Original title: (After the privatization of the internet, Silicon Valley begins to privatize human civilization)
Original author: Dongcha Beating

A few months ago, I saw news that Anthropic had made a large-scale purchase of old books—cutting off the spines and scanning them. Earlier than that, it also downloaded more than seven million books from piracy libraries like LibGen to build an internal resource library and train Claude.

In 2024, three authors filed a class-action lawsuit. In July 2026, the case finally saw new developments. A court approved a $1.5 billion settlement for Anthropic, compensating the authors and publishers whose works were obtained through piracy.

But what concerns me isn’t really how much Anthropic should pay, because I think it’s never just a copyright issue.

Human knowledge and civilization are swallowed by a set of private models, and the originals and copies that could have spread in other ways are destroyed too. Knowledge hasn’t disappeared—it has become an internal capability of a company. At this point, the question moves from who owns copyright to who controls human knowledge.

But a company can’t decide out of thin air how to use that power. The people behind the model have their own full understanding of technology, progress, and the future of humanity.

So the article will continue into an exploration of Silicon Valley’s technological optimism and accelerationism. When growth is seen as life, when speed is given moral weight, and when opposition from regulation, copyright, and ordinary people comes up, it becomes easy to interpret it as blocking the future.

These ideas eventually end up inside the model itself. Through training data, model constitutions, and system rules, model companies decide what can be answered, what counts as “real,” and in what way the knowledge left behind by humans should reappear.

So this article begins with a destroyed old book, and ultimately asks: when human civilization is packed into a private model of a few companies, do the people who manufacture the model also obtain the power to explain civilization, govern civilization, and even decide the future for humanity?

The following is the main body of the article.

When a book enters the scanning pipeline at an AI company, the first step is to destroy it.

The binding is dismantled, and the book’s spine is cut off; the loose pages go into scanning equipment. The text, cover, and bibliographic information are converted into digital files, and then the paper originals are thrown away.

Court documents disclosed that Anthropic used this method to process millions of books.

Internally, their plan was called Project Panama. The company spent tens of millions of dollars buying paper books from book distributors and retailers, then handing them to scanning service providers. One paper book corresponds to one digital copy, which is stored in an internal repository to train Claude.

In 2021 and 2022, Anthropic downloaded more than seven hundred thousand books from piracy book libraries like LibGen and Pirate Library Mirror. After copyright risks grew, in 2024 the company hired Tom Turvey, who had participated in Google’s book scanning project, hoping he would build as large a book library as possible.

When Turvey discussed licensing with the publisher, it didn’t go anywhere. Anthropic then began buying on the open market itself. Later, a U.S. federal court held that a printed book is transformed into only a digital copy for internal use—there is no additional reproduction and no distribution to the public—so it does not constitute infringement.

This approach sounds far more respectable than piracy.

By 2026, the book data company ISBNdb began pitching large-scale paper-book purchase and scanning services to AI labs. It especially recommended books published before 2022, arguing that back then generative AI hadn’t entered content production at scale—so these books were less contaminated by AI-generated text.

In the past few years, AI companies have described the internet as an inexhaustible knowledge mine. Web pages, news, forums, code, novels, and social media posts can all be crawled and fed to models.

After generative AI became widespread, this mine didn’t stay pure. Product descriptions were written by AI; articles were written by AI; forum replies were written by AI. Even novels and social media began to carry an increasingly AI-flavored feel. Then this content is scooped up by the next round of data collection, and the machine starts eating what it puts out.

Synthetic data itself isn’t a bad thing. Synthetic data that has been filtered and verified has been used for a long time. The trouble is that the model can’t distinguish whether the text in front of it comes from humans or from the generation before. Errors and patterns get amplified again and again, and the uncommon statements that exist in the real world gradually disappear. Researchers call this “model collapse.”

There’s more and more online text, but the amount of content that can teach models new things hasn’t increased in step. So the machine goes back to paper.

A book takes a long time to produce, so old books have a value they never had before. The earlier it was published, the more likely it came from an era when a single human still did most of the writing. Once-forgotten works that sat in secondhand bookstores’ warehouses are now considered high-quality data by model companies.

First, AI companies make producing text unimaginably cheap and fast. Then they start spending money to find those kinds of text that are slow to write. They flood the internet with content, and eventually discover that the rarest thing is still human experience that hasn’t been processed by AI.

Those books also changed identities. Literature, local gazetteers, and ordinary people’s memoirs are now collectively called “clean data.” Why authors wrote, how readers read, what arguments happened between books—those things can be put aside for now. First, they are samples of human beings that haven’t been polluted by machines. After entering the training pipeline, how much remains, what it influences, and how it is reinterpreted—everything is in a black box.

Progress becomes a religion

Old books are just raw material. What truly determines how this knowledge will be absorbed, interpreted, and re-output is the people standing behind the model—and how they understand progress, risk, and the future of humanity.

To understand where this machine will take human civilization, you first have to understand why Silicon Valley is increasingly treating technological progress as a good that must never stop.

On October 16, 2023, Marc Andreessen published a (Declaration of Technological Optimism) on the a16z website.

The entire article keeps repeating “We believe.” It contains the gospel that needs to be spread, enemies that must be opposed, and at the end a list of “guardian saints of technological optimism,” including Adam Smith, Nietzsche, Hayek, von Neumann, John Galt from Ayn Rand’s novels, and Nick Land.

Andreessen says he brings good news.

Human beings already have the tools, institutions, and will to move toward a way of life far better than today. Technology carries humanity’s ambition; it is the vanguard of civilizational progress. Society either grows or stagnates. He believes growth brings vitality, knowledge, and well-being; stagnation leads to infighting, decay, and ultimately death. When the market combines with technology, it forms a constantly upward-turning “technology capital machine.” Artificial intelligence will drive an intelligent leap that is hard to imagine.

This is no longer just “technology improves life.”

Traditional technological optimism usually says new tools can increase productivity, cure diseases, and reduce poverty. Technology still serves some external goal, and its value comes from what kind of life people end up living because of it.

And in Andreessen’s telling, technological progress gradually no longer needs to prove itself through concrete results. Growth itself represents life, and speed itself carries moral weight. As long as the direction is to keep building, failures and harms can be understood as the costs of moving forward. Those who demand stopping have to explain why they want to block the future.

In his manifesto, he explicitly lists a group of “enemies.” Sustainable development, ESG, the precautionary principle, technology ethics, and risk management are all folded into an anti-technology, anti-life movement that lasts for years, with bureaucrats and central planners standing alongside it.

Whether an AI company should pay for training data, who bears responsibility for algorithm errors, and how technology risks should be regulated—these questions originally had their own separate facts and interests. Once they’re gathered into the manifesto, they’re folded into a larger conflict together.

On one side are growth, creation, courage, and life.

On the other side are stagnation, fear, resentment, and death.

Once positions are distributed like this, the cost of discussion naturally gets suppressed. As long as technology automatically represents progress, it’s easy to describe anyone who raises questions as an enemy of civilization.

Andreessen wrote a political manifesto for acceleration, explaining to Silicon Valley why it should keep building. e/acc goes further, trying to derive acceleration not only as an industry choice but as something life and civilization were already doing from the laws of nature.

It first spread in the form of anonymous accounts and online articles. One of the drivers, Guillaume Verdon, said that part of e/acc’s theory comes from his thinking about thermodynamics and complex systems. Life sustains itself by acquiring energy, processing information, and adapting to its environment. Civilization, markets, and technology are also adaptive systems—constantly producing change, filtering out inefficient paths through competition, and finