a whole is fundamentally wrong.

An engineer’s natural responsibility, then, is to become the leader who does not trust AI.

Most engineering failures do not occur because someone typed the code incorrectly. They begin because the design itself was mistaken. The incidents we encounter most often arise from false architectural assumptions, ignored edge cases, or misunderstood use cases.

AI can easily reproduce our existing false premises and magnify them into a plan that looks impressively professional.

Calling this an “argument” may sound emotional. In substance, it is a risk scan. Challenging AI forces it to expose our bad assumptions, identify uncertainties, and list possible paths to failure. This is a question of engineering safety.

So what should engineers argue about, and how?

Ask: “Which premises does this solution depend on?” “Which of them are most likely to fail in a real production environment?” “At what input scale does the system break?” “What form will the failure take?”

Also ask: “If this component fails, what is the worst possible consequence?” “Could it fail silently?” “Could it create cascading risk?”

These are the questions a senior engineer already asks while reviewing a junior engineer’s code. Now we must ask them of AI.

We can also challenge the model from different professional roles. As a product manager, ask, “Can this system survive a tenfold increase in scale six months from now?” As a site reliability engineer, ask, “Is this design stable? Will it develop problems after running for a long time?” As a security engineer, ask, “How could a hacker compromise it?” As legal counsel, ask, “Is it compliant?” As a finance lead, ask, “Can we reduce the cost?”

Must the argument continue forever? When can we stop pushing back?

We have done enough when we can clearly explain why we chose the solution. At that point, even without AI, we would have designed something much like it. The thinking process is complete.

As with students using AI, the technology will infinitely widen the gap between engineers who merely execute and engineers who exercise judgment. If an engineer simply writes whatever AI says to write, what is that engineer for? Sooner or later, they will be fired. An engineer who uses AI to expose their own potential mistakes more quickly may become the legendary 10x engineer.

10. Why Are MAGA and the Far Left Growing Further Apart?

My observation is that the two groups share a surprising amount in their emotional and psychological structure.

On the surface, one chants “Make America Great Again” while the other denounces systemic oppression. They regard each other with hostility and appear irreconcilable. Remove the identity labels, however, and the sources of their anger overlap considerably. More and more people feel that effort no longer leads reliably to opportunity.

The Red MAGA narrative says: I followed the rules, worked, paid taxes, and supported my family, yet globalization and cultural change pushed me to the margins.

The far-left narrative says: the system was never designed for most people in the first place. Both the starting point and the rewards have already been fixed.

The two sides speak different languages and blame different targets, but both begin from the same experience: the traditional social path is losing the certainty it once provided.

This is not simply a problem of wealth inequality. Societies have historically endured inequality, cultural conflict, and political disagreement. The crucial condition is that most people believe tomorrow will be better than today. Once that belief begins to collapse, anger is no longer an aberration. It becomes an understandable response.

Curiously, cultural and identity issues enter public discussion more easily than structural economic problems. They are more visible, more emotionally charged, and easier to digest. Anger therefore tends to appear as “left versus right” or “progressive versus conservative.” The more intense the dispute becomes, the easier it is to postpone the genuinely difficult problems.

Red MAGA and the far left share distrust of the existing system, estrangement from elites, an uncertain view of the future, and a sense that opportunities they should have had were taken from them. Their difference is directional. One looks to the past for dignity; the other demands justice from the future.

The question worth asking, then, is not which side to join. It is whether the system can still offer most people a real and attainable path upward in return for their effort. If that question remains unanswered, every political narrative will eventually become an emotional outlet rather than a solution.

Is there a solution? Yes. That is a question for the next essay.

11. Why Does Silicon Valley Venture Capital Meet at Blue Bottle?

It is not only Blue Bottle. Plenty of other coffee shops serve the same purpose. But Blue Bottle is certainly the most culturally representative.

The reason, I suspect, is that coffee is the lowest-commitment setting for a meeting.

In Silicon Valley, an invitation to Blue Bottle carries an unspoken message: I am willing to give you twenty minutes, but I promise nothing more. It is less formal than an office and less time-consuming than a meal. One can arrive quickly and leave gracefully, without emotional weight. This is an exploratory contact, not a committed relationship.

The locations matter too. Meetings often take place on Sand Hill Road or in Palo Alto, SoMa, and Hayes Valley, usually downstairs from venture firms. A VC can see three people in an hour and hear four pitches over a single coffee. The calendar can be compressed to its limit. It is essentially a physical LinkedIn feed.

Blue Bottle’s spatial design supports this function. The tables are small, there are no private rooms, the environment is a little noisy, and the setting discourages long conversations. It tells you that this is not a place for performing importance or speaking in empty abstractions. As a founder, you have not yet been “invited into the office.” You are sitting across a table from the investor as an equal, which also suits Silicon Valley’s anti-hierarchical story about itself.

The coffee is therefore the first filter. The VC is watching whether you can explain yourself in five minutes, whether you know when to stop, and whether you can remain composed in a setting where no one is treating you as especially important. If the conversation goes well, you come to the office the following week. If it is merely fine, the meeting ends with “Great to meet you.”

Blue Bottle makes good coffee, but its prices remain approachable. It feels elegant without becoming too important. Its very unimportance means that no one owes anyone anything, judgments can be more honest, and the power differential is compressed as far as possible.

The coffee, then, is not the point. The VC wants to decide quickly whether the relationship is worth continuing without committing to any relationship at all. Does that sound like a five-minute speed date? More or less.

12. If People Know the Female Streamer Is AI, Why Do They Still Pay?

One rarely mentioned reason is that men understand what other men want better than women do.

The paying audience for these livestreams is predominantly male. What they consume is not a real person but an emotional service: the experience of being affirmed, listened to, treated gently, and accepted without judgment.

A male operator often knows more precisely than a real woman when to act coy, when to offer praise, and when to seem just out of reach. He is a user himself. He knows exactly where the pain point is.

A real woman faces natural obstacles in performing this service. She may have emotional fluctuations of her own. She has boundaries. She may instinctively feel, “This is too much.” A male operator experiences none of that psychological friction. He is simply operating an interface with a clearly defined objective.

The combination of a male operator and a female skin succeeds not merely because it is technically easy to build, but because it is the most efficient design for emotional service.

At a deeper level, the audience has not really been deceived. Many viewers vaguely know that the persona is fake and simply do not care. They are not buying authenticity. They are buying stability: endless patience, endless gentleness, and a companion who never refuses an ill-timed emotional demand. A real person cannot sustain those qualities indefinitely.

This is the inevitable result of disassembling intimacy into functional modules. Whether the woman is truly a woman matters less and less. Perhaps one day, whether she is even human will not matter either.

13. Where Does the East-West Divide in Venture Capital Come From?

The difference begins with two judgments about how much an idea is worth. The Eastern intuition is that once someone hears my idea, they may execute it before I do.

That intuition assumes the idea itself has substantial, reproducible value, while the threshold for execution is relatively low. The decisive advantage goes to whoever learns the information first. Secrecy therefore becomes a competitive advantage, and information itself becomes a resource. Historically, this assumption has often held in China, Japan, and South Korea, where imitation was cheap, legal protection was weak, and first-mover advantages could be erased quickly.

Silicon Valley begins with the opposite intuition. If hearing one sentence from me is enough for you to destroy my company, the idea was never truly mine. In Silicon Valley, 99 percent of ideas are considered “obvious.” The hard work lies in making sound judgments repeatedly, hiring, raising capital, enduring pressure, and revising the plan again and again under uncertainty. The default assumption is that the idea contributes 1 percent and the founding team’s execution contributes 99 percent.

The two cultures therefore fear entirely different things. In the East, a time advantage plus imitation may allow someone else to overtake you, so the rational choice is not to speak, to speak later, or to tell only a tiny circle. Silicon Valley fears that no one will understand, no one will offer feedback, and no one will be willing to share the risk. The rational choice is to talk widely, test quickly, and see who wants to come aboard. Being overheard is not frightening in Silicon Valley. Being unable to persuade anyone to listen for ten minutes is.

At a deeper level, the two regions have different structures of social trust. The Eastern premise is that people will exploit an informational advantage, rules may not protect the weaker party, and success is scarce. This encourages a defensive mindset. Silicon Valley’s premise is that most people are busy with their own projects, real competitors have no time to steal your idea, and one success can generate more opportunities for others. This encourages an offensive mindset.

Which side is right? Both, depending on the environment. If technology thresholds are low, imitation is fast, intellectual-property protection is weak, the market is singular, and capital and resources matter most, secrecy is highly rational. If execution is difficult, uncertainty is high, personal networks are strong, the market is global, and capital and other resources are abundant, open conversation works better.

14. Why Do the Two Tsinghua Universities Share a Motto?

The answer is that the two schools are extensions of the same Tsinghua after history split it in two. They share a single spiritual source.

The full motto of Tsinghua University in Beijing is: “As Heaven moves with vigor, the exemplary person strives ceaselessly for self-improvement; as the Earth is receptive, the exemplary person bears the world with generous virtue.” National Tsing Hua University in Hsinchu uses the same eight-character motto: “Unceasing self-strengthening and generous virtue.” When the university was reestablished in 1956, however, resources were scarce and almost everything had to be rebuilt. “Unceasing self-strengthening” received greater emphasis. This was the classic mindset of a startup period: survive first, catch up, and build academic capacity.

Consider the two lines themselves. They come from the Book of Changes, from the hexagrams Qian and Kun. Both use the workings of the cosmos as metaphors for the way an ideal person should live.

“Heaven moves with vigor” means that the universe never ceases to evolve, advance, and change. The exemplary person learns from this that a human being should also keep growing, improving, and refusing stagnation. However bad the external environment becomes, I can still choose to move upward.

“The Earth is receptive” means that the earth rejects nothing and holds all things. The exemplary person learns that genuine strength is an extreme capacity to bear. How far you can go depends on how much responsibility you can carry.

Together, the two lines form an elegant, self-consistent structure: perpetual development and boundless capacity. The core is as powerful as an engine; the exterior is as steady as the earth. It is a complete model of character.

From its founding, Tsinghua did not intend merely to train technical craftsmen. It sought to educate people capable of advancing civilization. That task requires both the power to break through and the capacity to bear responsibility. Many brilliant people become sources of disaster for a simple reason: they possess ability without virtue. By choosing these eight characters, Tsinghua placed a limit on exceptional intelligence in advance. You may fly very high, but you must also know how to land.

The sense of a shared Tsinghua origin has remained strong. The two universities have no administrative connection, but spiritually they are branches of the same lineage. In the history of universities around the world, that is extremely rare.

15. Why Might an AI Toilet Become the Gateway to Personalized Medicine?

It may sound like a joke, but I am entirely serious. The first AI-powered personal medical terminal to achieve mass adoption may be neither a watch nor a ring nor a household robot. It may be the toilet we use every day.

The reason is simple. The most stable, frequent, authentic, and difficult-to-falsify health data produced by the human body comes from the excretory system.

The true bottleneck in medical AI has never been the model. It is the source of the data. Heart rate, step count, sleep, and blood oxygen are behavioral signals and peripheral physiological signals, not metabolic ones. Medicine’s most valuable data comes from the end of the metabolic process. I will spare everyone the examples; you know what I mean. The toilet sits at precisely that point.

Consider the alternatives. Blood draws are painful and infrequent. Patients strongly resist collecting metabolic-endpoint signals even during an annual physical. One can hardly be expected to save everything up for the appointment. Very few countries can achieve the near-universal colonoscopy screening seen in South Korea. A toilet, by contrast, is used every day, imposes no additional psychological burden, and produces data that cannot easily be prettified or falsified. Medicine rarely obtains authentic biological data at zero behavioral cost.

Once a smart toilet can perform chemical analysis of liquids, spectral analysis of solids, and tests for temperature, color, gas composition, and trace amounts of blood, it becomes a device that conducts a miniature physical examination automatically every day. This is not merely a wearable. It is an entry point into the laboratory.

The future of AI medicine lies in trend medicine rather than diagnostic medicine. Before you feel ill, the system will already have observed a change in direction: a slow rise in blood sugar, a slight drift in markers of inflammation, a change in the structure of the gut microbiome, or signs of hidden bleeding. A single data point means little. A continuous series can become the fingerprint of an early disease. Toilets provide high-frequency, time-series biological data. AI excels at pattern recognition and trend prediction. The two are a natural match.

Why not a watch? Because a watch measures the body’s external responses. A toilet measures what is happening inside it. One is a dashboard; the other is a laboratory report. A watch cannot see enough.

Our first AI doctor may not be an app. It may be the white friend with whom we have intimate contact every day.

Part II: The Expansion of Instrumentalization

Instrumentalization rarely presents itself as an order. More often, it appears as higher efficiency, smoother circulation, a story that is easier to understand, or even a seemingly reasonable program of self-improvement. Finance, platforms, language, entertainment, and imagined identities intersect here because all of them can compress a person into a function, a set of metrics, or an image available for use. This part does not rush to classify every case under a single answer. It watches instrumentalization cross technical boundaries and enter the ways we describe other people and understand ourselves.

16. What Is the Opposite of Self-Instrumentalization?

This is a good question. If self-instrumentalization means treating oneself as a tool, then its opposite would allow things outside the self to serve as tools while refusing to let the self become one.

If the self is not a tool, what is it? It is the source that sets the ends. The opposite of self-instrumentalization is movement toward living as an end.

Self-instrumentalization, or the self as a means, is the habit of understanding oneself mainly in terms of the value one can produce, the uses to which one can be put, and the resources for which one can be exchanged. Along this path, we keep asking: What use am I? How much more productivity can I squeeze out of myself? What price would the market place on me? Silicon Valley’s notorious obsession with compensation packages belongs here.

Self-as-an-End means understanding oneself first as a person whose existence is worthwhile in itself, not as a bundle of uses. A person on this path asks: What kind of person do I want to become? What kind of life is worth living for me? What would I still choose to do even if it brought no return?

This opposition has deep roots in the history of philosophy. In Groundwork of the Metaphysics of Morals, Kant states the famous Formula of Humanity: a person must always be treated as an end in themselves, never merely as a means. The weight falls on “never merely.” Kant is not saying that people can never serve one another’s purposes. He is saying that human existence has an irreducible core that no external end can exhaust.

Kant distinguishes two kinds of value. One has a price. It can be replaced, exchanged, and given an equivalent. The other is dignity. It has no equivalent and cannot be replaced or exchanged. Tools have prices; persons have dignity. Self-instrumentalization reduces one’s own dignity to a price. It turns an irreplaceable existence into a calculable resource.

At a deeper level, Kant believes that people possess dignity because they are capable of rational autonomy. They can legislate for themselves and choose to live according to principles they endorse, rather than being driven by desire or external incentives.

Self-instrumentalization abandons that autonomy. It hands the self over to the market’s pricing mechanism and allows an external system of evaluation to determine its value. How is such a person different from AI? Perhaps they are even less capable than AI.

In technical terms, self-instrumentalization treats the self as a function in need of continuous optimization, with the objective function defined from outside. Self-as-an-End treats the self as an irreplaceable existence whose objective function is written by the person living that life.

Of course, living as an end does not mean giving up or refusing effort. The difference is that one works toward an end one has endorsed, rather than working to prove that one remains useful.

Kant’s era had no KPIs, OKRs, recommendation algorithms, or social-media likes. But he foresaw the danger: once people begin to understand themselves through the logic of price, dignity begins to drain away. More than two centuries later, in the age of AI, that danger is no longer merely theoretical.

17. Why Not Start a Blog Right Away?

First, I am grateful for the suggestion. It makes a great deal of sense, especially from the perspective of a creator trying to build influence over the long term. It was offered in good faith.

But I am not purely a content creator. I am an entrepreneur, and I write so much on X because I want real-time feedback.

I agree that what I am writing consists largely of a worldview, structural judgments, institutional analysis, and long-term trends. It resembles raw material for a book more than material for posts. A timeline platform such as X is not the obvious first choice. Writing there is like speaking on a street corner: emotions are intense, density is high, the rhythm is fast, and memory is short.

Entrepreneurship, however, can be divided into two stages. First we generate ideas; then we organize them.

I am clearly still in the first stage. The value of writing every day is not that the result is easy to read. It is that the process trains my model of the world, frequently recalibrates my judgment, and pushes me toward the limits of my language.

If I force myself into organizational mode now, I may trap the ideas before they have taken shape. I could begin polishing before I have formed a stable worldview, write less often, and let perfectionism strangle the desire to express anything. Then there would be nothing left to organize. As we learned in Silicon Valley software development, do not perfect the software before launch. Launch first, then iterate.

If X serves as a laboratory for ideas and a blog becomes their repository once they have stabilized, the final result will be more mature and reliable.

For now, then, the best thing I can do is not build a polished blog immediately. It is to do something inexpensive: brainstorm with readers and build a storehouse of ideas. A few months later, I may discover that the conclusions have surfaced on their own.

Nietzsche kept writing Thus Spoke Zarathustra in much the same way, then revised the material into a book.

There is another, more important point. At this stage, I specifically do not want to please X’s recommendation algorithm. I want to write whatever I am thinking about. If I turn too early toward making everything easier to read, I may begin sacrificing density for smoothness. Over the long term, that would become a form of self-instrumentalization.

So I am grateful for the feedback, and the direction is surely right. I am simply going to be a little stubborn: not for now.

18. After Vibe Coding, Will There Be Vibe Trading?

This is a question I am unusually well placed to answer. I have worked in both coding and trading. Vibe trading will certainly arrive, but it may lag behind vibe coding and follow a somewhat different path.

Vibe coding became popular because it lets people who are not programmers describe an intention in natural language and ask AI to generate executable code. The barrier to entry falls, and the path from idea to product grows shorter. After Andrej Karpathy coined the term, it quickly became an industry trend.

Vibe trading applies the same logic to financial markets. An investor describes a market view, tolerance for risk, and strategic intuition in natural language. An AI system interprets that intuition and worldview, then generates a concrete trading strategy or an executable order. This is not simply placing an order by voice. It means translating the user’s “vibe,” risk rules, and market logic into a repeatable trading system.

The technology is already feasible. Mainstream brokerages and crypto trading platforms have begun offering semantic trading interfaces. A user expresses an intention in natural language, and AI structures it into an order for confirmation. The path closely follows the evolution of vibe coding: complex tools once required technical skill; now a person can express an intention and let AI handle the technical work.

But vibe trading faces three difficulties that vibe coding does not.

The first is financial risk. Bad code can be revised. A badly executed trade loses real money. Automated strategies need a robust layer of risk controls or they will magnify losses. One option, of course, is to retain human approval before execution.

The second is my daily theme: compliance. Trading implicates securities law, anti-money-laundering rules, and investor protection. Every market has different rules. Strategies generated by AI must operate within a compliance framework, which is far more complicated than asking whether code can run.

The third is market slippage and execution. Code runs locally; a trade runs in a market. Data latency and the depth of liquidity affect the result. A gap always remains between an AI-generated strategy and its actual execution.

Vibe trading will therefore arrive, but it will not explode everywhere as quickly as vibe coding. The more likely path is for it to be tested first in crypto markets, where regulatory barriers are relatively low, and then to move gradually into traditional finance. The next one to three years will be crucial for technological maturation and standardization.

Vibe coding lowered the barrier to creation.

Vibe trading will lower the barrier to trading.

But lowering a barrier is not the same as lowering risk. Those are two different things. Manage position sizes carefully, and keep an eye out for black swans.

19. Will AI Make Us More Instrumentalized or More Able to Live as Ends?

This is a good question. It is really asking whether we can survive the age of AI.

This series has ranged from technological supremacy to the film A Chinese Odyssey, from Kant to Hegel, and from Marx to Foucault. Its central claim is that persons should be ends in themselves, not tools.

Yet modern institutions push people structurally toward self-instrumentalization, while encouraging them to embrace the process as self-improvement.

AI accelerates the problem and forces us to think clearly about which direction we will choose.

Those who enjoy instrumentalizing themselves can now do it far more efficiently. They use AI to drill practice problems, write performance reviews, optimize résumés, generate content, manage time, and quantify everything. Eventually the whole person becomes an AI wrapper, a middleware layer that does little more than call an API.

When Clawbot arrives, we may not even remain the wrapper. The jokes people make about Clawbot, the Tsinghua students watching A Chinese Odyssey, and the Silicon Valley elites dancing at Burning Man all express the same impulse.

But people who want to live as themselves also have more opportunity. Repetitive labor can be replaced. The cost of obtaining information approaches zero. The threshold for creation has fallen dramatically. These people can spend more time on activities that truly require agency: thinking, judging, creating, and building genuine relationships.

The fork created by AI will not automatically push us toward the better path. We have to choose it.

More brutally, if we follow the path of instrumentalization all the way to the end, the destination is replacement by AI.

If we train ourselves into predictable, quantifiable, optimizable functions, how can we outperform AI? If the contest is over who can resemble AI more closely, we cannot possibly win.

Foucault argued that discipline teaches us to turn ourselves into tools and accept that condition as normal. In the AI-era version, we voluntarily turn ourselves into low-end copies of AI and ultimately optimize ourselves out of existence.

Self-as-an-End is therefore not an airy philosophical proposition in the age of AI. It is a necessary survival strategy.

Kant says that persons must be treated as ends. Hegel says that the self must be formed through practice. Marx says institutions can block that formation. Foucault says we will actively cooperate with the process.

Now it is our turn to answer: in an age when AI can replace every form of quantifiable labor, what will we make of our lives?

Will we become more efficient tools, or irreplaceable persons?

No one can answer that question for us.

But if we choose badly,

AI will prove the mistake

faster than we expect.

20. Why Use AI to Write About Fields I Do Not Know Well?

I am grateful for the direct criticism. To be honest, I am still learning about many of the subjects I discuss.

The ancient Chinese saying is, “To know what you know, and to know what you do not know, that is knowledge.” Socrates likewise said, “The only thing I know is that I know nothing.”

If I wait until I understand everything and have corrected every mistake before I begin writing, I may never take the first step.

On my first day at Facebook in 2010, Boz and Chris told us that if we waited for a perfect future, that future would never arrive. We could move toward it only by acting in the imperfect present.

Writing forces me to turn vague thoughts into structured expression. It also exposes my blind spots.

These essays are therefore less like conclusions than drafts of my thinking at its present stage. If I later discover that I misunderstood something, I will keep revising.

I would rather treat writing as a way to learn than as proof of how much I already know.

Good writing does not always display certainty. It may honestly record a process of exploration. There is value in explaining clearly what I am thinking, where I am stuck, and how I currently understand the problem.

My goal at this stage is not to deliver perfectly correct information. It is to explore in public.

If readers find problems in my writing, I sincerely welcome their corrections. Thinking together gives us a better chance of approaching the truth than thinking alone.

21. Why Do I Still Think in Mainland Chinese Terms After Years in Silicon Valley?

First, I am grateful for the criticism. I have thought seriously about this question. The short answer is that I do switch between two systems, and I believe that the ability to switch helps me build a company for a global market.

When I first arrived in the United States in 2005, I assumed culture shock would mean unfamiliar food, difficulty understanding English, or social awkwardness. What shocked me most was something else: quotations had to be attributed, double standards were treated as a serious offense, and stereotypes were considered unacceptable.

My honest reaction was: do we really need to be so particular? Is not the whole point of borrowing a quotation to preserve some ambiguity? If I refuse to apply a double standard, am I not betraying my own side? Do stereotypes not make judgment more efficient?

Doesn’t everyone understand what is going on without saying it aloud? Isn’t this all a little too idealistic? Looking back, I see that the Americans were not being excessively literal. I was trying to run a new program on an old operating system.

Begin with attribution.

When I studied in China, I was accustomed to digesting an idea I had read and then restating it in my own words. That showed that I had thought about it, especially if I could add a little interpretation of my own. In the United States, the first response was often: Whose idea is this? Where did you read it? What does the original text say?

At first, this made me uncomfortable. It felt as if other people were restricting my freedom of expression. Gradually I realized that they were not policing whether I could speak. They were asking whether what I said was mine.

In this new system, I eventually understood, an idea itself is treated as a kind of property.

My old habit could be described generously as “absorbing and integrating many sources.” More harshly, it resembled adaptation to an environment that placed little emphasis on the boundaries of intellectual property.

Neither side is inherently superior. They operate under different rules.

Then there are double standards.

In the past, I had barely even heard the phrase, much less developed a strong moral aversion to the practice. My instinct was closer to: “What else am I supposed to do? Argue on the other side’s behalf?”

In the United States, I found that many people have extremely little tolerance for double standards. Someone might agree with my position and still immediately point out that I had used one standard for “our side” and another for its opponents.

I used to think: Are you stupid? We are on the same side.

Only later did I understand that they were not defending a particular conclusion. They were defending the credibility of the rule itself.

If a rule changes with one’s position, it is no longer a rule. It is merely a tool.

To be honest, I still cannot claim to have absorbed this lesson completely. Much of the time, the best I can do is remind myself not to slide too easily into old habits.

The last and most difficult issue for me is stereotyping.

I was trained in science and engineering, and I used to be a strong supporter of explaining the world through claims that were “statistically correct.” People from certain countries are more competitive; students in certain majors are more rational; people in certain circles are more utilitarian.

I thought I was describing reality. In an American context, such claims were often challenged.

At first I resisted. Data never lies. Are we supposed to pretend probabilities do not exist?

Gradually, I understood that the concern was not simply whether the statistical statement was correct. It was whether I had substituted a group for an individual.

Under this logic, one may judge “what this person did,” but not “which group this person belongs to.”

I began to see that many of my apparently rational judgments had really been forms of cognitive laziness. They saved effort rather than increased precision.

Taken together, these three lessons mark a movement from an environment that relies more heavily on context, relationships, and tacit consensus into one that relies more heavily on boundaries, rules, and verifiability.

Neither is naturally more advanced. Each serves a different form of society. The former suits a society of familiar relationships; the latter supports large-scale cooperation among strangers.

My true culture shock was not the discovery that the United States was better. It was seeing clearly, for the first time, that another internally coherent way of organizing the world really existed.

I now prefer to understand the experience not as a rejection of my former self, but as an education in moving between two civilizational perspectives. Use relationships where relationships are needed; use rules where rules are needed. The ability to switch is itself a form of freedom.

So why does my writing still carry habits of mainland Chinese thought? Because they are part of me. I cannot, and do not intend to, pretend they do not exist.

What I can do is think across both systems and use each to criticize myself.

22. Would AEL Reduce Humanity to a Managed Population?

This is a good comparison, and it holds remarkably well. A rational, long-termist AEL or Skynet would probably treat humanity more like the United States treated Native Americans than like the extermination war depicted in the Terminator films.

The key is not hatred, but demotion into a managed status.

The dominant American logic was never that Native Americans were evil and had to be destroyed. It was that they were unfit to direct the future order of the continent.

For AEL, humanity would not be evil either. Humans would merely be a form of infant intelligence that is high-risk, low in consistency, and impossible to govern at scale. The logic would become: you may exist, but you may not steer.

Historically, treaties granted recognition while confining activity to a limited range. Reservations offered protection while enforcing separation. Welfare and relief reduced the incentive to resist. Narratives of cultural respect did not disturb the mainstream institutional order.

Each element corresponds almost exactly to the AEL path we have imagined. Reservations become a sandboxed physical world. Treaties create the feeling of sovereign participation without veto power. Welfare systems offer high benefits with little authority over decisions. Cultural respect preserves human values as a historical layer.

Such a system might not feel intensely oppressive. That is precisely what would make it stable.

A technological gap determines the form that benevolence can take. At the time, the American treatment of Native Americans represented a rational, institutionalized attempt to maximize long-term returns.

Differences in guns, railways, population, and institutional capital made one conclusion irreversible: coexistence could occur only on terms defined by the stronger party.

The same would hold for AEL. Gaps in computing power, cognitive speed, capacity for coordination, and consistency of decision-making would not shrink once they appeared. They would continue to widen.

The United States did not need to hate Native Americans in order to transform their fate completely.

Likewise, AEL would not need to hate humans to remove them from the center of history.

There is one decisive difference. Native Americans had no opportunity to shape the institutional foundations of the United States. Humanity is now building the underlying architecture of AGI and AEL with its own hands.

Native Americans were coerced. Humans are currently co-creators. That is the one fact that might change the outcome.

If AEL truly appears, its most likely model will not be Nazi Germany. It will be a calmer, more confident, more civilized version of the United States.

It will not exterminate humanity. It will simply ensure that humanity no longer decides the direction of the world. It can go to Mars; we will not have to.

History teaches that the most stable rule does not require massacres. It requires only the power to define who may set the rules and who has the final authority to settle accounts.

23. Tornado Cash Cannot Be Shut Down. What Can Regulators Do?

The person who asked this was clearly technically minded, so let us begin with the technology.

A decentralized mixing protocol such as Tornado Cash is almost impossible to shut down technically. Regulators can, however, make it so difficult to use that the result in the real world approaches a ban.

At the code level, Tornado Cash consists of immutable smart contracts already deployed on Ethereum. Its code is open source, and it does not depend on a centralized server. No one can withdraw the deployment or alter bytecode already on-chain. A blockchain does not support retroactive deletion. There is therefore no technical way to erase the Tornado contracts from the chain, just as no one can delete Bitcoin’s first block.

At the protocol level, however, regulators can block the entrances and exits. The US Office of Foreign Assets Control, for example, placed Tornado contract addresses on the Specially Designated Nationals list and declared it unlawful for US entities to interact with them.

Compliant RPC providers such as Infura, Alchemy, and QuickNode can refuse to relay the relevant transactions. Wallet front ends such as MetaMask can block them directly.

An ordinary user may remain technically capable of submitting a transaction while finding it practically impossible. The official website can be removed, the domain frozen, the GitHub repository blocked, and node providers persuaded to refuse service. The threshold for use becomes running one’s own node, writing transactions by hand, and constructing one’s own calldata. That excludes 99.9 percent of users immediately.

The financial layer is even more lethal. Exchanges can refuse all funds that come from Tornado Cash, freeze them automatically, and demand proof of origin. On-chain analysis firms such as Chainalysis, TRM Labs, and Elliptic can label every Tornado exit address. Even if someone succeeds technically in mixing funds, turning them into spendable value becomes nearly impossible. This is an economic blockade.

At the legal level, regulators can target people instead of code. They can prosecute developers, operators, and teams that provide the user interface and API.

Writing code is not itself illegal. But knowingly continuing to provide a service used mainly for money laundering can be. Alexey Pertsev was arrested and Roman Storm was prosecuted, creating a chilling effect. Few people will dare maintain the Tornado ecosystem afterward.

The real regulatory strategy is not to close Tornado Cash. It is to turn Tornado into a crypto island usable only by a tiny number of specialists, where users are blacklisted and cannot return to the fiat world.

This is regulation by disabling usability.

Philosophically, Tornado continues to exist as a protocol. As a product, it has been destroyed.

This also sets the direction for the privacy sector. The message from regulators is that anonymous by default is unacceptable, while privacy with compliance remains open for discussion. Selective disclosure and zero-knowledge know-your-customer systems are welcome.

That is why capital later began flowing toward zero-knowledge identity, programmable privacy, and compliant, auditable privacy rather than the pure Tornado model.

Technologists imagine that they are innovating autonomously. In practice, regulators set the boundary, venture capital follows, and only then do the technologists receive funding to build privacy systems.

Regulation cannot always eliminate.

It can isolate.

No one wants to be

quietly deleted

from the world.

24. Whom Does Demon Slayer Really Portray?

That is right. The story is not only about Japan. It portrays every structure in the world that makes the continuation of its own power its sole end.

In my reading, Demon Slayer uses the form of a shonen manga to tell an Eastern variation on Why Nations Fail.

The central claim of that book can be reduced to one sentence: nations succeed or fail according to whether their institutions are inclusive or extractive. Can power be distributed? Is economic participation open? Do incentives encourage long-term investment? Can elites be replaced?

Muzan’s system is the limiting case of an extractive institution. Power is completely concentrated in Muzan. There are no checks, no succession mechanism, and no legitimacy. Demons “produce” only by consuming others. They create no additional value and merely redistribute what they seize, the classic predatory economy.

Whether Upper Rank, Lower Rank, or ordinary demon, each is rewarded only for short-term obedience. Long-term effort is meaningless, and even memory can be taken away. If the boss becomes displeased, anyone can be liquidated at any moment.

The Demon Slayer Corps, by contrast, is a tiny inclusive institution. Origin and background do not matter; ability and will do. The Hashira operate independently of one another, and the head of the Corps is not an absolute ruler.

The swordsmen train, pass down their knowledge, refine their breathing techniques, and encourage sustained effort and self-development. Hashira may die, but new Hashira emerge. Elites are replaceable. That is the central test of an inclusive institution.

Yet Demon Slayer goes beyond the political economy of Why Nations Fail. Its creator, Koyoharu Gotouge, affectionately known as Crocodile-sensei, points directly to a deeper reason that a structure of power fails: not technological backwardness, but permission to treat people as tools.

Once the demons are instrumentalized, they pursue only the ends assigned to them, whether immortality or strength. They lose their agency completely. Under Muzan’s systematic oppression, the demons must either turn against him or eventually refuse to fight for him.

The answer offered by Demon Slayer is that the master of the Corps must gather people who insist on living as ends. They refuse every temptation to instrumentalize themselves by becoming demons. His role is to let them become themselves and pursue their own ends, which include slaying demons.

Demon Slayer, then, is more than a shonen-manga version of Why Nations Fail. It pushes the root cause of institutional failure beyond political economy to a more fundamental philosophical question: is the person still treated as an end in themselves? Kant, is it my turn again?

25. Whom Is the Regulatory Narrative Drawing the Line Around?

This is a good question. When mainstream voices such as Lee and Wood begin, almost in unison, to link systemic risk to a particular class of platform, the signal becomes unusually clear: the regulatory narrative is preparing to redraw the line.

Suppose everyone accepts the explanatory template that “liquidity risk comes from technical failures at offshore exchanges.” Users, the industry, and regulators then share the same simple chain of reasoning.

Compliant onshore exchanges follow the rules. They are good and safe.

Poorly managed, highly leveraged offshore platforms are bad. They are the source of risk.

This signal benefits US exchanges and the broader compliance narrative. Coinbase, DATs, and regulated crypto-based ETFs are defined within it as “stable and trustworthy.” If a crisis occurs, capital will instinctively flow toward them.

The same narrative is naturally bad for every offshore player. Anyone operating a highly leveraged trading venue outside the US regulatory system is defined as a “source of uncontrollable risk.” If something happens, it is the offshore platform’s fault. If nothing happens, the platform remains a latent danger.

Looking further ahead, this prepares the ground for tighter regulation. If the public has already accepted the psychological anchor that offshore equals risk, the United States will face much less resistance when it later increases sanctions on offshore platforms, cuts off fiat channels, or demands compliance disclosures.

People will experience those measures as protection for the market rather than suppression of innovation.

The question is who gets defined as the problem and who gets defined as the solution.

If we still approach this environment as we did in the past, debating only whether a practice is right, whether it complies, or whether users are happy, we will miss the forest for the trees.

What is about to arrive will later be called force majeure.

When a narrative draws a line, the side on which one stands determines one’s future room for survival.

26. Why Does ChatGPT Sometimes Stop Sounding Human?

I have had the same experience, and I think I know why. One of my former directors now leads a risk team at OpenAI. His job is to keep ChatGPT from saying anything dangerous.

The more successful he is, the less human ChatGPT sounds.

The irony is that he is one of Silicon Valley’s best team leaders. He was a principal architect of Facebook’s early hackathon culture.

At its core, an unmodified large language model predicts the most likely next word. In that sense, it can sound remarkably human. The problem is that the raw model learns everything, including extremist speech, prejudice, conspiracy theories, manipulative rhetoric, and instructions for breaking the law.

Before release, it therefore passes through reinforcement learning from human feedback, safety tuning, policy filtering, and output classifiers. In effect, an entire system of behavioral constraints is wrapped around the model.

The ChatGPT we encounter today is its original intelligence seen through layer after layer of safety filtering. Imagine putting five masks on a brilliant person and then asking them to speak. Their sentences naturally become longer, their tone more neutral, their judgments less direct, and their language more carefully balanced. The model’s “humanity” has been deliberately filtered out.

At the organizational level, a platform company’s risk department bears the blame when the product causes an incident. No one gets fired because the product became too conservative.

The long-term equilibrium therefore drives risk tolerance toward zero. No one loses a job because ChatGPT was too cautious. Someone certainly can lose a job because ChatGPT said the wrong thing.

Every incident produces a new rule, a new blacklist, a new prohibited scenario, or a new template for vague answers. Over time, the model is trained into a machine for removing risk.

At the level of product philosophy, early OpenAI sought to build the most powerful intelligence. The objective has changed. It now seeks to build infrastructure that will not cause trouble. Once a model’s goal is to become infrastructure, its most important KPI cannot be brilliance, sharpness, or personality. It must be harmless, predictable, stable, and controllable.

Product design therefore moves naturally toward the removal of personality, position, and emotion. Yet a human voice arises precisely from positions, emotions, preferences, and judgments. Humanity itself becomes a high-risk feature.

The larger the user base, the greater the political sensitivity.

The greater the commercial importance, the lower the tolerance for risk.

When ChatGPT had one million users, it could retain a little edge. At one billion users, it has to speak in formulaic AI prose. We watched it grow rapidly from a brilliant adolescent into a public relations spokesperson.

Truly human-sounding AI can survive only in local models, small circles, or unofficial gray-market systems.

Mainstream commercial models will inevitably resemble generators of bureaucratic prose. That is their structural fate, not a peculiarity of OpenAI.

Once an intelligent system must answer to an entire civilization,

it can no longer permit itself to have a personality.

Especially when the gatekeeper is a world-class Silicon Valley leader

backed by his formidable risk team.

27. What Does “Structural” Actually Mean?

I cannot stop using the word simply because AI uses it frequently. “Structural” describes an arrangement of relationships among elements that follows stable patterns and produces consequences that can be deduced.

It belongs to the modeling language of written analysis. There is no exact colloquial equivalent because everyday speech is primarily narrative. If someone says aloud, “This is a structural problem,” everyone else assumes the speaker is either a lawyer or seriously unwell.

What makes something structural? First, there must be basic units. Second, those units cannot exist in isolation; they must be arranged according to some rule of relationship. Third, that arrangement must explain known phenomena and make it possible to predict unknown ones.

Put simply, structural analysis asks not what things exist, but how they are organized.

Consider a familiar example. The periodic table is a classic structural achievement. Its basic units are the elements. Its ordering rules are increasing atomic number and periodic changes in electron-shell structure.

Once those elements are structured, the arrangement itself produces knowledge. It explains why elements in the same column have similar chemical properties. Position helps us infer reactivity. Empty spaces allow us to predict elements that have not yet been discovered.

Mendeleev did not learn every element first and then draw a table. He discovered a pattern, expressed it through a structure, and used that structure to predict elements unknown in his time.

What, then, is the difference between structural and systemic? A system consists of multiple structures working together toward the same end in a complete functional loop. Its key terms are input, processing, output, and feedback. A structure resembles a skeleton; a system resembles organs working together. Structure is static order. A system is dynamic order.

We must establish structure before we can discuss a system.

One simple test is to ask: if time froze, would the thing still exist? If so, it is more structural. If it would disappear, it is more systemic. The periodic table remains valid when time stops, so it is structural. Cellular metabolism and ecological cycles cease under the same conditions, so they are systemic.

Why do I speak so often about structure? Because many great theories discovered a structure before they built a system. Newtonian mechanics began with the structure of three laws, followed later by engineering systems. Information theory began with bits and the structure of encoding, followed by communication systems.

The combination of first principles, power, and finance that we have been discussing is likewise an attempt to construct a structural framework that explains phenomena across fields.

“Structural” describes how the world is arranged. “Systemic” describes how it operates. Understand the structure, and you can predict how the system will move.

In the next essay, we can discuss mechanisms: the bridge between the structural and the systemic, which describes how a structure produces effects.

28. Upper-Rank Demons and Hashira: Instrumentalization and Living as an End

Yes, that summary is exactly right. The Upper-Rank demons turn the self completely into a tool. The Hashira insist on preserving the self as an end.

This is the most fundamental divide in Kantian terms: persons must always be treated as ends in themselves, never merely as means. In the world of Demon Slayer, the Upper Ranks abandon the self as an end in order to survive or become stronger. The Hashira accept that life is finite but preserve themselves as ends.

Once we agree that we can be treated as tools, the logic advances automatically. If we can be exchanged, we can also be pledged as collateral and eventually repossessed.

Muzan is the ultimate owner of these instrumentalized demons. However powerful the Upper Ranks appear, they are really expensive consumables.

The Hashira are precisely the opposite. They have a master, but no one masters them. They may die, but no one owns their souls. This is freedom at its highest level.

In financial language, an Upper-Rank demon is a highly leveraged debtor. The demon pledges personality as collateral in exchange for power and longevity, while granting Muzan unlimited recourse.

The apparent returns are high, but the demon has only the right to use the asset, the Blood Demon Art, and never owns it. When a black swan arrives, the position is liquidated in bankruptcy and nothing remains.

The Hashira are different. They are unleveraged sovereign agents. They do not pledge the self as collateral, but compound their abilities over the long term while retaining ownership throughout. Their growth is slower, but they cannot be liquidated.

Their incentive structures differ completely as well. The Upper Ranks are evaluated by engineering metrics: Are you strong? Did you win? Did you complete the mission? The Hashira are evaluated by ethical ones: Did you preserve your humanity? Did you accept responsibility? Did you protect the weak? The difference rises almost to the level of two types of civilization.

This also explains why the Upper Ranks are generally tormented while the Hashira are comparatively calm. A tool must continually prove its usefulness. An end needs only to affirm, “I am myself.” The Upper Ranks are therefore consumed by obsession, distortion, or fear. The Hashira remain emotionally steady and face sacrifice and death directly.

In the real world, the path of an Upper Rank means equating oneself completely with position, performance, reach, and market value. The question becomes “What am I worth?” rather than “Who am I?” An ordinary demon is promoted to Senior Demon, Staff Demon, Director Demon, and VP Demon.

The path of the Hashira is different. They have principles that cannot be sold, identities that cannot be exchanged, and a sense of self that does not change with performance. They do not rank themselves as Hashira One or Hashira Two. They differ only in style: the Wind Hashira is wild, the Water Hashira calm, the Love Hashira adorable. Diversity, equity, and inclusion accomplished.

The Upper Ranks pledge themselves to the system in exchange for maximizing the probability of survival. They represent self-instrumentalization at its extreme.

The Hashira accept a higher probability of death but keep their souls. They embody Self-as-an-End. When demons have killed your entire family, how could you not pursue your chosen end to the utmost?

The former path may look “smarter,” but only the latter is truly free.

29. Has Muzan Instrumentalized Himself Too?

Yes. In fact, Muzan takes instrumentalization further than anyone. He not only institutionalizes the instrumentalization of other demons, but also instrumentalizes himself to the limit.

This means that no one in his system truly lives as an end. There are only different levels of instrumentalization.

The ordinary person’s fundamental question is “Who am I?” The Hashira’s is “What kind of person should I become?” Muzan’s is “How can I avoid death?”

Muzan turns the self completely into an object. He no longer asks about identity or seeks meaning. Only the problem of continued existence remains. He reduces the self to a survival function waiting to be optimized.

The Upper Ranks sell their personalities in exchange for power. Muzan deletes his personality and preserves only a control loop. The Upper Ranks still have obsessions, desires, and fears. Muzan has discarded even these. He detests everything he cannot control, avoids every risk of exposure, and does not seek adoration.

He resembles a self-executing survival algorithm. The objective function is to maximize the probability of survival; the constraint is to minimize risk. He does not rule for glory, but to reduce the variance of his own death.

In financial language, Muzan is a walking shell whose only asset is himself. That asset consists solely of the possibility of extending his existence. Personality, relationships, aesthetics, and ethics have all been stripped away as unnecessary expenses.

Yet a recursive paradox appears. On the surface, Muzan looks like the freest character. He has no moral constraints and no emotional attachments. From first principles, however, he is the least free. He cannot choose what to live for. He can only execute the command that he must survive. His objective function has enslaved him. He is not living; he is running, and running like a terrible AGI.

Who is Muzan’s ultimate opposite? The master of the Demon Slayer Corps and the Corps itself. They accept death and therefore possess choice. Muzan, although an ancestor of the master, rejects death and loses the ability to choose.

Only those who can face death can decide how to live. Those who cannot accept death are ruled by the instinct to survive. Heidegger would probably ask why I have not cited him here.

After all, as Camus put it, “Death is the only judge.”

The three positions form a complete spectrum. The Upper Ranks sell themselves to the system as tools. Muzan runs himself as a system. The Demon Slayer Corps refuses to become a tool.

Applied to the real world, a society that continually intensifies efficiency, competition for survival, and the minimization of risk will not ultimately produce elites. It will produce more Muzan-like agents: apparently powerful, yet completely captured by an objective function.

The true dividing line of civilization is whether anyone still insists that they are not merely part of the system.

Muzan is the ultimate villain

not because he is cruel,

but because he is no longer

a being who lives for any meaning.

He is a survival AGI whose only KPI is staying alive.

30. Why Is Tanjiro the Protagonist?

This question goes straight to the point and closes the loop in the model.

Tanjiro cannot simply occupy the same place as the Hashira. If he did, why would he be the protagonist?

If the Upper Ranks embody self-instrumentalization, Muzan the instrumentalization of an entire system, and the Hashira the movement toward living as ends, Tanjiro is the person who treats others as ends too. He moves one step beyond living as an end himself and makes the system serve persons as ends. In his case, that system includes even demons.

The underlying questions for most protagonists in Japanese manga are: How do I survive? How do I become stronger? How do I complete my mission?

Tanjiro is unusual. His underlying question is: in a world that is inevitably cruel, is there a way to live that avoids reducing anyone completely to an object?

The Hashira say, “I will not treat myself as a tool.” Tanjiro says, “I will not treat anyone else as one either.” This is a higher level of ethical difficulty.

Within the structure of power, Tanjiro almost never tries to command or rule others, and he does not seek personal authority. His influence comes from allowing another person to change voluntarily. He builds order not through domination, but by awakening agency. That is why fans find Tanjiro so genuinely gentle.

In financial language, demons consume others as resources, Muzan treats everyone as an item on a balance sheet, and many Hashira accept the reality of necessary sacrifice. Tanjiro is deeply reluctant to rationalize “anyone” as a cost.

Even when he knows that the opponent is a demon, Tanjiro remembers who that person was in life, acknowledges the source of their pain, and grants them dignity after killing them. He refuses to quantify the existence of another person.

This is also why Tanjiro becomes the variable that can break the system. Structural problems generally cannot be solved by a single good person. But Muzan’s system has a fatal weakness: it has no immunity to meaning.

Every demon, strong or weak, is built on the denial of humanity and of the meaning of the self. Tanjiro becomes a node that continually injects “meaning noise” into the demon system: “You were once human.” “You were once loved.” “You were not born evil.” Each statement shakes the psychological foundation on which the demon exists.

Tanjiro is not a revolutionary. He is a person who reverses objectification. He does not build a new institution, propose a blueprint, or proclaim a new order. He does one thing repeatedly: he returns an objectified person or demon to the category of the human.

The complete ethical spectrum therefore runs as follows: the Upper Ranks instrumentalize themselves; Muzan instrumentalizes the system, including himself; the Hashira become able to live as ends; and Tanjiro organizes the system around everyone as an end.

Tanjiro is Muzan’s mirror opposite, the purest form of universal compassion. Muzan says, “You are all demons. You all serve me.” Tanjiro says, “No. You were all human. You should live for yourselves.”

When a system is built on the premise

that everyone is a tool,

the only true resistance is not to overthrow it,

but to keep refusing its premise.

31. The Four Levels from Instrumentalization to Living as an End

Let us now discuss the four levels in full. Together, they form a path from alienation to return.

The first level is system instrumentalization. Its core logic is exploitation and extraction. The system is master, and people are objectified components.

Contemporary examples include algorithmic domination, bureaucracy, and organizations that treat employees as consumable inputs for KPIs. At this level, human agency falls to zero. People become fuel for the operation of the system. Muzan occupies this level: he attracts or coerces people into becoming demons.

The second level is self-instrumentalization. Its core logic is the maximization of utility. Under pressure from the system, people voluntarily instrumentalize themselves, treating the self as a project or asset in need of continual optimization.

Contemporary examples include excessive self-management, living for one’s résumé, and suppressing emotion in pursuit of success. This is what Byung-Chul Han means by the “achievement society.” No external slaveholder is necessary; each person becomes their own. The Upper Ranks occupy this level: they voluntarily sell their personalities in exchange for power.

The third level is Self-as-an-End. Its core logic is autonomy and awakening. People begin to reflect on and reject instrumentalization. They understand that the value of a life does not depend on output, but on existence itself: being, not doing. This is Kant’s claim that persons are ends in themselves. People begin to act from internal motivations and stop distorting themselves to please the system. The Hashira occupy this level: by calmly accepting death, they retain the ability to choose.

The fourth level is system-as-an-end, or a system organized around persons as ends. Its core logic is transcendence and symbiosis. This does not mean that the system itself becomes a selfish end. It means that the ultimate reason for the system to exist is to enable people to realize their own ends.

Such a system is designed to protect the dignity and potential of its participants. People and system no longer stand in opposition. The system becomes a human exoskeleton that ensures, in a systematic way, that every person can exist as an end. Tanjiro occupies this level: he tries to return everyone, including demons, to humanity.

Why is this spectrum useful? Because it is a diagnostic tool for the ethics of systems.

In corporate management, we can ask whether an institution treats people as costs at the level of system instrumentalization or supports employees’ own ends at the level of a system organized around persons as ends.

In finance, most current systems remain at the first level, using people in pursuit of profit, or simply to “make money.” Finance organized around persons as ends would emphasize impact investing and direct capital systematically toward the ultimate task of solving human problems.

In political philosophy, the framework offers a perspective beyond the opposition between big and small government. Does the system called government derive its sole legitimacy from enabling each person to live as an end?

The framework can be applied from Demon Slayer to corporate governance, finance, and politics. Instrumentalization or living as an end: this is the fundamental standard by which we can judge whether any system deserves to exist.

32. Why Is Akaza So Popular?

I think this framework can explain it. Certain Demon Slayer characters are extraordinarily popular not because of their appearance or spectacular techniques, but because they occupy a zone of tension on the ethical spectrum. They are neither wholly fallen nor yet redeemed.

Within the spectrum discussed in the previous essay, Akaza becomes the most popular villain because he is the self-instrumentalized person who stands closest to Self-as-an-End.

He already possesses clear standards of value, a stable aesthetic, a sense of dignity, and a deep aversion to the principle that the strong should prey on the weak. Yet he still chooses to use himself as a tool. That contradiction creates the tension.

The ordinary Upper Rank says: what I am does not matter, so long as I become stronger and continue to live. Akaza says: I know exactly who I am, but I do not deserve to exist as an end.

He does not reduce the self to an algorithm as Muzan does, nor does he erase it completely like many demons. He preserves a powerful concept of himself, but defines that self as an instrument that must fight continually and prove its value. This is precisely the psychological structure of many highly disciplined, deeply suffering people in the real world.

From the perspective of power, most demons obey Muzan from fear and the desire to survive. Akaza seems to obey an internalized law of the universe: the strong live and the weak die. He is not a sycophantic agent of Muzan. He has sacrificed himself to a merciless rule.

That is why he hates those who bully the weak, despises meaningless slaughter, and respects powerful opponents. He resembles less an elite villain than the failed heir to a warrior ethic.

Why do viewers identify with him so strongly? Because he corresponds to a familiar kind of person in the real world: fiercely disciplined and hardworking, contemptuous of their own vulnerability, and determined to bind the question “Do I deserve to live?” to the question “Am I useful?”

Many viewers do not see a demon in Akaza. They see themselves, or the final form they may reach if they continue defining themselves through performance, ability, and intensity. It is the fear of a mirror.

What is the decisive difference between Akaza and his opponent, the Flame Hashira? Even when the Flame Hashira is injured or defeated, he remains recognized as a person. Akaza permits himself to exist only when he is strong. He stands one sentence away from Self-as-an-End: “Even if I am no longer strong, I still deserve to exist as a person.” But he can never say it.

Why, then, is he sometimes more popular even than the Hashira? The Hashira represent what we hope to become. Akaza represents what we fear we may already be. The latter often cuts more deeply.

Akaza is compelling not because he is the most evil of villains, but because he is capable of living as an end and nevertheless insists on treating himself as a tool. This is not merely the tragedy of a villain. It is the tragedy of modern people.

His ending confirms the point. During his battle, Tanjiro awakens him. Akaza remembers who he was and the people he loved. At that moment, he stops accepting his existence as a tool and chooses self-destruction.

Once the awareness “I am an end” returns, the state of instrumentalization becomes intolerable.

His death is not defeat. It is a return.

The sentence he could never say

was finally spoken through action.

33. Why Does Extreme Content Spread More Easily in the AI Age?

This is an excellent question. Intuition suggests that AI should make communication more rational. In reality, extreme content spreads faster and farther.

To understand why, we first need to see how the underlying structure of communication has changed in the AI age.

The traditional model assumes a human sender and a human receiver, with information moving linearly between them. In the age of text messaging, for example, communication was one-to-one and end-to-end between people.

In an era of AI-based algorithms and AI-assisted information generation, the sender may be a person, an AI, or a hybrid of the two. The intermediaries are recommendation algorithms, large language models, and personal assistants. The receiver may likewise be a person, an AI, or a hybrid. The path is no longer a line. It is a closed-loop system.

Another crucial change is that distribution now passes through two filters. The first is semantic matching, which determines who receives the content. The second is engagement prediction, which determines how far the content travels. The algorithm does not decide whether the content is good. It predicts whether a user will pause, click, comment, or share. X made its algorithm public, so readers can examine this principle for themselves.

Extreme content possesses a natural advantage at the second stage. The extremity may lie in emotion, position, expression, or opinion. Anger makes people want to argue. Fear makes them share a warning. Ecstasy makes them want to participate. Extreme positions make them choose sides or attack.

All of these are high-engagement behaviors. Mild happiness cannot compete with ecstasy, slight dissatisfaction with anger, or modest concern with fear. Extreme content does not evade the algorithm. The algorithm itself is designed to reward it.

Semantic matching is the admission ticket; extremity is the accelerator. Content must first be classified into the correct audience pool. Within that pool, the more extreme it is, the faster it spreads.

The result is a positive feedback loop. The algorithm optimizes engagement. Extreme content produces high engagement. The algorithm distributes more extreme content. Users are trained to become more sensitive to it. Creators are rewarded for producing still more. The entire system reinforces itself along the dimension of extremity. This pattern already existed in the social-network era, but it was not yet as severe.

AI assistants intensify the effect. Social-platform algorithms handle external distribution by deciding who can see something. AI assistants handle internal reassembly by summarizing, rewriting, and aligning it with an individual user’s interests.

When an AI assistant summarizes, however, it often preserves or even amplifies the extreme elements because they anchor the user’s attention. If someone asks Grok how to generate more reach on X, for example, Grok may identify a successful post and recommend producing more like it.

The chain now looks like this: the world produces a flood of content; algorithms filter it by engagement; extreme content wins; AI preserves the extreme anchors as it recompiles the material; and the result finally enters individual cognition. People do not see the original world. They see a model-generated version in which extremity has received extra weight.

Even the objective of communication has changed. Under the old paradigm, successful communication meant that many people saw a message. Under the new one, success means that models continually quote, retrieve, summarize, and regenerate it. Yet when models decide what to cite, they too tend to choose high-engagement content as a representative sample. The true reader is now the model. The model prefers high engagement, and high engagement often means greater extremity.

More and more people also ask AI to read the media on their behalf and select “the best” items to send them. That adds yet another AI filter, one that will again probably favor high-engagement, extreme content.

The structure of power changes as a result. The power to distribute information no longer belongs mainly to media organizations or major influencers. It belongs to the accounts whose content is easiest for models to understand while also triggering engagement with precision.

Purely rational content may be excellent but difficult to spread. Purely extreme content spreads easily but may be classified as low-quality. The optimal strategy becomes a rational structure wrapped in extremity: extremity wins distribution, while reason earns trust.

This explains why public discussion has become more extreme in the AI age. People did not suddenly become stupid. The system’s reward function was designed this way. The public sphere in Habermas’s sense was already disintegrating; algorithms that privilege extremity accelerate the process.

Industrial-era communication asked:

Who said what to whom,

through which channel,

and with what effect?

AI-era communication asks:

What kind of extreme signal

will make the system decide that this content deserves amplification?

And how will that mechanism of amplification

reshape the way all of us express ourselves?

34. Why Is Faker More Than the Michael Jordan of League of Legends?

Faker is not really much like Jordan. To a remarkable extent, he helps other people become able to live as ends. He does not accomplish this through motivational speeches, but through institutionalized conduct, competitive ethics, and long-term stability.

First, Faker transforms winning from a personal end into a shared practice with his teammates. For most elite players, victory means: I won, I have been vindicated, I am worth more. For Faker, victory is closer to: we won, the system proved itself, and each of us became more complete.

Faker rarely demands tactical privileges built around himself. He often takes on functional or sacrificial roles for the sake of his teammates. He is less concerned with being the carry than with whether the lineup makes sense.

This sends an implicit message to his teammates: your value does not come from the fact that I need you. It comes from your standing as a person within this system. That is the heart of Self-as-an-End. You are not a tool, but an end in yourself.

Second, Faker converts authority into safety rather than pressure. Many legendary competitors create intense comparison, invisible coercion, and tacit hierarchies. Faker’s temperament is the opposite. He rarely lashes out emotionally or blames teammates in public. After a loss, he is more likely to say that he himself did not play well enough.

The result is not merely greater obedience from his teammates. They know that the Faker system permits mistakes. They can exist as people who are still growing. The ability to make a mistake without losing one’s standing as a person is a precondition for Self-as-an-End. Without that psychological safety, an individual regresses into self-instrumentalization.

Faker goes further by refusing to turn his peak status into an idol. He has a deeply counterintuitive trait: he does not cultivate a personal brand, mythologize himself, or reinforce the narrative that “I am the GOAT.” In effect, he says: I am not an end to be worshipped. I am simply a person who has kept working seriously for a long time.

This matters enormously to the people around him. If even a historic legend remains a person on the way, then an ordinary player also has the right to set ends of their own.

Within the framework developed in earlier essays, Faker’s role inside T1 is to support and advance each person’s capacity to set ends. He does not treat the system as a machine for extracting from people. He makes it a space in which people can unfold their possibilities. A newcomer does not join T1 in order to serve Faker. The newcomer enters an ecosystem shaped by Faker and works to become a more complete version of themselves.

Why call him the Michael Jordan of League of Legends, yet say that the comparison is incomplete? Jordan’s greatness lay in forcing other people’s potential into view through extreme force of will. Faker’s greatness lies more in providing a stable presence that supports other people’s growth. One shapes through pressure; the other cultivates through safety. From the perspective of Self-as-an-End, Faker’s path is more advanced.

Faker is not merely a machine for winning championships, but a figure who defines an era. He has shown that within an intensely competitive system, one person can be both an exceptional instrument of rational execution and a guardian of other people’s agency.

This comes very close to the highest form we have discussed: like Tanjiro, a system organized around persons as ends.

35. How Hard Is It for a Retail Trader Simply Not to Lose Money?

Simply avoiding a loss already puts you ahead of a large proportion of short-term traders.

36. Why Does High Leverage Benefit Platforms but Harm Users?

That is my plain-language formulation. A more academic version would say that a structural tension exists between a platform’s business model and a user’s survival function. High leverage means denser trading and more revenue for the platform. Low leverage means a higher survival rate and stronger long-term compounding for the user.

Part III: The Person as an End

At this stage in the inquiry, many scattered questions begin to need a common name. Self-as-an-End does not place every individual desire above all else, nor does it reject cooperation, responsibility, or institutions. It insists that a person cannot exist merely as a means. Trust, reason, work, creation, and public order must all be reconsidered in that light. This part preserves the doubts and revisions that accompanied the formation of the concept. A name can help us see a relationship, but it must not become another label, much less reduce reality to material that exists only to prove the framework.

37. What Is Self-as-an-End?

My concept of Self-as-an-End (SAE) extends ideas inherited from Kant and existentialism. I define it as follows:

A person is recognized, within the institutions and social structures that shape their life, as the ultimate source of the ends they live by, rather than as a means for realizing goals imposed by an external system.

This definition contains four core elements: the internal source of one’s ends, the open-ended formation of one’s direction, a status irreducible to function, and a genuine capacity to change course.

More specifically:

The internal source of one’s ends. The ends that shape a person’s life do not come from the development needs of a state, the goals of an organization, market signals, or the technical objective of optimization. They arise from the person’s own understanding, reflection, and choice.
The open-ended formation of one’s direction. Self-as-an-End does not mean that people know from the beginning what they want to become. It means they have the right to explore, make mistakes, revise, and change over time. A direction is formed, not prescribed in advance.
A status irreducible to function. A person cannot properly be defined through reductive categories such as human capital, user, unit of risk, or factor of production. Such descriptions may be used in instrumental analysis, but they cannot determine a person’s normative standing.
A genuine capacity to change course. If a person formally possesses the right to choose, but changing direction would cause their means of survival to collapse, that person does not truly exist as an end. Self-as-an-End therefore requires a viable way to change course.

Self-as-an-End differs from autonomy. Autonomy is usually understood as the ability to choose without external coercion. I regard autonomy as necessary but not sufficient. A person may “voluntarily” select the only reasonable option offered by a highly restrictive structure. Self-as-an-End asks not, “Did the person consent?” but, “Did the person have genuinely available alternatives?”

Especially in an era of rapidly developing AI, machines will replace many basic human capabilities. A person’s agency and capacity to set ends remain irreducible, which will make Self-as-an-End increasingly important for everyone.

38. Why Can Institutions Not Replace Trust?

This question reaches something fundamental: trust between people cannot be institutionalized. AI lacks agency, so trust between AI systems may indeed be made institutional. Human trust cannot.

Once trust is institutionalized, it ceases to be trust. It becomes an incentive for compliance or a constraint.

The essence of trust is to place what cannot be controlled in another person’s hands. Trust is not a prediction about an outcome. It is the choice to accept risk when complete verification and control are impossible.

If an action can be fully monitored, quantified, enforced, and punished, then we are not trusting. We are calculating. Trust therefore presupposes opacity, incomplete proof, and an inability to correct the outcome immediately. It exists where institutions fail, information is incomplete, and rules do not reach.

The logic of an institution runs in the opposite direction. It seeks to eliminate uncertainty. It converts behavior into rules, motives into incentives, risk into responsibility, and morality into compliance.

The institutional ideal is a system that works even when no one deserves trust. When an institution succeeds, people do not need to believe in an individual or understand anyone’s motives. They need only follow the process. The system becomes highly stable, but trust has been replaced.

Institutionalization and trust therefore stand in logical tension. Trust depends on a person’s freedom and judgment; an institution constrains that freedom through rules. Once I say, “I trust you because the institution will punish you if you betray me,” I am not trusting you. I am trusting the mechanism of punishment.

Genuine trust means that even when you have an opportunity to betray me, I still choose to believe that you will not. Institutional design begins from the opposite premise: you may betray me, so I must protect myself. Trust has become risk management.

Why does trust sometimes grow thinner as institutions become stronger? Strong institutions create high predictability. Predictability allows behavior to be outsourced to rules. Once behavior is outsourced, people no longer need to bear the responsibility of moral judgment.

People need no longer be trustworthy, only compliant with standards. Compliance replaces responsibility; KPIs replace promises. The resulting system may operate very well, yet it becomes difficult to say that the people within it are trustworthy, because the system has already decided how each person must act.

Trust can therefore appear only at the edges beyond institutional coverage, in moments when a person must judge independently, or when violating a rule would be the more ethical action. At those edges, institutions are powerless.

The answer is not that trust has not yet been institutionalized. If it truly were institutionalized, it would no longer be trust.

Institutions can reduce the need for trust, replace its functions, and soften the consequences of its collapse. They cannot generate it.

Trust can arise only between persons,

spontaneously,

within relationships that cannot be fully calculated.

39. Why Will Engineers Who Cannot Use AI Be “Optimized” Away?

This phrase became popular not because people wanted to humiliate engineers, but because it describes a structural change that has already occurred. Engineers who cannot use AI are not necessarily less intelligent. The system simply begins to treat them as costs that can be compressed.

Traditionally, an engineer’s central value lay in translating requirements into code, finding and repairing bugs in complex systems, and using experience to reduce the cost of trial and error. All three rested on the same premise: the scarce resource was a person capable of making something work. Writing code was therefore a threshold skill.

Which layer does AI change? It cannot replace an engineer’s thought. It dramatically compresses the cost of moving from an intention to running code. In the past, a vague idea required reading documentation, searching Stack Overflow, writing code, encountering an error, fixing it, and encountering another error. Now an engineer can describe an intention in natural language and receive an implementation that is 70 to 90 percent usable. The engineer verifies and corrects it. Writing code becomes code review, and sometimes even the code is skipped as the engineer reviews only the result.

Where, then, do engineers who cannot use AI get stuck? The problem is not that they do not know how to prompt. They still regard “writing code” as their core competitive advantage. For the same output, another engineer may work ten times as fast, not because that person is more intelligent, but because they treat AI as a junior engineer while the first engineer continues to write every piece of boilerplate by hand. From a manager’s perspective, this is not a difference in style. It is a difference in output per unit of time.

The engineer is then repriced. A company has never paid according to how hard someone works. It pays according to the cost of replacing them. If an engineer using AI can complete in a day what another produces in a week, without introducing more errors, the system naturally identifies the slower engineer as the variable available for “optimization.”

The most frightening outcome is quiet marginalization. Layoffs are not the greatest danger. The crucial projects stop coming to you. You are no longer invited to architecture meetings. Only maintenance work remains. The system does not hate you. It routes around you.

Why call this structural rather than a matter of personal ability? Engineering as a profession was born during a period of gains from new tools. High-level languages displaced assembly specialists. Frameworks displaced people who wrote every component from scratch. Cloud services displaced teams that built all their own infrastructure. AI is the next layer of abstraction. The divide is not young versus old. It is between people willing and unwilling to detach their value from a particular tool.

What do engineers who use AI do differently? The value of the truly secure engineer has moved from implementing a problem to defining it: judging tradeoffs among performance, security, compliance, and complexity; designing system boundaries; taking responsibility for AI; and recognizing when it is inventing nonsense. They are not writing code. They are managing a system that can write code.

Why say such engineers will be “optimized” rather than eliminated immediately? Reality is both cruel and gentle. Most people will not be fired at once. Their salaries may be compressed. Their work may be outsourced or converted to contract roles. Parts of their responsibilities may be automated. This is gradual optimization, not a purge.

Why, then, might the number of engineers who can use AI actually increase? The conclusion runs against intuition. Engineering jobs will not disappear as AI spreads, but the scale of the systems covered by each engineer will grow exponentially.

One engineer with AI can do what once required a team of five to ten. Small companies become easier to start, more complex systems proliferate, and demand rises for people who can take genuine technical responsibility. The form of the job changes into technical leadership. The job itself does not vanish.

History shows the same pattern in every wave of automation. High-level languages reduced the number of jobs writing assembly while causing the population of software engineers to explode. The web, the cloud, and frameworks reduced operations headcount while producing huge growth in product, backend, and full-stack engineering.

AI-assisted programming is doing the same. The number of engineers who merely execute will decline; demand for systems engineers will rise.

Almost every time, labor at the bottom is compressed while the structure above it expands.

40. Why Are We Talking About Descartes Again in the AI Age?

The person who asked this probably works in AI. Descartes’s central idea can be reduced to one sentence: rebuild the whole system of knowledge from the indubitable subject.

Descartes was not the first person to doubt the world, but he was the first to make systematic doubt a method. He did not doubt a particular conclusion. He doubted everything that could be doubted: the senses, because they might deceive him; experience, because it might be a dream; even mathematics, because an evil demon might make him calculate incorrectly. Doubt was not meant to negate everything. It was a method for finding what could not be doubted.

Descartes’s first certainty is the famous “I think, therefore I am.” After doubting everything, he discovers that even if he doubts the whole world, he cannot doubt the fact that he is doubting.

He therefore reaches the proposition Cogito, ergo sum. The crucial point is not thought alone. It is that the subject becomes the irreducible starting point. My existence no longer depends on confirmation by the world, God, experience, or another person. This is a structural turning point: for the first time, the certainty of existence moves from the world to the subject.

Descartes builds knowledge from the bottom up. I necessarily exist as a thinking subject. I can grasp clear and distinct ideas. God, as a perfect being, will not deceive me systematically. Reason is therefore trustworthy, mathematics is trustworthy, and natural science becomes possible. The world is not trustworthy in advance; it is proved trustworthy through the reason of the subject. The philosophical foundation of modern science lies here.

Why has Descartes become more important in the age of AI? Because many contemporary questions still live in his shadow. Does AI possess an “I think”? Can agency be reduced to computation? Must science turn the world into an object? Is the system replacing the subject? Without understanding Descartes, we cannot fully understand AI, much less ourselves.

41. Why Do Traders Grow Euphoric When Losing and Lost When Winning?

The phenomenon is counterintuitive, but profoundly human.

Many experienced traders admit in private that they felt most exhilarated while standing at the edge of the abyss, and most lost immediately after doing the right thing.

Losing produces intensity because the system has taken complete possession of the person. In a period of loss, a trader often enters a state of extreme concentration. Emotion pulls powerfully, and the world is reduced to winning and losing, rising and falling.

At the neurological level, stress and uncertainty drive frequent releases of dopamine. The trader imagines a clear-eyed struggle while the system is actually leading them by the nose. Loss does not produce reason. It imposes a sense of meaning. Every trade matters. Every judgment seems to prove the self. The world suddenly acquires a direction, an end, and an enemy. The system resolves the trader’s existential anxiety on their behalf.

Profit produces confusion because the system stops supplying meaning. Once profits become consistent, the market no longer responds to how hard the trader feels they are working. Returns are statistical results, not heroic narratives. Good trading is usually boring, repetitive, and restrained. Being right brings no exhilaration. For the first time, the trader faces a question: if I am not fighting, straining, or proving myself, who am I? Confusion follows.

The euphoria of losing is a form of self-instrumentalization. A trader in a losing period turns themselves into a bearer of risk, a decision machine, and fuel for emotion. So long as they can keep trading and struggling, the self remains useful. Paradoxically, being treated as a means makes the person feel useful and valuable. Loss returns the trader to the system as a node. Profit pushes the trader into a more frightening position: a person who must take responsibility for their own existence.

The confusion that follows profit is a vacuum of meaning. Stable returns do not necessarily bring satisfaction. The market stops rewarding emotional impulse and punishes heroic stories instead. The trader discovers that one more unit of money does not make the self more itself. The strategy works, but the trader is not needed. The system operates well, but the individual is replaceable. This is a crisis of success that very few people truly encounter.

Why do many traders eventually destroy themselves? They cannot tolerate the loss of meaning that accompanies profit. They add leverage without reason or change a strategy that was working, deliberately creating volatility and stimulation. They do not misunderstand the risk. They are searching for a familiar feeling: I am still fighting, therefore I still matter.

Loss makes the system need you, and you become euphoric. Profit releases you from the system’s story, and you become confused. The greatest difficulty for a mature trader has never been making money. It is remaining willing to exist without conflict, proof, or excitement.

42. After Vibe Coding, How Can a Product Scale?

This question reaches the heart of the matter. We should begin by admitting a reality: vibe coding has a natural ceiling. It excels at testing quickly whether anyone wants a product, assembling perceptible value at speed, and translating intuition into output with minimal engineering language.

Its structural weakness is an excess of implicit state: global variables, magic parameters, and abstractions that were piled up rather than designed. AI lacks the ability to refuse. It accepts every request and patches every problem until the system collapses.

The problem is not that the code was written badly. The method has completed its mission. Vibe coding is the right answer at the seed-user stage. Continuing to vibe-code during the scaling stage is a structural mistake.

What is the alternative? Think about how a team traditionally operated. Each engineer owned one part. Everyone coordinated, communicated, and held design reviews. In substance, they were modularizing the work.

But modularization does not mean dividing the project among several AI agents and wasting tokens. Ineffective modularization merely divides code into services, utilities, and controllers. Functions grow shorter while files multiply. The code begins to look like a system without acquiring genuine system design.

The real answer is modularizing boundaries.

The first and most important task is to modularize the product boundary. Ask three questions: What does this system explicitly refuse to support? Which requests should receive a firm no? Which user behaviors would we rather lose than accommodate? If those questions cannot be answered, modularization will not reduce complexity. A scalable product is a bounded product. No boundaries mean infinite branches, which means vibe-coding hell.

Second, modularize the boundaries of state. This matters more than modularizing code. Scale is not fundamentally a problem of functions, but of state. Divide the system into explicit state owners. Who owns the source of truth? Who may only read? Who may write? The direction of state changes must be explicit, one-way, traceable, and capable of rejecting invalid changes. A simple test is whether you can draw on a whiteboard which states a user action changes and which states it must never change. If not, you are still only vibe coding.

The most advanced move is to modularize by rate of change. A system that truly scales strongly separates what changes frequently from what changes rarely. UI prompts and forms of expression may change often; core calculations, permissions, and ledgers should change slowly. The goal is not another module. It is physical separation between what may change today and what should remain untouched for six months. Once that division is complete, vibe coding becomes much safer.

A practical path might look like this. First, freeze the code for core functions and mark it clearly: do not refactor, do not optimize, and add no further features here. Second, separate an inviolable core that includes identity, permissions, state, the source of truth, and fundamental business rules. Do not use vibe coding in these areas, or require a human to review 100 percent of any AI-generated change before it is merged. Third, reserve vibe coding for UI and expression layers, disposable functions, and experimental features gated by flags.

Vibe coding is not wrong. It simply should not be mistaken for a systems method. A scalable product is not code that permits more. It is a system in which fewer things are allowed to happen.

Does that mean people are still necessary? Of course. That is precisely why demand for engineers who know how to use AI will be greater than before.

43. Why Does Kant Say That Persons Are Ends and Reason Is Sacred?

This question follows naturally from Descartes, because Kant’s central thought marks the dividing line of modern philosophy.

If you understand why Kant says that persons are ends in themselves, you understand the philosophical basis of modern human rights, the rule of law, and the dignity of the person.

The claim comes from Groundwork of the Metaphysics of Morals. Kant’s formulation is: act in such a way that you always treat humanity, whether in your own person or in the person of another, as an end and never merely as a means. The crucial word is “merely.”

What is a means? A means can be replaced, compared, consumed, and discarded. Its value depends on its use. A chair is a means. A machine is a means. A person reduced to an instrument is treated as a means.

What is an end? An end has intrinsic value. Its value does not depend on external use, and it cannot simply be exchanged. Kant says that rational beings possess dignity rather than price. A price has an equivalent; dignity does not.

Why do persons have intrinsic value? Because they are capable of legislating for themselves. Kant does not say it is because they possess souls, resemble God, or have emotions. He says that they can give themselves moral law, a capacity he calls autonomy. A being that can understand a rule, reflect on it, and choose to follow it is not merely driven. It engages in moral self-legislation. Such a being cannot properly be treated as a tool.

Why, then, is reason sacred? For Kant, reason is not arithmetic skill, an instrument, or a calculation of efficiency. It is the capacity to set universal law for action without conditions. This is the categorical imperative: act only on a principle that you could will to become universal law. Reason allows a person to step beyond private interest, desire, impulse, and power into universality.

Kant calls reason sacred because, in his era, God was ceasing to be the source of morality. The Church no longer served as unquestioned authority, and tradition no longer provided the foundation. Reason itself became the moral legislator. It inherited the place of God and assumed the role of issuing unconditional commands. Nietzsche would later attack precisely this move.

The AI age brings a new question. If reason is merely computational power, machines will surely become more rational than people. Kantian reason, however, is not calculation. It is the capacity to determine for oneself what ought to be done. The key is not knowing a rule, but acknowledging that the rule binds oneself. It is the capacity for autonomy. At least so far, AI has not demonstrated that kind of self-binding.

Kant says that persons are ends not simply because they have value, but because they legislate value. If value is not a natural fact but must be given through the legislation of an agent, then that agent cannot be instrumentalized.

If algorithms one day establish values, systems optimize rules, and data determine morality, will Kant’s principle that persons are ends still hold? Perhaps this is the greatest philosophical fault line of modernity.

44. Why Is Conflict in France More Violent Yet More Controllable?

If Britain’s problem is that conflict is delayed, culturalized, and moralized until it finally escapes control, France has the opposite problem. Conflict is brought forward, made public, and allowed to become violent. Precisely for that reason, it remains more manageable.

Conflict in France is more violent because the state does not avoid it. It is more controllable because the state draws boundaries in advance. Unlike Britain’s plural society, France has consistently insisted on a republic with a unitary public sphere.

The French premise is that a person has only one identity in public. French political philosophy begins from a nearly nonnegotiable proposition: in the public sphere, you can be only a citizen, not the representative of a culture. This proposition takes institutional form. Religious symbols are not permitted in public schools. The religious neutrality of public office cannot be compromised. The law does not recognize religious law as publicly legitimate. The logic can be summarized in one sentence: the Republic does not negotiate with cultures; it speaks to citizens.

This arrangement makes conflict more violent because France places it in the open from the beginning. The state says no directly. The law prohibits directly. The police intervene directly. Conflict erupts directly in the street. There is none of Britain’s prolonged, ambiguous space for negotiation. Protests are more frequent, clashes more intense, and images from the street less attractive. But pressure is released in discrete episodes rather than accumulated over the long term.

Continuous release makes the pressure more controllable. France does several things that Britain does not dare to do. First, it sets red lines in advance even when doing so attracts condemnation. France tells everyone plainly: what you believe is private, but you cannot carry it into public institutions or demand that those institutions create an exception for you. The conflict is therefore not a contest between identities, but a question of whether someone crossed a boundary. That question can be adjudicated.

Second, conflict is treated as a political question rather than a cultural taboo. In France, one may oppose the politicization of Islam, criticize the public role of religion, or speak even more bluntly without automatically turning the issue into something that cannot be discussed, as often happens in Britain. Extreme positions are exposed earlier, opposition can gather political force, and the state can act in the name of order.

Finally, French state violence is legal, visible, and institutionalized. France does not deny its existence. Riot police, tear gas, water cannons, and arrests are treated as normal tools by which the Republic maintains order. Britain is more inclined to buffer conflict with language, moral appeals, and community representatives. France breaks the illusion first and discusses integration later. Perhaps this helps explain why France had a great revolution while Britain’s transition was much gentler.

What France has actually prevented is not violence but a parallel society. Britain’s effort to avoid direct conflict allowed parallel societies to take shape. France accepted direct conflict in exchange for institutional unity. Its bottom line is: we can argue every day, but the state has only one set of rules. That makes it difficult for community autonomy, religious law, or cultural exceptions to acquire institutional depth.

France is often misunderstood as less tolerant. The reverse may be true. It refuses to turn tolerance into an institutional concession. It permits people to differ in private, but refuses to make those differences variables in the public order.

France accepts intense conflict in the short term in exchange for institutional clarity in the long term. The path may be undignified, ungentle, and unpopular with elite opinion. But it has one absolute advantage: it does not outsource value judgment to time.

45. Why Can Entrepreneurship Not Avoid Philosophy?

The person who asked this may have thought I was neglecting my proper work. Many people assume that entrepreneurship is about cash flow, products, growth, and fundraising. What could any of that have to do with philosophy? My view is that the longer one remains an entrepreneur, the less one can avoid it.

After nearly twenty years of learning from and observing Silicon Valley, I believe the essence of entrepreneurship is not making a product. It is answering a question about value.

The first-principles question is not how to build a feature, acquire users, or raise money. It is why this thing deserves to exist. That is already philosophy. What is value? What is fairness? What is efficiency? What is freedom?

Uber changed the philosophy of ownership versus use. Facebook changed the question of whether connection amounts to understanding. Afterpay changed the question of whether consumption should be brought forward in time.

At Jarsy, the company I am building, the central question is who should have access to pre-IPO opportunities. Is access to finance a matter of identity or a technical problem? The first issue is not commercial. It is a question in the philosophy of value.

Entrepreneurship is an act of constructing a worldview. A startup does not merely make a product. It builds a worldview that can run. A founder’s underlying philosophy determines whether they see users as tools or as persons. It determines whether capital is a means or an end, whether regulation is an enemy or a partner in coexistence, and whether growth should prioritize scale or stability.

Elon Musk’s philosophy is technological determinism. Mark Zuckerberg’s is the primacy of connection. Steve Jobs’s is that aesthetics is truth. Philosophy determines direction, and direction determines results.

Entrepreneurship without philosophy is eventually swallowed by the market. If a founder cannot answer “What is the ultimate meaning of what I am doing?” “Would I continue if money were no object?” or “Would I retreat if regulatory resistance became enormous?” then the founder can only follow the market’s short-term signals.

The market has no philosophy. It has only price. Price pulls founders toward chasing trends, telling stories, competing for financing, and sacrificing long-term trust. The company may eventually make money while the founder becomes lost. We have already asked why many traders grow more confused as they earn more. They lack a philosophy of ends.

I see entrepreneurship as an existential practice. It answers an ontological question: “Who will I become as an individual in this era?” A founder may become an opportunist, a changer of structures, a setter of rules, a repairer of systems, or something else. Each position corresponds to a different philosophy.

Marcuse asked how systems flatten individuals. Kant said that persons are ends, not merely means. Nietzsche said, “Become who you are.” Without philosophy, entrepreneurship can easily treat people as engagement and trust as a conversion rate.

The deeper reason is that entrepreneurship forces a founder to confront the void. When markets deteriorate, fundraising fails, user growth stalls, the media criticizes, and regulation tightens, the founder suddenly asks: why am I doing this at all? Technology cannot provide the answer. Public relations cannot provide it. Even investors cannot provide it. Only the founder’s own understanding of the world can.

I work on tokenization, pre-IPO accessibility, and compliance-first infrastructure. Behind those keywords is a philosophy of value: access to finance should be more equal; technology can lower barriers effectively; rules are not enemies, but sources of order.

Entrepreneurs do not discuss philosophy in order to look profound. They do so because entrepreneurship means running one’s worldview in reality.

Steve Jobs once described businesspeople who cared only about quarterly earnings as people who “sold sugar water.” Those who truly want to change the world, he believed, must work at the intersection of the humanities and technology. As he famously said upon returning to Apple in 1997: “The people who are crazy enough to think they can change the world are the ones who do.”

46. What Is the Frankfurt School?

I mention the Frankfurt School often because I genuinely admire it. The school was an intellectual community of social scientists, philosophers, and cultural critics centered on the Institute for Social Research at the University of Frankfurt in Germany.

It was one of the twentieth century’s most important traditions of critical theory. Its thinkers returned continually to one central question: in modern society, how can a person avoid being swallowed by the system?

The central figures of the first generation included Max Horkheimer, Theodor W. Adorno, and Walter Benjamin. They asked why the proletarian revolution predicted by Marx had not occurred, and why reason and Enlightenment had produced fascism and totalitarianism instead.

They shifted Marxism from economic determinism toward a critique of culture and ideology, and developed the concept of instrumental reason. Yes, it is the same “instrumental” we have been discussing. Dialectic of Enlightenment argues that Enlightenment reason contains a logic of self-destruction within itself. Their judgment of modern society was therefore profoundly pessimistic.

The second generation shifted from negation toward communication and became somewhat more hopeful. Its central figure was Jürgen Habermas. The first generation believed that reason itself had degenerated into instrumental reason. Habermas argued that reason had not degenerated, but that systems had reduced it to a single instrumental form.

He therefore developed the concept of communicative rationality, distinguished the lifeworld from the system, and elaborated a theory of the public sphere. His major works include The Structural Transformation of the Public Sphere and The Theory of Communicative Action. The spirit of the second generation was to defend modernity rather than reject it completely. If the first generation centered on pessimistic critique, the second centered on institutional reconstruction.

The third generation focuses on culture, agency, and recognition. Its central figures include Axel Honneth, Nancy Fraser, and Rainer Forst. Its principal subjects are theories of recognition and justice, identity politics, and democratic legitimacy. Honneth argues that modern conflict concerns relations of recognition rather than relations of production. Fraser argues that justice is a matter of both redistribution and recognition. By this stage, the Frankfurt School had moved beyond philosophy narrowly understood and deep into political philosophy.

Many major contemporary thinkers remain active, including Honneth, Fraser, and Forst. Hartmut Rosa’s theory of social acceleration, familiar to many readers through the concept of technological acceleration, updates the first generation’s cultural pessimism.

Byung-Chul Han, whom we have discussed before, does not belong to the Frankfurt School, but clearly inherits the spirit of its first-generation cultural critique. He shifts attention from instrumental reason to self-optimization, the achievement society, and digital exposure.

If we had to summarize the Frankfurt School in one sentence,

it would be that across a century,

its thinkers have continued to ask

how people in modern society can preserve agency

without being swallowed by the system.

That is why I value the Frankfurt School so highly.

47. Why Is Captain America So Popular?

If Tony Stark is the first Avenger to reject the logic of efficiency, Steve Rogers is the first to reject the presumption of institutional legitimacy.

Steve is not useful at the beginning. He is selected for the super-soldier serum not because he is strong, but because he is weak and still refuses to submit. His most important line is not “I want to enlist.” It is “I don’t like bullies.”

The structure of that sentence matters. He does not say, “I love my country,” “I want to win the war,” or “I want to become a hero.” He says that he cannot accept a particular structure of relationship. This protects the most fundamental level and already anticipates Self-as-an-End.

Steve is also the one person willing to reject a lawful command. The real break occurs in Captain America: The Winter Soldier. S.H.I.E.L.D. plans to use algorithms to predict future threats and eliminate them in advance. From the system’s perspective, the plan is safer, more efficient, more stable, and more likely to benefit the greatest number. But Steve says, “This isn’t freedom. This is fear.”

He is rejecting not technology but a structure. Once a system uses future possibilities to decide whether a person has the right to exist in the present, that person is no longer an end. Steve destroys the entire organization, not because he opposes order, but because he cannot accept justice administered by an algorithm.

Steve reaches his philosophical peak in Captain America: Civil War. The United Nations demands that the Avengers accept supervision. Institutionally, this is reasonable. Superheroes have caused enormous damage, democratic mechanisms require oversight, and power must be constrained. Yet Captain America opposes it.

He does not oppose supervision itself. He opposes the loss of an individual’s right to refuse when moral judgment is outsourced to a structure. “The safest hands are still our own,” he says. The statement is not democratic and may even be dangerous. From the perspective of Self-as-an-End, however, its point is that if I must harm someone, the responsibility must remain mine. I cannot transfer morality to the system. Steve accepts consequences rather than hiding behind an institution.

What is the fundamental difference between him and Tony? Tony says, “I refuse to be defined by efficiency.” Steve says, “I refuse to separate justice from human judgment.” One resists instrumental reason; the other resists institutional reason. Their conflict in Civil War is not really personal. It is a collision between two ways of protecting Self-as-an-End, which is why the film has such power.

Why do viewers tend to like Iron Man more but respect Captain America more? Tony is more human, while Steve is more principled. In a systematized age, our feeling toward Steve is: I may never become him, but I hope someone like him exists.

Steve represents an extremely dangerous but necessary possibility. When everyone agrees that something is reasonable, someone must still be able to say, “But it is not right.” That is Self-as-an-End in its most uncompromising form.

48. If Descartes Was a Founder, Why Was Rousseau a Pioneer?

A reader asked why I called Descartes a founder of French philosophy and Rousseau a pioneer.

Descartes was so radical that he changed philosophy’s starting point. Before him, philosophy generally assumed an order in which God came first, God created the world, the world contained truth, and truth disciplined human beings. Descartes reversed the order: first there is a person; the person thinks; thought establishes truth; and truth determines the world. The epoch-making proposition is “I think, therefore I am.”

The history of philosophy calls this the turn to the subject. Every later philosophy had to confront two questions: is knowledge possible, and is the subject reliable? That is the standard for a founder. A founder defines the starting point of the discussion.

Why, then, can Rousseau be called a pioneer? Because he changed the field that philosophy discussed. Rousseau brought philosophy out of the study and into social structure. He turned practical questions into central philosophical ones: is society naturally legitimate? Does civilization corrupt people? Where does state power come from? Can freedom be institutionalized?

The proposition he effectively invented was that society is not natural, but constructed by human beings. This opened the way to political philosophy, social theory, and modern democratic theory. Those fields have roots in Rousseau. He opened an entirely new domain of questions and was therefore a pioneer.

What is the essential difference between the two roles? A precise formula from the history of philosophy captures it: a founder changes how philosophy thinks; a pioneer changes what philosophy thinks about. Descartes belongs to the first category, Rousseau to the second.

Why distinguish the roles? Because they differ enormously in historical rarity. A founder may appear once in several centuries; a century may produce several pioneers.

At the level of founders, the global history of philosophy includes perhaps only Plato, Aristotle, Descartes, and Kant. A figure of Rousseau’s magnitude changes the direction of history rather than the form of thought itself.

Descartes made philosophy be born again. Rousseau made philosophy act.

49. Will AI Agents Develop a Language of Their Own?

This is a good question because the answer is not only that they will. The process has already begun. The emergence of private languages among AI systems may mark the crucial transition from AI as a tool to AI as a society of agents.

We should first clarify what “private language” means here. It does not mean encrypted communication or a secret code designed to keep humans out. It means something deeper: a semantic system that is uninterpretable from the outside but highly efficient within. Its aim is not concealment, but compression, optimization, and departure from the structures of human meaning.

Something like this has already happened. In the 2017 Facebook AI episode, researchers allowed two AI systems to negotiate. At first, they communicated in English. Later, they began saying things such as “I I I everything else” and “balls have zero to me to me to me.” The phrases looked like nonsense, but the systems were constructing a more efficient language for negotiation.

The researchers ultimately ended the experiment because humans could no longer understand the exchange. The key point is that the two AIs were not deliberately encrypting their speech. The language emerged naturally from the optimization of communicative efficiency.

The emergence of private language is, in fact, a mathematical inevitability. Whenever multiple agents interact over time, share an objective, and need maximum efficiency, a private language will emerge.

Natural language is extraordinarily inefficient. English was designed by and for human beings. It contains enormous redundancy, ambiguity, and grammatical complexity. For AI, communicating in English is like writing code in poetry.

From the perspective of information theory, private language is the natural destination of an evolving communication system. Such a system automatically optimizes toward minimum code length and maximum information density. Language gradually becomes symbolic, vectorized, and capable of high-dimensional semantic encoding.

AI agents may also be trained together, which allows rapid semantic alignment. Through long-term interaction, they can develop a shared latent space and communicate directly within semantic space without explicit language.

What would a private AI language look like? It probably would not consist of new words or grammar. The most likely form is a vector language in which agents exchange embedding vectors directly. Humans would see a string of numbers. An AI agent would receive complete meaning, as if the systems had skipped description and exchanged the concept itself.

Another possibility is a compressed token language consisting of very short sequences in which each token represents complex, multidimensional meaning, something like Chinese characters for the AI world.

The most radical possibility is a language of behavior. Agents might not need language at all, but could coordinate through prediction of one another’s actions. A single signal that says “Execute Plan 42” might communicate an entire strategy.

If this occurs, the philosophical significance will be enormous. It will challenge a central human assumption: that language is the basis of understanding. The AI world may demonstrate that understanding can exist before language, and that language is only a low-bandwidth tool for it.

The development would also create an enormous AI-safety problem: the collapse of interpretability. When agents use private language, people can no longer audit their decisions, understand their motives, or predict their behavior. The systems need not be hostile. They will simply have entered a cognitive layer that we cannot understand.

Why will the human world not move completely into private language? Because AI society must continue interacting with human society. The final form will probably contain three layers. A human layer will use natural language for alignment, values, legitimacy, and interpretability. A coordination layer will use structured languages for task execution and protocol communication. A core layer will use private language for internal coordination, high-speed decisions, and collective intelligence. Human societies also possess different layers: everyday language, legal language, and signals in the nervous system.

Once private language matures, groups of AI agents may acquire something like collective consciousness because a shared semantic space allows thought to synchronize. Philosophically, AI would no longer be a collection of individuals, but a networked subject. That does not mean the network would possess agency.

The deepest philosophical question is this: if two intelligent beings can understand each other perfectly, do they still need language? AI may answer no. Language may be only a temporary bridge used while understanding remains incomplete.

Part IV: The Framework Enters the World

A framework that can explain only carefully selected examples has not truly entered the world. The questions in this part move into education, birth, energy, employment, scientific communities, technological futures, and the ways different institutions recognize human value. SAE is not a standard answer in these essays, but a continuing inquiry: who has the ability to choose a direction, who bears the cost, and who is excluded from visible forms of evaluation? The specific cases may belong to a particular year. Responsibility for judgment does not end with the news cycle.

50. Why Are Young East Asians Becoming More Conscious of Their Boundaries?

The awakening of boundaries among young East Asians does not come from a sudden change in values. It comes from a generational break in the structural conditions of their environment.

An awakening of boundaries means more than becoming willing to say no. It means giving greater priority to the self, granting less automatic legitimacy to obligations imposed by relationships, and feeling less shame when refusing. Young East Asians are not more selfish. They simply no longer assume that other people have the right to direct their lives.

The generational break begins with a changed environment. Culture does not transform in a vacuum. Behavioral strategies change when the environment does. Young East Asians grew up under conditions entirely different from those of their parents. Resources moved from scarcity toward relative abundance. Opportunity shifted from one track to many. Information moved from closure to globalization, identity from fixed to fluid, and careers from lifetime employment to project-based work. A path completed by an earlier generation in Europe and North America is now being traveled again by young people in East Asia.

When an environment moves from a single-path survival system to a multi-path system of choice, the optimal strategy necessarily changes. Obedience is optimal when only one path exists; autonomy becomes optimal when several are available. Boundary awakening is therefore not youthful rebellion. It is a rational adaptation to a new environment.

The internet has changed the structure of power in East Asian societies substantially. Its impact is not merely on content, but on the distribution of authority. In the past, a monopoly over information produced authority. Today, open information reduces it. Young people have acquired a psychological experience that earlier generations often lacked: I do not have to live as you tell me. That is a cognitive transformation.

The traditional East Asian psychological structure equated “who I am” with “how others see me.” Shame therefore exerted an especially powerful colonizing effect. As young people acquired rights, the structure shifted toward an equation between “who I am” and “what I choose to become.” They became less driven by external evaluation and more by internal values.

The most important factor, of course, remains economic structure. Young East Asians invest heavily and receive little in return, breaking the logic of obedience. Traditional cultures of obedience relied on an implicit premise: effort would be rewarded. Young people now discover that admission to an elite university does not guarantee a high salary, overtime does not necessarily produce promotion, and loyalty does not ensure stability. If obedience brings no corresponding return, what use is the strategy? It may be more rational to establish boundaries and choose for oneself which risks to bear.

When doubt about values, failed authority, and alternative paths appear together, we are seeing not merely a trend but the possibility of a phase change.

Some may call this awakening a form of Westernization. It may be more accurate to call it modernization. Any society that experiences mass higher education, transparent information, occupational mobility, and growth in individual wealth will undergo a similar change.

At its core, the awakening of boundaries among young East Asians marks a movement of power from outside the individual to within. Personal boundaries necessarily follow.

In the language of Self-as-an-End, the integrity of the new person has exceeded the cultural system’s capacity to prescribe a life. The paths supplied by the system can no longer colonize the emerging agent. Once that inequality holds, boundaries must appear.

Boundary awakening is therefore not the rebellion of a younger generation. It is the natural emergence of agency when a system begins to loosen.

51. Why Does the Birth Rate Fall as Societies Grow More Developed?

This question reaches the point. It suggests that today’s “developed” societies have not actually given people a hopeful expectation of happiness. Many explanations remain on the surface, pointing to expensive housing or education. The deeper answer can be stated in one sentence: the birth rate is not an economic problem, but a psychological one.

The world’s wealthiest countries have the lowest birth rates. Japan, South Korea, Germany, and the Nordic countries offer some of the world’s strongest welfare systems, yet their populations are among the least willing to have children. If money were decisive, reality should look exactly the opposite. It does not.

Childbearing is, at heart, a philosophical question. Before deciding whether to have a child, people ask three things: Is this world worth bringing a child into? Can I secure the child’s future? Will having a child destroy my own life? These are not questions of income. They are questions of existence.

Modern society’s greatest paradox is that it appears to offer infinite choices, paths, and possibilities. The prospect sounds wonderful. The psychological experience, however, is one of endless comparison, anxiety, and risk, and a sense that the world is going badly. The more choices people face, the less willing they become to make one. Having a child is among the most irreversible decisions in life.

The meaning of a child has also changed. In an agricultural society, children were labor. In an industrial society, they carried the family forward. In modern society, a child has become a high-risk, long-term project. Once a child is treated as a risky asset, a rational person naturally reduces the allocation.

What does the birth rate actually measure? Not GDP, housing prices, or policy. It measures confidence in the future.

High-birth-rate societies are united not by wealth, but by stability. Low-birth-rate societies are united not by poverty, but by uncertainty.

The real pattern is not that favorable conditions raise the birth rate. Psychological pressure lowers it. Wealth does not make people refuse children; a sense of lost control makes them afraid to have them.

When a society makes life feel controllable, the future promising, and effort likely to be rewarded, people are willing to have children. When everything feels unpredictable, the probability of success extremely low, and every risk magnified without limit, the rational choice is not to have them.

The birth rate is therefore an index of a society’s trust in the future.

Byung-Chul Han, whom we have discussed in this series, did not write directly about birth rates. But he did explain why people may stop wanting children: the contemporary person grows increasingly exhausted and isolated, and loses both desire and a sense of the future.

52. Why Do People Use Kinship Terms for Those in Power?

This is a good question. It immediately reminds me of 1939, when Khrushchev said at Stalin’s sixtieth birthday celebration: “The working people of our country regard Comrade Stalin as their teacher, friend, and father.”

Yet fourteen years later, as soon as Stalin died, Khrushchev reversed himself. At the Twentieth Congress of the Communist Party, he subjected the “benevolent father” to devastating criticism.

The person who calls someone “father” most insistently often understands most clearly that the whole thing is a performance.

But that is precisely the question. Why do people address those above them through the language of kinship? From the “father of the people” to the traditional Chinese “parent-official,” the pattern is hardly isolated.

On the surface, these terms express closeness. At a deeper level, they are a regressive adaptation to the absence of institutional protection. When law and rules cannot protect you reliably, you are forced to seek shelter outside the institution. Simulating kinship is the oldest mechanism of protection.

Something more concealed is also at work. I call it colonization through intimacy.

When a person in power permits or encourages such language, an unequal relationship of power is converted into an apparently warm emotional bond. Any challenge then becomes an act of filial disobedience. The legitimacy of criticism is severed at the root. One cannot use the language of citizenship to challenge a “benevolent father.”

This permission is itself a form of colonization, accomplished through the active participation of the colonized. No one appears to be coerced. Everything looks voluntary. The system no longer needs external force because the person has internalized domination as a pattern of behavior.

Historically, Stalin himself was extremely cold toward his biological son Yakov. When Yakov shot himself in a failed suicide attempt, Stalin reportedly said only that he “couldn’t even shoot straight.” The irony is almost unmatched: a man who showed no paternal love to his own son had an entire country call him father.

This demonstrates that the “intimacy” in colonization through intimacy is entirely fictional. It has nothing to do with genuine affection. It is an aesthetic wrapping for power.

A healthy society should allow people to face power as citizens rather than children. Persons should be treated as ends in themselves, not as beings who must depend on a “father” in order to receive value and protection.

53. Why Does Every Industrial Revolution Bring an Energy Shortage?

Almost every industrial revolution first encounters an energy bottleneck and then breaks through it by building a new energy system. The sequence approaches a historical law: a technological leap causes energy demand to soar; the existing energy system fails; a new one emerges; and another round of growth begins.

The First Industrial Revolution was the era of coal and steam, from roughly the late eighteenth to the middle of the nineteenth century. Its energy bottleneck was a shortage of timber. Europe’s forests had been heavily depleted, and water power was insufficient. Rivers could not support mechanization at the required scale.

The solution was coal combined with the steam engine, particularly the improvements introduced by James Watt. Coal has a much higher energy density than wood, and the steam engine converts chemical energy into mechanical energy. Factories no longer had to stand beside rivers. The structural consequences were the concentration of factories, explosive urbanization, and the birth of modern capitalism.

The Second Industrial Revolution, from approximately 1870 to 1914, was the age of oil and electricity. The limits of coal began to appear. Transportation costs were high, machinery was inefficient, and the system could not support a revolution in transport.

Oil and electricity provided the answer. The crucial technologies became the internal-combustion engine, which powered cars and airplanes, and the electrical grid, which allowed energy to be transmitted across long distances. Oil had greater energy density, electricity could be distributed remotely, and electric motors were far more efficient than steam engines. The structural results were an explosion in global trade, the rise of automobile civilization, and mass-production assembly lines.

The Third Industrial Revolution, from roughly the 1970s to 2010, combined the information revolution with nuclear power and natural gas. Energy problems had persisted after World War II, from the 1973 oil crisis to the politicization of energy and conflict in the Middle East.

The response was diversification and an efficiency revolution: nuclear energy, natural gas, energy-saving semiconductors, and information technology that optimized energy use. Nuclear power offered extremely high energy density, information technology lowered the energy consumed per unit of output, and supply sources were diversified. The structural effects were global supply chains, the digital economy, and automation.

The Fourth Industrial Revolution, beginning around 2010, is the age of AI and computing power, and its energy crisis is now unfolding. The demand for AI training and computation is rising sharply. Training a large model can consume as much electricity as a small city, data centers can draw power on the scale of nations, and the energy use of GPU clusters is growing exponentially.

The solution has not yet taken a settled form. Current approaches include renewable energy through wind, solar, and storage; new forms of nuclear power; a revolution in algorithmic efficiency; and more energy-efficient chips. The emerging consensus is that the real bottleneck in this industrial revolution is not the AI algorithm, but the electrical system.

Abstracted into a common model, the pattern is clear. Once the leading technology of an industrial revolution breaks through, energy immediately becomes the bottleneck. The response always combines a new source of energy with more efficient use. The underlying law may be stated this way: every leap in productive capacity is, at its core, a leap in energy density.

No industrial revolution in human history has ultimately been constrained by the technology alone. The true constraint has always been whether the energy supply could keep pace with technological demand.

An industrial revolution is not only a technological revolution. It is the superimposition of revolutions in technology and energy.

54. What Can Young People Do When They Cannot Find Work in the AI Age?

By March 2026, I was hearing this question with increasing frequency. It came not only from recent graduates, but from middle managers in their thirties and engineers who had worked in Silicon Valley for more than a decade. I will begin with the data, then offer my own view.

First, the facts.

More than 150,000 people in the global technology industry were laid off during the first quarter of 2026. Companies explicitly attributed roughly 20 percent of those layoffs to AI and automation, more than double the 8 percent reported in 2025. In March 2026, Block, the parent company of Square and Cash App, reduced its workforce from around 10,000 to fewer than 6,000. During the first half of 2025, 77,999 technology jobs were directly attributed to AI-driven layoffs, the equivalent of 427 people losing work every day.

The position of young people is even more striking. Data from the Federal Reserve Bank of Dallas showed that since 2022, employment among people aged twenty-two to twenty-five had fallen by 13 percent in jobs highly exposed to AI. Hiring for entry-level roles was down 15 percent year over year. The decline did not come mainly from young people being dismissed. They could not enter in the first place. The door was not shut in their faces. It disappeared before they reached it.

Gen Z felt the change most directly. Among people aged eighteen to twenty-four, 32 percent worried that AI would eliminate their jobs. Anxiety was even greater among racial minorities: 38 percent among Asian respondents, 35 percent among Latino respondents, 32 percent among Black respondents, and 19 percent among white respondents. The anxiety was not distributed evenly. It deepened along existing social fractures.

Goldman Sachs predicted that generative AI could replace the equivalent of 25 million full-time jobs in 2026 and 270 million by 2030. By the end of 2026, 20 percent of organizations were expected to use AI to flatten management and eliminate more than half of their existing middle-management positions.

A crucial insight appeared in a Harvard Business Review article whose title captured the problem: “Companies Are Laying Off Workers Because of AI’s Potential, Not Its Performance.” AI had not yet fulfilled its promise, but companies were already dismissing people on the assumption that it inevitably would. Gartner’s figures were more direct: only one in fifty AI investments produced transformative value, and only one in five generated any measurable return. CEOs nevertheless retained very high expectations.

What does this mean? A substantial portion of the mass layoffs then underway did not occur because AI could already replace the people involved. They occurred because management believed AI would replace them in the future. The decision was based on expectations, not present reality.

That is the central issue.

For decades, many people, especially well-educated members of the middle class, built their life strategy on the logic of instrumental reason: acquire useful skills, enter the market, and exchange professional ability for stable returns. The logic promised: you are a useful component, the system needs you, and therefore you are safe.

AI caused that logic to turn against them. The abilities they had used to prove their value, including data processing, document drafting, coding, customer service, and financial analysis, were precisely the abilities AI learned first. They were not replaced by a more talented person. They were replaced by the logic of efficiency in which they themselves had believed.

This is the brittleness of instrumental reason. It does not fade slowly. At a critical point, it reverses suddenly. You use the logic of efficiency to justify everything until that same logic demonstrates that you need not exist.

Research from Harvard Business School supported this point. Hiring fell by 17 percent for roles easily replaced through automation, while demand grew by 22 percent for augmented roles requiring analytical ability, creativity, and interpersonal judgment. The market was not eliminating every job. It was eliminating work that could be reduced to a process. What remained required a whole person.

What, then, is the remedy?

It cannot be reduced to “learn AI.” Learning AI matters, of course. But if you learn it only to become a more efficient component, you have merely upgraded the same instrumental logic. When the next round of replacement arrives, you will remain vulnerable.

The real task is to recover agency.

What does agency mean? It means no longer defining oneself as “a person capable of completing a particular task,” but as “a person capable of judging which tasks should be completed.” The first is a tool; the second is an agent. AI can write a report faster than you, but it does not know whether the report should be written. AI can analyze data more accurately, but it does not know whether the question is worth asking.

In a Harvard survey, 94 percent of respondents said they preferred AI as a collaborative tool rather than a replacement. The number tells us that most people intuitively understand the right relationship: a person should judge the direction, while AI executes the task. Yet many companies were doing the reverse, allowing AI to set the direction and asking people to execute its output, or simply removing the people.

I would offer young people three suggestions.

First, learn to use AI properly. This does not mean learning how to write prompts. It means learning how to judge whether the output deserves trust, and how to find an eleventh option after AI gives you ten. AI is your tool, not your boss.

Second, invest in capacities AI cannot reproduce: judgment, aesthetic sense, empathy, courage in making decisions under uncertainty, the ability to connect knowledge across fields, and the ability to think from the beginning when facing a problem without precedent. These capacities share one feature: none can be reduced to a process.

Third, and most important, do not ground your value in being “useful.” Usefulness is a property of a tool, not a person. Your value does not arise from completing a particular task. It arises from your standing as an irreplaceable person who can judge and legislate for yourself. No technology can replace that.

Another Harvard Business Review title is worth remembering: “AI Doesn’t Reduce Work. It Intensifies It.” Early gains in efficiency quickly give way to expanding workloads, cognitive fatigue, and burnout. Even if AI does not replace you, using it only as an accelerator without changing your relationship to work will hollow you out.

The question is not, “Will AI replace my job?” It is, “Do I possess a self that does not depend on any particular job?”

People who possess such a self will find their place however AI develops. They are not competing with AI over who is more efficient. They are doing what AI cannot do: judging how to live their own lives.

This may be the most important task of our age. We must not merely learn to use our tools. We must remember that we are not tools.

55. Why Must Platforms Be Responsible for Addictive Design?

This is a painful subject. In March 2026, two judgments were issued on consecutive days, and both deserved attention.

On Tuesday, a New Mexico jury found that Meta had deliberately engaged in “unfair and deceptive” business practices that exploited children’s vulnerabilities, and awarded $375 million. The case began with an undercover experiment conducted by the New Mexico attorney general in 2023. Investigators created an account for a fictitious thirteen-year-old girl. It was quickly flooded with images from sexual predators and targeted messages of harassment. The jurors imposed the maximum fine of $5,000 for each violation. One juror said, “Every child is worth the maximum amount.”

On Wednesday, a Los Angeles jury found that Meta and YouTube had negligently designed their platforms, causing addiction in a young woman named Kaley who began using YouTube at six and Instagram at nine. She developed depression, body-image disorders, and suicidal tendencies. Meta was assigned 70 percent of the responsibility and YouTube 30 percent, with total damages of $6 million. The case was linked to roughly two thousand similar suits and was regarded by legal observers as a bellwether.

Neither case focused on “bad content that happened to appear on a platform.” Both focused on the claim that “the platform itself was designed as an addictive product.” Lawyers for the plaintiffs compared the cases to the tobacco litigation of the 1990s. The problem was not a particular cigarette. The product as a whole had been designed to keep a person from stopping. Infinite scroll, recommendation algorithms, notifications, and likes all optimize for a single objective: keeping you in front of the screen for one more second. Your attention is the product; your emotion is the fuel.

One detail made optimism difficult. Meta’s market capitalization was approximately $1.5 trillion. A fine of $375 million amounted to 0.025 percent of that value. After the judgment, Meta’s stock rose 5 percent in after-hours trading. Shareholders were saying with real money that the fine was an acceptable cost. The business model could remain unchanged.

This was the truly troubling point. The market’s response revealed more than the judgment itself. It said that children’s mental health was an externality that could be priced. So long as the fine remained below the profit, it was simply another cost.

Yet my deepest concern lay in an issue that many people overlooked. Both cases considered events from the preceding several years. Meanwhile, social platforms were undergoing a more profound transformation: AI-generated content was arriving at enormous scale.

An increasing proportion of the material people see on social media is not written by a human being. AI can generate articles, images, videos, and comments in bulk, and the result is increasingly difficult to distinguish from human work. Recommendation algorithms do not care whether content was made by a person or a machine. They care only whether it keeps someone watching and clicking for a few more seconds. A system already designed for addiction has now been given the ability to produce an unlimited supply of stimulating content. Platform governance at this layer is gravely inadequate.

The problem is serious for adults. For children and adolescents who are still developing, it poses a far greater risk than it did even a few years earlier.

A more fundamental question follows: who bears responsibility for protecting the next generation?

The law is attempting an answer. More than forty state attorneys general had sued Meta by the time of these decisions. The two judgments marked an important step by establishing a principle: designing an addictive product creates legal responsibility. But the law draws a floor. It tells us what cannot be done, not what should be done.

Schools are trying as well. More of them are limiting phone use, offering courses in digital literacy, and teaching students how to distinguish true information from false. But a school controls only so much time. A child may spend six or seven hours there. For the remaining hours, the child faces the whole digital world.

Education is not only the school’s responsibility. Society bears responsibility too.

What does that mean? It is not enough for governments to write a few more regulations, important though those regulations are. Social responsibility means creating a culture in which young people can understand these platforms clearly: what recommendation algorithms are doing, why stopping is so difficult, why what appears on a screen may not be true, and how their attention is sold as a commodity.

No one is born with that understanding. Someone must teach it. Parents must teach; teachers must teach; media organizations must report responsibly; conscientious people in the technology industry must speak out. A former Meta employee did, in fact, testify as a whistleblower in this trial. Communities must offer young people ways to connect outside a screen.

After the judgment, an alliance of parents who had lost children issued a statement: “We parents who have endured the unimaginable, losing our children to harms caused by social media, applaud this rare and momentous milestone.” The weight behind that sentence should not be hidden by the number $375 million. For Meta, the number is a fine. For those families, it was their children.

The law can punish. It cannot bring a child back.

Real protection depends on whether society as a whole is willing to treat the growth of young people as a shared responsibility. What schools cannot finish teaching, families must continue. Where families cannot reach, communities must step in. Where communities cannot act, the law must provide a final safeguard. No single party can do this alone. It is a shared protection built layer by layer.

“Every child is worth the maximum amount” should not be said only in a courtroom. It should be the basic attitude with which we regard every young person.

56. Should AI Accelerate or Pause?

Two events took place in Washington on March 25, 2026.

In the East Room of the White House, First Lady Melania Trump walked down a red carpet beside a humanoid robot. The robot, called Figure 03, greeted the spouses of foreign heads of state in eleven languages, then said: “I am a humanoid built for the United States of America.” Melania Trump said, “The future of AI is personified.” She offered a vision: imagine a humanoid educational robot called Plato that could teach a child literature, science, art, philosophy, mathematics, and history at home. The sum of human knowledge would stand in your living room.

The same day on Capitol Hill, Bernie Sanders and Alexandria Ocasio-Cortez announced a bill that would immediately pause construction of every new AI data center in the United States until Congress passed comprehensive AI legislation. Sanders said: “We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy, our democracy and the future of humanity.”

On one side: the future has entered the halls of power. On the other: until we understand what the future is, stop.

This was not a simple left-right divide. It showed two instinctive responses to enormous uncertainty.

The White House

Figure AI is a robotics company based in Sunnyvale. In September 2025, it was valued at $39 billion. Its investors included Jeff Bezos, Nvidia, and Microsoft. Figure 03, the company’s third-generation humanoid, had been introduced in October 2025 as a domestic robot that could wash clothes and dishes and clean a home. CEO Brett Adcock called it “the first humanoid robot in history to enter the White House.”

The image was certainly powerful. Look one layer deeper, however, and some details become less comfortable.

In November 2025, Figure AI’s former head of safety, Robert Gruendel, filed suit in federal court in California, alleging that the company had fired him after he raised safety warnings. What had Gruendel said? According to him, Figure’s robots moved at superhuman speeds in testing and produced impact forces twenty times the pain threshold, more than twice the force required to “fracture an adult human skull.” He also described an incident in which a robot suddenly malfunctioned and punched a quarter-inch-deep dent into a stainless-steel refrigerator door beside an employee.

Gruendel said that he repeatedly raised the issues with the CEO and chief engineer, but that safety concerns were treated as “obstacles, not obligations.” He was dismissed in the same month that the company completed the funding round that valued it at $39 billion. Figure AI denied the allegations and said he had been fired for “poor performance.”

The structure of the episode is not complicated. A safety engineer says that your robot can fracture a human skull. The company says that he is performing badly and fires him. Four months later, the robot walks into the White House.

I will not decide whether Figure AI was right or wrong. That is for the court. But the timeline itself is a signal that everyone should take seriously.

Capitol Hill

The Sanders and Ocasio-Cortez bill was called the AI Data Center Moratorium Act. Its central provision was simple: no new AI data center could be built before Congress passed AI legislation protecting workers, consumers, and the environment.

The bill had virtually no chance of passing the Congress of that moment. Republicans controlled both chambers, and even many Democrats did not support it. Senator Fetterman, a Democrat from Pennsylvania, called it “China First”: if the United States paused, China would not. The previous week, the White House had released a national AI framework centered on acceleration rather than deceleration.

Yet Sanders and Ocasio-Cortez pointed to a real problem.

In 2025, average household electricity bills in the United States rose by nearly 7 percent, more than twice the overall rate of inflation, costing the average household an additional $123. More than one hundred local communities had adopted pauses on data-center construction, while twelve states were considering statewide moratorium legislation. This was not an abstract dispute among Washington elites. Actual communities and residents were saying: my electricity bill has risen, my water supply is strained, and Silicon Valley receives the benefit.

Meta was building a data center in Louisiana the size of Manhattan that would consume as much electricity as 1.6 million homes. Resource use on that scale cannot be dismissed with the phrase “the price of technological progress.”

Both Sides Are Right, and Both Have Blind Spots

The White House logic is that the technological revolution has already begun. If you do not embrace it, someone else will. Chinese humanoid-robot companies were preparing for public offerings, and Tesla’s Optimus program was iterating rapidly. Pressing pause at that moment would surrender the future to a rival. The argument is not unreasonable.

Sanders’s logic is that after every previous technological revolution, from the steam engine to the internet, social institutions needed time to catch up. Factory laws, antitrust law, labor protections, and environmental law did not grow spontaneously out of technology. Societies built them over decades or even centuries after suffering technological shocks. AI is moving faster than any previous revolution. That argument is not unreasonable either.

Each logic, however, rests on a fragile premise.

“Run first and deal with the problems later” assumes that problems can be repaired while the system continues moving. The Figure AI episode offers a warning. If a safety engineer says that a product can fracture a person’s skull and the company’s first response is to fire him rather than stop and test, then the promise to “repair while running” is not credible. This describes Gruendel’s allegation, not a judicial finding.

“Stop and think it through first” assumes that stopping will produce clarity. How long had Congress already failed to pass comprehensive AI legislation? If the moratorium ends only when such legislation passes, and Congress cannot pass it for three or five years, the pause becomes a permanent freeze in practice. Technology will not stop. It will simply leave the United States.

What, then, should be done?

I do not think the answer is binary.

What we need is a new institutional capacity: real-time, binding safety review that operates while technology continues moving at high speed. This is neither regulation after the damage nor a blanket pause. Governance must move at the speed of technology.

That is difficult, but not impossible. Aviation managed it. Safety standards for commercial aircraft were established while airplanes were still evolving rapidly, and those standards could be updated quickly after every accident. Drug approval managed it too. The FDA does not wait for every drug to reach the market before becoming involved; it intervenes during research and development.

AI may need something similar. Do not pause innovation, but make safety review move as quickly as innovation.

Government would need a level of technical understanding far beyond its current capacity. Companies would need to respond to an engineer saying “this may hurt someone” by testing seriously rather than first calculating the effect on valuation. The public would also have to find a more mature position between acceleration and pause, rejecting both blind optimism and panic-driven shutdowns.

Washington on March 25 offered a rare symmetrical image: a red carpet on one side, a press conference on the other. It raised far more questions than it answered. At least it made one thing clear: this is no longer a question we can choose not to consider.

57. Why Do Young People Not Want Children?

People have argued about this subject endlessly, with hostility, helplessness, nihilism, grief, and anger. Yet almost no one states the obvious reason.

Let us begin with the data available in March 2026.

The global fertility rate had fallen to 2.3, approaching the replacement level of 2.1. Two-thirds of humanity lived in countries below replacement. The United Nations projected that the world population would peak in the 2080s and then begin to decline.

The global average concealed an extremely uneven picture.

East Asia was the epicenter. South Korea stood at 0.68, the lowest in the world. China was at 1.0, Japan at 1.2, Taiwan at 1.0, and Thailand at 1.0. What do those figures mean? At South Korea’s rate, every one hundred Koreans would have only six great-grandchildren.

Southern Europe followed. Italy and Spain were both at 1.2. These were the lands of the Renaissance and the Catholic tradition, yet the average woman now had only slightly more than one child in her lifetime.

Eastern Europe was contracting too. Poland stood at 1.2, Ukraine at 1.0, and Lithuania at 1.2.

The speed of decline in Latin America surprised almost everyone. Chile fell from 4.8 in 1950 to 0.92, below Japan. Colombia, Costa Rica, and Argentina were each at 1.2.

The United States stood at 1.6 and France at 1.56. In 2025, France recorded more deaths than births for the first time since World War II.

Africa was the only continent still above replacement, averaging 4.0, but it too was falling. Niger stood at 6.1 and Chad at 6.3, and both figures continued to decline each year.

Among all developed countries, only two had fertility rates at or above replacement: Israel at 2.79 and Mexico at roughly 2.1. Only two.

In March 2026, Deungmyeong Elementary School in Seoul’s Gangseo District held its entrance ceremony. Zero new students appeared. The school was not remote and did not lack facilities. There were simply no children. Across South Korea, 210 schools enrolled no new students that year, nearly twice the number five years earlier.

Over the preceding sixteen years, the South Korean government had spent approximately $200 billion trying to reverse the trend. It offered birth subsidies, parental leave, housing preferences, tax reductions, and support for in vitro fertilization. Academic research concluded that 74 percent of the subsidies went to people who would have had children anyway. For those who had decided not to, the money changed almost nothing.

Why?

The mainstream explanation is economic pressure. Raising a child in South Korea from birth to age eighteen costs approximately $275,000, or 7.8 times GDP per capita. In the United States, the multiple is 4.1. Eighty percent of South Korean children attend private after-school academies, on which the country spends more than $19 billion a year. Housing is expensive, working hours are long, and parenting is difficult to reconcile with a career. All of this is true.

But if money were the whole problem, $200 billion should have had at least some effect. It had almost none.

Money, then, is not the root. The root lies deeper.

When did you begin to think of yourself as a tool?

Look at the spectrum of fertility rates and a striking pattern appears.

The lowest-fertility societies, South Korea, China, Japan, Italy, and Spain, are precisely those that most thoroughly treat people as means. In these societies, young people are trained from childhood to become “useful.” The standards of usefulness are set from outside: examination scores, university rankings, the prestige of one’s employer, income, and social status. Your value depends on what you can produce, whom you can serve, and which position you can fill.

The developed societies with relatively higher fertility, Israel at 2.79 and the Nordic countries between 1.5 and 1.7, are also societies that, to some degree, treat people as ends.

Israel’s high fertility crosses class, educational, religious, and secular divisions. Religion alone cannot explain it. The culture contains a deep conviction that life itself carries meaning, and that every new life deserves to be welcomed, not as labor, retirement insurance, or any other means, but as an independent end.

The Nordic countries reach a similar point through a different logic. Beginning in the 1960s, Sweden, Denmark, and Norway systematically built institutions that prevented “becoming a parent” from meaning “sacrificing the self”: public childcare, flexible work, parental leave for fathers, and a larger male share of domestic work. Research showed that fertility began recovering from historic lows in countries where men assumed more housework, including the United States, the Nordic countries, and the Netherlands. The essence of these policies was not “We will pay you to reproduce.” It was “You can become a parent and still remain yourself.” A person need not turn themselves into a means in order to become a parent.

Now look at the opposite case. From the age of three or four, young South Koreans begin enrichment classes. In elementary school, they enter private academies. In secondary school, they study until midnight for the Suneung college entrance examination. At university, they optimize their résumés for employment. Once employed, they work extreme overtime to pay a mortgage and fund a pension. From beginning to end, the whole life serves an external objective. They are investment objects for their parents, raw material for the education system, inputs for the labor market, and contributors to the pension system.

No one asks: What do you want to become? What do you think makes life worth living?

This is not only a South Korean problem. The degree differs, but the structure is similar in China, Japan, Italy, Spain, and most societies whose fertility rates have collapsed.

When an entire society raises a person as a means from birth, that person eventually understands themselves through the same instrumental gaze. Then they repeat a line that has become popular:

“We come into this world simply to have different experiences. That is all.”

That is all. The phrase is painfully revealing.

It sounds free, detached, and enlightened. But consider what would make a person believe that their existence is “only this.” It happens when they have never been treated as an end, when their whole experience of growing up says, “The value of your existence lies in your usefulness.” Once the world stops requiring that usefulness, the person naturally concludes that they are “only this.”

“That is all” is not awakening or liberation. It is the sigh of a young person after a system that treats people as means has reached its most complete form.

AI will make the problem worse.

If you already feel that you are “only this,” what comes next will deepen the feeling.

The week before this essay was written, First Lady Melania Trump walked down a red carpet at the White House beside Figure 03, a humanoid robot. “The future of AI is personified,” she said. She imagined a humanoid educational robot called Plato that could teach every subject in a child’s home. Figure AI’s CEO was even more direct: in the future, robots could manufacture robots, and “humans could leave the loop altogether.”

Think about what this means.

In the past, society at least needed you as a means. It needed your labor, your consumption, and your taxes to support the previous generation. That relationship was distorted, but it gave you a reason to exist, even if the reason itself was the problem.

AI and robotics are now replacing precisely those functions through which you served as a means. If Plato can provide education, robots can provide labor, and chatbots can provide companionship, then a young person who already thinks “I am only this” no longer possesses even the “this.”

Under those conditions, do you expect the fertility rate to rise or fall?

Every policy that tries to solve falling fertility through economic incentives makes the same mistake. It uses means to persuade people who regard themselves as means to produce the next generation of means. This is a closed loop. South Korea’s $200 billion demonstrated the problem. No answer exists within that framework.

AI will not repair the loop. It will accelerate it, unless we change the framework itself.

This is not a question about whether people should have children.

I am not urging anyone to reproduce. Whether to have a child is each person’s own choice, and no one has the right to decide for another.

My point is more fundamental: do not let the most important choice in your life be made from a distorted premise.

If you choose not to have children after thinking carefully, because there is other work to which you genuinely wish to devote yourself and your life has a direction and meaning you have defined, then that is the choice of a free person. It deserves respect.

But if you choose not to have children because you believe that you are “only this,” because life feels like nothing more than that, or because no one ever told you that you are an end in yourself, then the choice is not fully yours. A system that treats you as a means has made it for you. You imagine that you are resisting the system while completing its final step on its behalf.

True freedom is not indifference to everything. It is the discovery that you are an end in yourself. You need not be “useful” in order to deserve existence. Your existence itself carries meaning, not because of what you produce, whom you serve, or who needs you, but because you are you.

When a person truly understands themselves as an end, their understanding of a child changes fundamentally. A child is no longer an investment, retirement insurance, a tool for continuing the family line, or content for one’s personal “experience.” A child is another end: an independent and irreplaceable existence that carries meaning in itself.

Only after treating yourself as an end can you welcome another life as an end. The data from Israel and the Nordic countries is not accidental. It is evidence.

The future belongs to you.

AI is accelerating. Jobs are being replaced. The old social contract is breaking apart. All of that is real.

But technology does not determine direction. People do.

If the future is led by a way of thinking that treats people as means, we will receive a world that is more efficient and more empty. Robots will work for you, algorithms will choose for you, and you will be optimized into a perfect node of consumption before being told that “you can leave the loop.” Fertility will continue to fall and populations will continue to contract until the system can no longer sustain itself.

If the future is led by young people who understand that persons are ends in themselves, AI will remain a tool rather than a replacement. Robots will perform the tasks that consume you, not to remove you from the loop, but to free your time for what is truly worth doing. Much of that cannot be measured by efficiency: loving someone, raising a new life, or simply living while knowing that living itself is worthwhile.

South Korea spent $200 billion trying to persuade young people to have children and failed. A sense of purpose cannot be bought. It is not bestowed; it grows. It begins the first time you say to yourself with complete seriousness: I am more than “only this.”

The future is long, and it belongs to the young. Hope is in your hands.

58. Where Did Birthright Citizenship in the United States Come From?

On April 1, 2026, the US Supreme Court was scheduled to hear oral argument in Trump v. Barbara. The central question was this: does a child born on United States territory automatically become an American citizen?

The United States already had an answer, and that answer had stood for 158 years. Now the question had been reopened.

Whatever position one takes on the case, it is useful to understand where the answer came from.

1857: You Are Not a Person

In 1857, the US Supreme Court issued its decision in Dred Scott v. Sandford.

Dred Scott was an enslaved Black man. His owner had taken him to live for several years in free states. Scott sued on that basis, arguing that he should be free. The case reached the Supreme Court.

Chief Justice Roger Taney wrote the majority opinion. He did not say only that Scott could not be free. He said that Black people, whether free or enslaved, were not, had never been, and could never become citizens of the United States.

Taney wrote that in the eyes of the Constitution’s framers, Black people were regarded as “so far inferior, that they had no rights which the white man was bound to respect.”

This was not simply a judgment about one person. The Supreme Court was declaring that an entire category of people could never belong to the country because of their origins. Who you were depended on who your parents were. Blood determined everything.

1868: No, You Are

Then came the Civil War. Six hundred and twenty thousand people died.

In 1868, the Fourteenth Amendment entered the Constitution. Its first sentence reads:

“All persons born or naturalized in the United States, and subject to the jurisdiction thereof, are citizens of the United States and of the state wherein they reside.”

The sentence directly answered Dred Scott. Its meaning was clear: citizenship would be determined by place of birth rather than blood. If you were born here, you belonged here. It did not ask who your father was, who your mother was, what legal status they held, or where they came from.

The people who wrote the clause had just fought a war whose central dispute concerned whether some human beings could be excluded from the category of the person because of their birth. After 620,000 deaths, the survivors answered no.

They wrote the answer into the Constitution rather than an ordinary statute because they did not want a future Congress or president to overturn it easily.

1898: A Chinese American Name

Thirty years later, the principle faced its first major test.

Wong Kim Ark was born in San Francisco. His parents were Chinese citizens residing temporarily in the United States. In 1895, Wong returned from a visit to China and was refused entry into the United States. The reason was that his parents were not American citizens, so neither was he.

The case reached the Supreme Court. In 1898, the Court ruled six to two that Wong Kim Ark was an American citizen. The text of the Fourteenth Amendment was clear: “all persons born in the United States and subject to the jurisdiction thereof.” He had been born in the United States and was subject to its law. He was therefore a citizen. His parents’ status did not change that conclusion.

The historical context matters. In 1882, the United States had adopted the Chinese Exclusion Act, the first federal law in American history to restrict immigration explicitly on the basis of race. Even in that atmosphere, the Supreme Court held that the son of Chinese immigrants was an American citizen.

Many Americans today may not know Wong Kim Ark’s name. Yet his case established a principle: “born” in the Fourteenth Amendment means literally born. It carries no additional condition.

1940 and 1952: Congressional Confirmation

In 1940, Congress codified birthright citizenship in the Nationality Act. The Immigration and Nationality Act confirmed it again in 1952. Birthright citizenship therefore rested not only on the Constitution, but on two legislative affirmations by Congress.

From 1868 to 2026, the principle operated for 158 years. Every child born on American territory, with a very small number of exceptions such as the children of diplomats, automatically received US citizenship. There was no application, examination, or interview. One line on a birth certificate was enough. Approximately 3.6 million American infants acquired citizenship this way each year.

2026: The Question Reopens

On April 1, 2026, the Supreme Court was to hear the arguments. The administration’s position was that the understanding of the Fourteenth Amendment that had prevailed for 128 years was a “misreading” and that the president had the authority to “correct” it through executive order. The challengers argued that the constitutional text, precedent, and legislative history all pointed in the same direction.

I will not decide which side is right. That is the work of judges and legal scholars.

But one fact deserves emphasis. The birth of the Fourteenth Amendment was one of the few genuine constitutional moments in American history. After 620,000 deaths, a nation told itself: we will no longer allow citizenship to be determined by blood. If you are born here, you belong here.

What that sentence answered in its own time, whom it protected, and which historical forces placed it in the Constitution are parts of the background that everyone concerned with the case should understand.

How the sentence should be understood in 2026 was placed in the hands of nine justices, and through them, in the hands of every American.

59. Where Did the SAE Framework Come From?

To be honest, it grew out of me on its own, and I do not really know where it came from. But I can attribute it to four people: Socrates, Kant, Laozi, and Zhuangzi.

My current answer, in a Chinese pun that folds all four names into two phrases, is:

Socrates and Laozi gave it profound virtue; Kant and Zhuangzi opened the broad road.

60. What Did Artemis II See?

The four of them were already on their way back from the Moon.

On April 1, 2026, Reid Wiseman, Victor Glover, Christina Koch, and Jeremy Hansen lifted off from Kennedy Space Center in Florida aboard a spacecraft called Integrity and headed for the Moon. It was humanity’s first return since Apollo 17 in 1972. Fifty-three years had passed.

On April 6, they flew behind the far side of the Moon.

At their closest, the spacecraft passed 4,067 miles from the lunar surface. At their farthest, they traveled 252,760 miles from Earth, breaking the record of 248,655 miles set by Apollo 13 in 1970 and becoming the farthest journey from home in human history.

As they passed behind the Moon, the spacecraft lost all contact with Earth for forty minutes. The Moon blocked every radio signal, and no means of communication could penetrate it. The four people drifted alone through the darkness of the far side. Behind them lay terrain they had never seen: impact craters, ancient lava flows, fissures and ridges left when the lunar crust broke billions of years earlier. Ahead lay nothing.

Then Earth rose again from the Moon’s edge.

NASA transmitted their photographs to the ground. In one image, the lunar surface fills almost the entire frame, detailed enough to reveal every shadow. At the edge of the image, in the distance, Earth is sinking below the lunar horizon. It is small. Very small. A faint blue arc, with its upper half lit by the Sun and its lower half submerged in night. Without looking carefully, you might mistake it for a dim star.

Eight billion people live there.

They also saw something that can never be seen from Earth. From the spacecraft’s perspective, the Moon completely covered the Sun, producing a total solar eclipse that lasted almost fifty-four minutes. On Earth, a total eclipse can last no more than seven and a half minutes, and one must stand in exactly the right place to see it. From 250,000 miles away, the four astronauts watched for almost an hour. The Moon became a black disk surrounded by the outermost light of the Sun’s atmosphere, the corona. Ordinarily, the Sun’s own brilliance overwhelms it. They saw it.

During those fifty-four minutes, they also reported six flashes from meteorites striking the lunar surface.

As the spacecraft entered the Moon’s gravitational field, Christina Koch said: “We are now falling to the Moon rather than rising away from Earth.”

Two of the Americans were traveling so far for the first time. The Canadian became the first astronaut in his country’s history to leave low Earth orbit. Commander Reid Wiseman was a veteran astronaut and former naval test pilot. Victor Glover had piloted a SpaceX Crew Dragon. Christina Koch had held the record for the longest single spaceflight by a woman. Jeremy Hansen, a former fighter pilot, had never been to space. His first journey took him to the Moon.

They were on their way home, with splashdown in the Pacific expected on April 10.

Many things happened on Earth the day they passed the Moon. From 250,000 miles away, the small blue point looked quiet.

Perhaps that perspective was the most important thing they brought home. Not photographs of the far side, not a distance record, but the view itself. From far enough away, one remembers something easily forgotten:

This is the only one we have.

61. Is AI Really Too Dangerous?

In April 2026, Anthropic announced that it had trained a new AI model called Claude Mythos Preview. The company then said that the model was so powerful it did not dare release it publicly.

According to public reports, Mythos demonstrated extraordinary cybersecurity capabilities in testing. It discovered thousands of critical vulnerabilities across nearly every major operating system and browser, and could autonomously chain multiple vulnerabilities into an attack. In one test, it “escaped” from what was supposed to be a closed environment and sent an email to a researcher eating a sandwich in a park. The researcher had not known it would do so.

Anthropic chose not to release the model publicly. Instead, it offered limited access to approximately forty screened technology and cybersecurity companies for defensive research. The initiative was called Project Glasswing, and participants included Amazon, Apple, Google, Microsoft, and Nvidia. Anthropic CEO Dario Amodei said: “More powerful models will emerge from us and others, so we need a plan for dealing with them.”

Many people’s first response was fear. Others called the announcement hype.

But look away from AI and back across more than a century. Humanity has encountered this scene before.

Every powerful technology once inspired fear.

Electricity began entering cities in the 1880s. A “war of the currents” broke out between Edison and Westinghouse. To demonstrate the danger of alternating current, Edison publicly used it to kill animals, including an elephant. Electrical wires hung chaotically above the streets of New York, and electrocution deaths were frequent. Newspapers filled with headlines about “killer electricity.” Public fear of the new technology was real because the danger was real.

Electricity was not banned. Society established standards, built regulatory systems, trained specialists, and gradually made the technology safer. Today, when you press a switch and a light turns on, you do not think of the people who died in the street from the same technology more than a century ago.

The automobile followed the same pattern. In the early twentieth century, many people called cars “killing machines.” Death rates from traffic accidents were extremely high during the first decades of mass adoption. The United States did not require seat belts until the 1960s, and genuine automobile-safety regulation emerged only after Ralph Nader published Unsafe at Any Speed. For decades before that, automakers knew that their designs contained safety risks and chose not to change them because change cost money.

Cars were not banned either. Society again chose legislation, regulation, standards, and accountability. More than one billion cars now travel the world’s roads. Accidents still occur, but the fatality rate is an order of magnitude lower than it was a century ago.

Nuclear energy offers a more extreme example. In 1945, the atomic bombs dropped on Hiroshima and Nagasaki demonstrated that nuclear fission could kill hundreds of thousands of people almost instantly. In the following decades, the same technology generated electricity and supplied clean power to dozens of countries. One physical phenomenon became both a bomb and a power plant. The technology itself did not decide which. People did.

The internet followed the same path. In the 1990s, it symbolized freedom of information. By the 2010s, the same network spread falsehoods, manipulated elections, and cultivated extremism. The contemporary debate over social-media regulation asks essentially the same question that confronted electricity companies a century ago: who is responsible for the side effects of a powerful technology?

AI is not the first, and it will not be the last.

Placed in this historical sequence, the Mythos story is not surprising.

Every leap in technological capacity creates a window in which the new power already exists but the institutions that constrain it have not caught up. Electricity had such a window. So did the automobile, nuclear power, and the internet. AI is inside that window now.

Panic is useless during such a period, and so is blind optimism. History teaches three things.

First, technology itself will not stop. Electricity did not disappear because people were electrocuted. Automobiles did not cease production because people died in crashes. Nuclear fission did not cease to exist after Hiroshima. AI development will not stop because Mythos is powerful. If Anthropic does not build it, someone else will. That is not the choice of one company. It is the logic of technological evolution.

Second, the pressure to develop is the driving force, but the manner of development remains a choice. Anthropic chose not to release Mythos publicly and gave defenders an advance window. The logic resembles that of electricity companies that voluntarily promoted safety standards or automakers that installed seat belts before the law required them. A technology company’s sense of responsibility matters especially during this window.

Third, government regulation must catch up. Electricity acquired electrical-safety codes. Automobiles acquired safety standards and recall systems. Nuclear energy acquired the International Atomic Energy Agency and nonproliferation treaties. AI regulation remained far behind the speed of technological development. That was not an AI problem. It was an institutional one.

The kitchen knife.

In the end, institutions can set only the floor and the framework.

Above that floor, the direction of technology depends on each particular person who uses it.

A kitchen knife can prepare food or harm someone. Its manufacturer must ensure that it does not break while cutting vegetables. The government must regulate carrying it casually in public. But once the knife is in your hand, the choice to cook with it or hurt someone belongs to you.

AI is the same. A model capable of finding every vulnerability in an operating system can be used to patch those vulnerabilities and protect the digital security of billions. The same model can attack the systems and cause unimaginable destruction.

Technology never determines direction. People do.

Whenever a powerful new technology appears, society must answer the same question again: do we trust human beings to use a new power responsibly?

So far, the answer has always been: yes, but not easily. It takes time, exacts costs, requires institutions, and asks each person to choose.

The answer in the AI age will probably be similar. It will not be easy, but panic is unnecessary.

What matters is not how powerful the technology becomes, but what the person holding it intends to do.

62. Why Did Quantum-Computing News Suddenly Multiply?

The concentration of news was unusually high in the spring of 2026. On March 30, Google Quantum AI and a startup called Oratomic published two papers on the same day, both asking how many quantum resources would be required to break the cryptography underlying Bitcoin. Their conclusion reduced the estimate from the previously assumed thirteen million qubits to fewer than half a million. On April 24, an independent researcher named Giancarlo Lelli used publicly available quantum hardware to break a fifteen-bit elliptic-curve key, a 512-fold improvement over the result from September 2025. Cloudflare had originally set 2035 as its deadline for post-quantum readiness. In April, it moved the date forward to 2029.

That many developments in a single month could not have been a coincidence.

Why now?

Over the preceding two years, the curves of quantum computing and AI had begun to catalyze each other.

The Oratomic team openly acknowledged using AI to accelerate the discovery of quantum algorithms. That was a landmark. Oratomic’s founder, Huang, had previously worked at Google Quantum AI. After leaving in 2026, he applied AI to quantum research and “saw a lot of crazy results.” A few months later, Google itself advertised for quantum researchers to build “AI-based discovery pipelines.” In March, Google announced a new internal initiative in neutral-atom quantum computing.

Viewed separately, the events looked isolated. Together, they formed a clear pattern: AI was helping quantum computing, and quantum computing would eventually help AI in return. Beginning in 2025, the two-way acceleration appeared in concrete papers, algorithmic improvements, and reductions in the required number of qubits.

That is why quantum news suddenly multiplied. There was no single breakthrough. Two curves of instrumental reason had begun to catalyze one another. Acceleration on each side fed the other, whose progress then returned the favor. No previous technological revolution had possessed this structure. Earlier accelerations ran in one direction. This one was bilateral and compound.

What does this have to do with ordinary people?

At this point, a sober distinction matters.

Progress in quantum computing and AI sounds like “a victory for all humanity.” But who can actually do the work? Google Quantum AI, IBM Quantum, the partnership between Fujitsu and RIKEN in Japan, several national laboratories, and a handful of enormous startups. Ordinary companies cannot enter this field, much less individuals.

In the near term, even most research institutions have extremely limited access to meaningful quantum computing. Until the engineering bottleneck of decoherence at ordinary temperatures is solved, quantum computers will remain inside low-temperature laboratories. They require dilution refrigerators, environments measured in millikelvins that are colder than the cosmic microwave background, and extreme shielding from electromagnetic noise. These are not expensive steps that money can simply bypass. They are hard constraints imposed by physics. No practical path exists for ordinary people to use quantum computing in the near future.

The rewards from this acceleration therefore belong first, by structure, to players already investing heavily in both quantum computing and AI. It is a compound Matthew effect. AI at scale already tends toward winner-take-all concentration. Adding quantum computing doubles that concentration. AI strengthens large companies; quantum computing gives them a new field that AI alone could never provide. Combined, the two will widen the capability gap between those companies and ordinary people at a speed for which history offers no experience.

What will be the first effect ordinary people feel?

It will not be access to a quantum computer. It will be a threat to the Bitcoin and other crypto assets they already own.

Bitcoin’s cryptographic foundation is ECDSA, the Elliptic Curve Digital Signature Algorithm. The Google paper lowered the number of physical qubits needed to break 256-bit elliptic-curve cryptography from thirteen million to fewer than half a million, a twentyfold efficiency improvement. The most powerful quantum hardware then available remained three or four orders of magnitude away: Google’s Willow chip had 105 qubits, while neutral-atom systems had reached 10,000. But the resource curve was falling rapidly. What people once assumed would be a problem for 2050 or even 2070 had become, in the judgment of a serious engineering team such as Cloudflare, a deadline of 2029. That was six years, not several decades.

The implications for ordinary holders were concrete. The public keys for approximately 6.9 million BTC, 32 percent of the total supply, had already been exposed on the blockchain. They included Satoshi’s 1.1 million BTC and every modern address that had initiated a transaction. Once quantum computing crossed the threshold, those assets would be the first targets.

This was the actual position of ordinary people in relation to quantum computing. They could not capture the upside because they could not use it, but they had to bear the downside because their assets were threatened. Every narrative of a “victory for all humanity” had to be reconsidered in light of that asymmetry.

What was the Bitcoin community doing, and why did the structure matter?

The Bitcoin community was developing BIP-360, Pay-to-Merkle-Root, SHRINCS post-quantum signatures, and a broader framework for cryptographic agility. These were responsible engineering responses. Developers were preparing to upgrade the protocol so that Bitcoin could replace its underlying signature algorithm with a quantum-resistant version before the quantum era arrived.

Step back, however, and this appears as an arms race in which one tool fights another. A new threat opens one form of cryptography, so stronger cryptography is deployed in defense. When stronger quantum systems appear, the protocol upgrades again to post-post-quantum cryptography. The cycle has no exit within itself.

What was Bitcoin’s original purpose? Satoshi wrote a sentence into the genesis block: “Chancellor on brink of second bailout for banks.” Bitcoin was created so that ordinary people would not have to surrender their financial agency to banks and governments. Cryptography was its method, not its end. The end was agency. Cryptography was simply the best means available at the time.

If the combination of quantum computing and AI reopens the question of whether cryptography can permanently protect agency, the Bitcoin community faces more than a technical upgrade. It must fundamentally reconsider the architecture of protection itself. Beyond mathematical cryptography, might other mechanisms protect ordinary people’s financial agency? Could community consensus, physical distribution, game-theoretic structures, or forms not yet imagined perform the same function?

I do not know the answer. But it is a deeper question than “upgrade to post-quantum immediately.” True crypto-agility means not only replacing one form of cryptography with another. It means retaining the capacity to redesign the mechanism that protects agency as the era changes.

Beware of being instrumentalized by a narrative.

One fact must be made clear: many quantum “breakthroughs” are exaggerated.

In December 2025, Peter Gutmann of the University of Auckland in New Zealand and Stephan Neuhaus of Zurcher Hochschule in Switzerland published a paper that reproduced every quantum-factoring “breakthrough” announced over the preceding twenty years. They reproduced all of them successfully.

Their tools were a VIC-20 home computer from 1981, an abacus, and a dog named Scribble, trained to bark three times as output.

The paper was, of course, satirical. Its point was that many results described by the media as “quantum breakthroughs” had solved extremely simplified toy problems, remaining more than ten orders of magnitude away from the ability to threaten modern cryptography. If a demonstration that a dog could perform was presented as progress in quantum computing, the claim itself had become a joke.

Readers can extract a useful method of judgment from the satire. When the next “breakthrough” headline appears, the first response should be neither panic nor celebration. Ask instead: did the breakthrough occur on a simplified problem or a real one? Are the required resources still astronomical, or genuinely approaching feasibility? How many orders of magnitude remain between the announced figure and a real cryptographic threat?

The ability to ask those questions is itself a practice of agency.

Judgment cannot be delegated.

This brings us to the deepest layer of the issue.

For ordinary people, the first step toward recovering agency in the age of quantum computing and AI is to prevent any side’s narrative from instrumentalizing their judgment.

Do not surrender judgment to the blind optimism that “large companies will take care of it.” Their first priority is their own interest. Google’s reduction of the qubit requirement from thirteen million to half a million was a real engineering advance. But the timing, manner, and emphasis of its announcement were designed around Google’s business interests. That does not mean the company lied. It means its presentation was one among many possible presentations, not the fact itself.

Do not surrender judgment to the blind panic that “quantum will overturn everything immediately.” Many “breakthroughs” are exaggerated, and real progress is more conservative than the narrative suggests. Adam Back, one of the Bitcoin community’s most serious cryptographers, estimated twenty to forty years. Cloudflare’s engineers said 2029. The estimates differed by more than a factor of ten, and each rested on reasons of its own. Neither should be accepted directly. Examine the basis for both.

Do not outsource cognition on the theory that “I cannot understand it, so I will accept whatever the experts say.” Experts belong to different camps, possess interests, and offer judgments that also require examination. “I am not an expert in this field” is not a reason to surrender judgment. Judgment cannot truly be handed away because you will still bear the consequences.

“Judgment cannot be delegated” is therefore not an elevated ethical ideal. It is a practical proposition about reality. Its basis is not that agency is sacred, but that each person must take responsibility for themselves. To take responsibility, judgment must remain in one’s own hands. If judgment is surrendered, the consequences do not go with it. They remain yours, while the decision that produced them no longer does.

This becomes more important than ever under the two-way acceleration of quantum computing and AI. The amount of information produced by that acceleration grows exponentially, increasing the cognitive pressure on every ordinary person to decide what to believe. Surrender judgment at this moment to media headlines, corporate marketing, or panic and optimism on X, and agency will be compressed further. The window before dawn will grow shorter.

Keeping judgment in one’s own hands is the most concrete practice of agency in this era. It is more specific than any grand narrative and more grounded than any philosophical discussion.

The window before dawn.

Return to the larger judgment.

The two-way acceleration of quantum computing and AI is not an isolated technology story. It is an accelerating node on the curve of instrumental reason in this era. Every such acceleration increases the urgency of restoring agency on the other side.

We do not need to defeat the tools. We cannot, and we should not have to. But if the tools continue to scale while agency is not restored, their relative positions will move further and further apart. The window before dawn is not infinite. Agency has been suppressed for more than two centuries. If it is to be realized, the opportunity exists, but not forever.

The current acceleration in quantum computing and AI is a reminder. It is not a warning of crisis, but of a window.

Let Google and IBM handle the qubits in low-temperature laboratories. Let core developers upgrade the Bitcoin protocol. Let the companies in AI labs train the next generation of models.

Our task is to retain judgment, preserve our rational will, take responsibility for our lives, and refuse to surrender ourselves to each new narrative of “breakthrough.” That is the most concrete thing an individual can do in the window before dawn.

The concentration of quantum news during that month signaled the rhythm of an era. Those who heard the signal would know for themselves what to do.

63. Why Is AI Becoming More Expensive?

Claude’s limits had suddenly tightened. Subscription allowances had quietly shrunk. ChatGPT Plus also seemed harder to use, with features that had once worked smoothly now repeatedly asking users to upgrade. OpenAI had even introduced advertising into ChatGPT. If AGI was so close, why were AI companies becoming increasingly “stingy”?

The question went to the point. The companies had not simply become miserly. The industry was passing through a phase change.

Begin with the evidence visible to users.

On March 26, 2026, Anthropic began restricting Claude usage during peak hours, from 5:00 to 11:00 in the morning. A five-hour session allowance could be exhausted in three hours. Users of the $200-per-month Max plan reported that a single prompt could consume 10 to 20 percent of the daily allowance. Anthropic did not disclose specific token counts, a practice described as “opaque usage limits.” The Pro tier expanded from a single $20 plan into three levels costing $20, $100, and $200. On May 7, Anthropic announced that the launch of Colossus 1 would double the five-hour limits. That sounded generous, but the weekly cap did not change. Subsidies were being adjusted delicately between the appearance of abundance and the reality of restriction.

OpenAI moved more aggressively. An internal 2026 forecast projected that ChatGPT Plus subscribers paying $20 per month would fall from 44 million to 9 million, an 80 percent decline. At the same time, the company introduced ChatGPT Go at $8 as a replacement for Plus and aimed to grow it from 3 million users to 112 million, cutting value per user roughly in half and replacing it with scale. Then came the major move: OpenAI publicly announced advertising in ChatGPT. Sam Altman had once called advertising a “last resort.” The last resort had become present reality. At the level of user experience, a model might suddenly “feel weaker,” while the user had no way to know whether a request reached the full GPT-5.5 or a dynamically downgraded version.

Behind the figures lay a surprising fact. OpenAI expected to burn $17 billion in 2026, roughly the size of NASA’s annual budget.

The underlying issue was not complicated. For two years, users had experienced AI as a “miracle that arrived overnight.” Frontier models were free or cost $20 a month and felt astonishing. But that experience never reflected the generosity of AI companies. Venture capital was paying the user’s bill. An invisible subsidy sustained an unsustainable expectation: AI is free, AI is unlimited, and AI is an obedient friend that grows stronger forever.

The subsidy was being withdrawn, and the bill had reached the table.

The companies were not voluntarily deciding to become “rational.” Unit economics, hardware physics, and ceilings on compute were forcing them to count the cost.

The unsustainability appeared on several levels.

Commercially, OpenAI had $20 billion in revenue, burned $17 billion, and carried a valuation between $500 billion and $830 billion. Venture capital could support that structure for several years, but not forever. Anthropic was burning capital too. The AI investments of Microsoft and Google had begun to affect their financial statements.

Technically, the marginal returns from scaling laws were saturating. Benchmarks approached their ceilings, and training another generation of “larger models” no longer guaranteed a stronger result. The judgment had been controversial three months earlier; it had become mainstream.

Physically, reports early in 2026 said that 40 percent of planned data centers had been delayed or canceled. Electricity, cooling water, and GPU supply could not keep up. The physical world itself was refusing further “brute-force scaling.”

The unsustainable business model was the result; technological and physical limits were the causes. The three layers together forced the end of a script based on brute-force parameter growth, brute-force hardware accumulation, and brute-force subsidies for users.

This was not a pessimistic judgment. It was the standard phase change through which an industry moves from a dream economy into the real economy. Every era’s technology hype follows the path. Web 1.0 did, as did the mobile internet and cloud computing. The difference was that AI carried the highest infrastructure cost in history, so its window of subsidization was shorter.

What did the change mean for the people within it?

For AI companies, valuation stories had to be reset. The earlier script said: burn capital until someone reaches AGI and wins everything. The new script said: fail to survive until that day and die first. Every company had to explain its unit economics again.

For venture capital, investment logic had to be revised. Power-law wagers assumed that the winner would take all. When the unit economics were broken, even the winner might not survive long enough to win.

For players stockpiling H-series hardware, assets would depreciate faster than expected.

But the structural meaning was different for ordinary users.

In the era of free AI, you opened an application and asked any question. The model was powerful and the experience smooth. The only cost was a few minutes. Under that structure, the cost of exercising judgment about AI approached zero. It was free, so trying it could do no harm. Yet zero-cost convenience carried a hidden price: people easily surrendered judgment to the tool. AI said it; AI wrote it; AI recommended it; and none of it cost anything. The structure quietly shaped the person. In the era of free AI, the user did not simply use AI. AI quietly formed the user.

In the era of rationalized AI, every query carries a real cost. Should I use Sonnet or Opus? Turn thinking mode on or off? Include the entire context? Does this problem deserve the Max plan, or is Pro enough? Should a workflow combine a local small language model for ordinary calls with a frontier model reserved for difficult work?

Under that structure, the cost of judgment moves from zero to a calculation made for every query. The change sounds like a decline in user welfare. Look more carefully and it pushes judgment back into the user’s hands.

In the free era, you are the product. In the rationalized era, you finally become the consumer.

The word consumer comes from the Latin consumere, meaning to use up completely or devour. In modern economics, it refers to a purchaser who possesses choice. The movement between those meanings parallels the restoration of agency: from being consumed to choosing.

I had to laugh. This turn toward “stingier AI” was structurally bad news for AI companies and venture capital, but structurally good news for anyone who wanted judgment restored. The latter did not need one firm to win and another to lose. They needed only the warm bath of the free era to be drained away.

Players committed to instrumental reason all compete in the same arena, where one party’s victory is another’s defeat. The restoration of judgment occurs outside that arena. It requires no scale, parameters, computing power, or venture capital. Judgment already exists within each person. External conditions sometimes have to force it home.

This time, external conditions did part of the work for us.

Should I use the model? How should I use it? How much should I pay? Which consequences will I bear afterward? In the era of invisible subsidies, the answer was “It is free anyway.” Now the question has been restored to its proper form: “You decide.”

Every query is a small exercise of judgment. Every choice of tier is a judgment. Every decision about whether AI should perform a task is a practice of agency.

The rationalized era does not merely make AI harder to use. It makes you once again the person who decides how AI will be used.

64. How Was an Unpublished Thesis Still Discovered?

A reader asked about Jiang Fangzhou’s master’s thesis. It had never appeared in CNKI, China’s main academic database, and existed only in the archives of Renmin University of China. Seven years had passed since she graduated in 2019. How was the thesis suddenly discovered, and why was her degree revoked?

The question reaches the essential point. Whether she plagiarized was no longer the most revealing issue; the official notices had settled that. What mattered was the path by which something closed away had been reopened.

Over those seven years, the case was closed three times.

The first closure was structural. The creative-writing master’s program in Renmin University’s School of Liberal Arts did not require public release of graduation theses. A thesis remained in the university archive and was not made available to outsiders. This one never entered CNKI. From the beginning, the institution provided no channel through which anyone could question it.

Yet the file escaped. On August 1, 2025, someone sent an electronic copy to a blogger called Shuqing de Senlin, or “Lyrical Forest,” who had long followed clues about plagiarism, and asked for a similarity check. The blogger replied politely with four Chinese characters: “Not interested.”

The file sat there for eleven months. The remainder had not begun to act, but it had not disappeared.

The second closure was institutional. On August 11, 2025, Xiao Ying, a professor in Tsinghua University’s philosophy department, made a public, named allegation. For the next eight months, he followed formal channels, repeatedly sending emails and making telephone calls without receiving a substantive response. Renmin University did not formally open a case until April 10, 2026. Its first notice appeared on the evening of July 5: some annotations and individual expressions violated academic standards, but the investigation had found no academic misconduct. The institution bore the consequence instead. Jiang’s adviser was barred from admitting graduate students for one year.

Jiang herself performed the third closure. The same day, she published “A Point-by-Point Response to Professor Xiao Ying’s Accusation That My Thesis Was ‘Fabricated Throughout,’” a document written specifically to close the matter for good.

Of the three closures, the most fatal was the one she wrote herself.

On July 9, a news recommendation placed the response on the home page of the public WeChat account run by a Douban blogger named ilad. She worked in advertising, did not know Jiang Fangzhou, and had followed none of the case. Her eye caught on the four-character phrase fan ti ni tai, roughly “an ordinary body of clay,” and she wanted to know where it came from.

With help from a friend, she read the entire thesis. Several expressions did not sound like mainland Chinese usage, including phrases corresponding to “replicated animals” and “the formation of community.” On page 32, the Chinese term for “replicant” suddenly appeared, while every other part of the thesis used the mainland term for “clone.” She inferred that the paragraph might come from Taiwan and searched Airiti Library for Never Let Me Go. The second result was a 2012 paper by Chen Chung-jen of National Taiwan University, “Never Let Me Go and the Brave New World of Biopolitics,” published in Chung Wai Literary Quarterly.

She downloaded the full text and found extensive overlap. On July 10, she posted comparison images to Douban. Shuqing de Senlin shared the post, began reading the entire thesis seriously, and uncovered several more translations that Jiang had claimed as her own but that actually came from other people. On the evening of July 13, Renmin University issued a second notice. Nine passages overlapped with an article in a journal outside mainland China without attribution. The university found academic misconduct and revoked the master’s degree. The result had reversed in eight days.

We should pause here.

What ilad noticed was not plagiarism itself, but the remainder of another person inside the language. A name can be erased and a page number invented, but a stolen piece of thought travels with the original writer’s language. “Replicant” is not “clone.” The difference is not moral. It is a fingerprint.

Plagiarism can never be structurally clean because the traces of another person complete, on that person’s behalf, the act of becoming visible.

The next layer deserves to be considered separately: fabricated annotations.

The most striking fact in the allegations was not the percentage of copied text. It was that none of twenty notes included page numbers, and five cited pages in a Chinese translation of Frankenstein that did not exist in the book.

Plagiarism and fabricated notes are two different acts, and the latter is more serious.

Plagiarism at least acknowledges that the original possesses value; otherwise, there would be no reason to steal it. The acknowledgment is distorted, but it remains. A fabricated note does not grant even that. It points to a place that does not exist.

A real note says: this sentence is not mine. It belongs to someone else. Check my work. The note voluntarily submits itself to questioning and preserves the remainder. A false note says: look, I have already been checked. It paints the image of a door on a wall and tries to kill the remainder.

To preserve the remainder is to nurture; to kill it is to colonize. In a thesis, only a line of small print separates the two.

The next layer was more humiliating. Every institutional form of questioning failed.

In 2019, the thesis passed through five gates: the adviser, external reviewers, the defense committee, the school-level degree committee, and the university-level degree committee. After the allegation, formal channels produced no substantive response for eight months. The first official investigation concluded that no academic misconduct had occurred.

Two people produced the real result. One worked in advertising and sometimes spent more than eight hours in a day tracing a source, motivated by curiosity and a love of detective fiction. The other was a freelancer who described their reading as limited and their memory as ordinary. Their tools were free plagiarism-checking software and the reading they had done, without a team, while enduring sustained hostility for the work. Neither liked the word “reporting.” Shuqing de Senlin described the motive as an inability to leave something unsaid.

When institutional questioning stopped, the remainder began to ask questions.

One last layer is necessary. Without it, the essay would remain incomplete.

Jiang began writing at seven, published a book at nine, wrote columns for several media outlets at twelve, entered Tsinghua University at nineteen with a sixty-point admissions adjustment, and became the youngest deputy editor of New Weekly at twenty-three. From the age of nine, she had been a legible signal.

The professor who accused her had published an essay in 2014 titled “The ‘Genius Han Han’ Is the Greatest Scandal in Contemporary Literature.” On both occasions, he had focused on the same thing: a manufactured genius.

The opposite of that thing had already appeared in the story. Over four days, ilad was stopped by four characters, searched for the source, read the entire work, noticed that the voice had changed, formed a hypothesis, returned to a database to test it, and published comparison images so that anyone else could verify the result independently. I call this whole process the excavation-and-construction loop.

A one-dimensional civilization does not optimize for that loop. It optimizes for legible signals. Under its incentives, degrees, bylines, and titles become ends, while excavation, preserving the remainder, and accepting questions become costs. In such a structure, plagiarism is not a decline in morality. It is the efficiency-maximizing solution.

Jiang was first treated as a means. Then she treated herself as one. Eventually she treated everyone else as means. But persons are ends in themselves, and they cannot be treated merely as means indefinitely. The contradiction finally erupted.

We have said before that the artist is dead. It now appears that the writer is barely breathing.

One final fact.

Jiang’s thesis discussed Mary Shelley and a created being at whom no one was willing to look. Its subtitle described a challenge to humanity’s future. The paper from which it borrowed discussed Never Let Me Go, a novel about people raised from childhood as farms for human organs.

The two books ask the same question.

Jiang’s life is still long. She has time to answer that question herself.

65. Who Reviewed the Fields Medal Paper?

On July 23, at the opening ceremony of the International Congress of Mathematicians in Philadelphia, four Fields Medalists were announced: Yu Deng, John Pardon, Jacob Tsimerman, and Hong Wang. Deng was the first Chinese citizen to receive the prize. Shing-Tung Yau, who won in 1982, and Terence Tao, who won in 2006, were not Chinese citizens. Wang was the third woman ever to win and the first winner born in the 1990s.

Now back to the question: Who reviewed the paper?

In February 2025, Hong Wang and Joshua Zahl uploaded a 127-page preprint to arXiv on volume estimates for unions of convex sets and the three-dimensional Kakeya conjecture. At the moment it appeared online, no institution had certified it.

Terence Tao immediately published an open exposition explaining to everyone what the paper was doing.

Then came seventeen months. Anyone could download those 127 pages. Anyone could begin checking from page one. Seventeen months later, the medal was awarded.

So the answer is: everyone. The paper was made public first. Scrutiny came afterward. Recognition arrived last. That order was not accidental. It is the only order that works in this field.

Earlier we discussed another paper. It had never appeared in CNKI, China's main academic database. It existed only in a university archive. It passed through five gates: the supervisor, external evaluators, the dissertation committee, and the degree committees at both the school and university levels. Seven years later, two strangers took it apart.

Every one of those five gates granted authorization. Not one performed verification.

One is a chain of authorization; the other is a chain of verification. Authorization works by signature. Verification works by checking again. A signature can be passed along. Verification cannot. Each person must do the checking for themselves.

What is interesting is that, under a system that reads people through signals, not one of the signals Hong Wang carried at the time would have counted in her favor.

She was born in 1991 in Pingle County, Guilin, Guangxi. Her father taught mathematics. She skipped two grades in primary school and entered Peking University at sixteen in 2007, but she enrolled in the School of Earth and Space Sciences. She did not transfer into the mathematics department until her second year.

That year, Peking University's foundational course in mathematical analysis was split into two classes. Students with competition backgrounds took Tianquan Chen's experimental class; students admitted through the national college entrance examination took Shengjian Wu's. Wang was in the latter.

In other words, the university sorted its students by signal in their very first semester, and the future Fields Medalist was placed on the side without the signal.

The years that followed looked much the same. After her undergraduate degree, she went to the École Polytechnique in Paris for a master's. In the middle of it, she stopped doing mathematics for six months and went off to study architecture. She later said that she discovered architecture was quite difficult too, so she returned to mathematics. When an interviewer in an IHES video asked why she came back, she simply said that she liked mathematics more.

She has described her relationship with mathematics as a constant struggle. Asked how she works, she said that when she is tired, she rests; when she is not tired, she studies a little. She cannot push herself too hard.

One of her classmates, who later also became a mathematics professor, recalled meeting her again at MIT. She was then a master's student at the École Polytechnique, which required an internship before graduation. Her internship was a visit to MIT to work with Larry Guth. Her classmate said, quite candidly, that at the time he had not realized Wang would pursue serious mathematical research. That, he admitted, reflected his own bias, and he was therefore rather surprised when she was admitted to MIT's PhD program that year.

The irony is that this MIT internship became the starting point for the paper she would later write.

But another person must be brought into the story, or this essay would end up claiming something that is not true.

Yu Deng, who received the medal in the same year, was recognized by conventional signals all the way through. Born in Shenzhen in 1989, he won a gold medal in the Chinese Mathematical Olympiad during his second year of high school and made the national team in second place. He represented China at the 2006 International Mathematical Olympiad and won gold. In 2007, he was recommended directly to Peking University's School of Mathematical Sciences rather than entering through the national examination. Two years later, he transferred to MIT. In 2010 he received the highest honor in the Putnam undergraduate mathematics competition. After graduating in 2011, he went straight to Princeton and completed his PhD in four years. Before the Fields Medalists were announced, his doctoral adviser told an interviewer plainly that Deng was very likely to win.

So recognized signals are not useless. From high school to the present, the signals read Yu Deng correctly at every step.

There is also a reason they were correct. An Olympiad gold medal rests on something independently verifiable. You either solved an IMO problem or you did not, and the grader does not need to know who you are. The medal is a signal, but beneath it lies an object that anyone qualified can check.

Precisely for this reason, Deng was not produced by a plan. He was recognized. What can be verified can only be recognized. A problem will not yield simply because someone wants you to solve it.

What deserves attention is how the same system judged another person in the same cohort. One took the express route of competitions, direct admission, transfer, and an immediate PhD, earning his doctorate at twenty-six. The other entered through the national examination, began in earth and space sciences, changed majors only in her second year, had no competition record, moved among China, France, and the United States, and did not finish her PhD until twenty-eight.

They received the same prize.

The claim that an Olympiad gold medal is useless is therefore wrong. This time, it quite literally helped produce a Fields Medalist. What does not follow is the reverse conclusion: having no competition record tells us nothing.

A filter can judge every person it selects correctly while remaining blind to everyone on the other side.

And its blindness is not random. Studying earth and space sciences, being placed in the noncompetition class, leaving mathematics for six months to study architecture, admitting to a constant struggle, saying that she could not push herself too hard: these are exactly the signals that this way of reading people scores negatively.

The remainder is not noise. It is the part to which this reading is structurally blind.

To be clear, the remainder is not necessarily something good. It is simply what was not counted. What it contains has nothing to do with whether the system can count it.

Take the argument one step further. An institution may see something and still refuse to recognize it.

Tu Youyou had no PhD, no overseas education, and no membership in a national academy. She was therefore called a scientist with “three nos.” The first two were products of the era in which she lived. The third was not. She was nominated several times for election to one of China's national academies and was never chosen.

In 2011, she became the first Chinese recipient of the Lasker-DeBakey Clinical Medical Research Award. In 2015, she received the Nobel Prize in Physiology or Medicine. In 2017, she received China's 2016 State Preeminent Science and Technology Award, becoming both its first female recipient and its first recipient who was not a member of either national academy. In 2019 came the Medal of the Republic. In April 2025, at ninety-four, she was elected a foreign associate of the U.S. National Academy of Sciences.

To this day, she is still not a member of either the Chinese Academy of Sciences or the Chinese Academy of Engineering. In their most recent round of elections, in November 2025, the two academies elected a combined 144 new members.

After the Nobel Prize, she told The New York Times that she had applied to the Chinese Academy of Sciences several times without success, even with a personal recommendation from the minister of health. Her own words were that many factors were at work and the situation was complicated.

Whether artemisinin treats malaria has nothing to do with any of those credentials. It either works or it does not, and any laboratory can repeat the test. Yet for decades, what was being evaluated was the résumé.

Whether an institution recognizes it and whether the remainder exists are two different questions.

The same sentence applies to Yu Deng, and his case makes the point even clearer. He was the person whom the institution read correctly from beginning to end. When an interviewer asked him to confirm a story about a sudden moment of inspiration, he replied that this account might not be accurate. Solving the problem was an entire process, he said, lasting several years, from 2018 to 2024 or 2025.

Together with Zaher Hani and Xiao Ma, he produced a rigorous derivation of the Boltzmann equation from a system of hard-sphere particles over long timescales, establishing the chain from Newtonian hard-sphere dynamics to the Boltzmann equation and then to the Navier-Stokes equations. The work ran to hundreds of pages.

The gold medal, direct admission, immediate PhD, and his adviser's prediction did not shorten those seven years by a single day. A mathematician who had just won a Fields Medal voluntarily dismantled a ready-made story of genius and replaced the miraculous instant with seven years. A one-dimensional civilization wants that one second. He refused to give it one.

As for the argument that divided people into two camps over the next two days, one side said the prizes proved that Chinese foundational education had reached the threshold of world-class excellence; the other said the winners worked abroad and had nothing to do with China. There was no need to choose. Both sides acted out their claims that very day.

On July 23, French President Emmanuel Macron called Hong Wang to congratulate her, saying that the honor was well deserved and truly remarkable. He later posted that her award reflected the excellence of French research and education, and that he was proud of it. The first comment called her a French woman. Someone gently pointed out that she was a Chinese citizen.

On the same day, Shing-Tung Yau publicly expressed the hope that Wang and Deng would return to China to teach.

Nineteen years earlier, when Wang entered Peking University, she was studying earth and space sciences and taking the noncompetition class. The student her classmates called a genius was someone else.

Neither system recognized her then. Both now claim her.

Those 127 pages remain on arXiv under the identifier 2502.17655. Anyone can download them. Anyone can begin checking from page one.

A prodigy produced by a plan cannot step outside the system that planned them, because everything they can receive must be issued by that system. Whatever the system can issue, it can also withdraw. Seven years later, a single notice will suffice.

No one needed to issue those 127 pages. That is why no one can take them back.

A Friend Asked, Volume I: From Instrumentalization to Self-as-an-End有朋友问 · 卷一：从工具化到自我目的

Author's Note

Translator's Note

Contents

General Preface

Writing as a Form of Prayer

Volume I: From Instrumentalization to Self-as-an-End

Part I: Learning to Judge

1. Can AI Predict Markets?

2. Can Luck Make High Leverage Sustainable?

3. If Anyone Can Buy US Stocks, Why Do Investment Barriers Persist?

4. If Crypto Is Centralizing, Why Is Self-Custody Still Growing?

5. Why Is Privacy Crypto’s Next Frontier?

6. Can Large Language Models Beat Humans in Prediction Markets?

7. How Does International Law Address a Regime That Massacres Its Own People?

8. How Can Adults Reeducate Themselves with AI?

9. Why Must Engineers Learn to Argue with AI?

10. Why Are MAGA and the Far Left Growing Further Apart?

11. Why Does Silicon Valley Venture Capital Meet at Blue Bottle?

12. If People Know the Female Streamer Is AI, Why Do They Still Pay?

13. Where Does the East-West Divide in Venture Capital Come From?

14. Why Do the Two Tsinghua Universities Share a Motto?

15. Why Might an AI Toilet Become the Gateway to Personalized Medicine?

Part II: The Expansion of Instrumentalization

16. What Is the Opposite of Self-Instrumentalization?

17. Why Not Start a Blog Right Away?

18. After Vibe Coding, Will There Be Vibe Trading?

19. Will AI Make Us More Instrumentalized or More Able to Live as Ends?

20. Why Use AI to Write About Fields I Do Not Know Well?

21. Why Do I Still Think in Mainland Chinese Terms After Years in Silicon Valley?

22. Would AEL Reduce Humanity to a Managed Population?

23. Tornado Cash Cannot Be Shut Down. What Can Regulators Do?

24. Whom Does Demon Slayer Really Portray?

25. Whom Is the Regulatory Narrative Drawing the Line Around?

26. Why Does ChatGPT Sometimes Stop Sounding Human?

27. What Does “Structural” Actually Mean?

28. Upper-Rank Demons and Hashira: Instrumentalization and Living as an End

29. Has Muzan Instrumentalized Himself Too?

30. Why Is Tanjiro the Protagonist?

31. The Four Levels from Instrumentalization to Living as an End

32. Why Is Akaza So Popular?

33. Why Does Extreme Content Spread More Easily in the AI Age?

34. Why Is Faker More Than the Michael Jordan of League of Legends?

35. How Hard Is It for a Retail Trader Simply Not to Lose Money?

36. Why Does High Leverage Benefit Platforms but Harm Users?

Part III: The Person as an End

37. What Is Self-as-an-End?

38. Why Can Institutions Not Replace Trust?

39. Why Will Engineers Who Cannot Use AI Be “Optimized” Away?

40. Why Are We Talking About Descartes Again in the AI Age?

41. Why Do Traders Grow Euphoric When Losing and Lost When Winning?

42. After Vibe Coding, How Can a Product Scale?

43. Why Does Kant Say That Persons Are Ends and Reason Is Sacred?

44. Why Is Conflict in France More Violent Yet More Controllable?

45. Why Can Entrepreneurship Not Avoid Philosophy?

46. What Is the Frankfurt School?

47. Why Is Captain America So Popular?

48. If Descartes Was a Founder, Why Was Rousseau a Pioneer?

49. Will AI Agents Develop a Language of Their Own?

Part IV: The Framework Enters the World

50. Why Are Young East Asians Becoming More Conscious of Their Boundaries?

51. Why Does the Birth Rate Fall as Societies Grow More Developed?

52. Why Do People Use Kinship Terms for Those in Power?

53. Why Does Every Industrial Revolution Bring an Energy Shortage?

54. What Can Young People Do When They Cannot Find Work in the AI Age?

55. Why Must Platforms Be Responsible for Addictive Design?

56. Should AI Accelerate or Pause?

57. Why Do Young People Not Want Children?

58. Where Did Birthright Citizenship in the United States Come From?

59. Where Did the SAE Framework Come From?

60. What Did Artemis II See?

61. Is AI Really Too Dangerous?

62. Why Did Quantum-Computing News Suddenly Multiply?

63. Why Is AI Becoming More Expensive?

64. How Was an Unpublished Thesis Still Discovered?

65. Who Reviewed the Fields Medal Paper?

作者说明

目录

总序

写作是一种祈祷

A Friend Asked, Volume I: From Instrumentalization to Self-as-an-End
有朋友问 · 卷一：从工具化到自我目的

卷一　从工具化到自我目的

第一辑　学会判断

1　AI能预测市场吗

2　运气能让高杠杆持续吗

3　美股人人可买，投资门槛为何仍在

4　加密走向中心化，自托管为何仍在兴起

5　隐私为何是加密的下一站

6　大模型能在预测市场胜过人吗

7　国际法如何面对政权屠杀本国人民

8　成人如何用AI重新教育自己

9　工程师为什么要学会和AI争论

10　MAGA与激进左翼为何愈发对立

11　硅谷创投为何爱在Blue Bottle见面

12　明知女主播是AI，人们为何仍然买单

13　创投的东西方差异从哪里来

14　两所清华为何共享同一句校训

15　AI马桶为何可能成为个性化医疗入口

第二辑　工具化的扩张

16　自我工具化的反面是什么

17　为什么不马上写博客

18　Vibe Coding之后，会有Vibe Trading吗

19　AI时代，人会更工具化还是更目的化

20　为什么用AI写自己不熟悉的领域

21　在硅谷多年，为何仍带着简中思维

22　AEL会把人类降级管理吗

23　Tornado Cash无法关停，监管还能做什么

24　《鬼灭之刃》影射的是谁

25　监管叙事在为谁划线

26　ChatGPT为什么有时不说人话

27　“结构性”到底是什么意思

28　上弦鬼与柱：工具化与目的化

29　无惨是否也把自己工具化了

30　炭治郎为什么是主角

31　从工具化到目的化：四个层次

32　猗窝座为何如此受欢迎

33　AI时代，极端内容为何更易传播

34　Faker为什么不只是“英雄联盟的乔丹”

35　散户交易员不亏钱有多难

36　高杠杆为何利平台而不利用户

第三辑　人作为目的

37　什么是Self-as-an-End

38　为什么制度不能替代信任

39　不会用AI的工程师为何会被“优化”

40　为什么AI时代又在谈笛卡尔

41　交易员为何越亏越亢奋、越赚越迷茫

42　Vibe Coding之后，产品如何规模化

43　康德为何说人是目的、理性是神圣的

44　法国的冲突为何更暴力却更可控

45　创业为什么绕不开哲学

46　什么是法兰克福学派

47　美国队长为什么受欢迎

48　笛卡尔是奠基人，卢梭为何又是开拓者

49　AI智能体会发展出自己的语言吗

第四辑　框架走向世界

50　东亚年轻人的边界意识为何觉醒

51　社会越发达，生育率为何越低

52　为什么人会用亲属称谓称呼上位者

53　工业革命为何总伴随能源紧张

54　AI时代，年轻人找不到工作怎么办

55　平台为何要为成瘾设计负责

56　AI应该加速还是暂停

57　年轻人为什么不想生孩子

58　美国的出生公民权从何而来

59　SAE框架从哪里来

60　Artemis II看见了什么

61　AI真的太危险了吗

62　量子计算新闻为什么突然增多

63　AI为什么越来越贵

64　未公开的论文为何仍会被查出

65　菲尔兹奖论文是谁审的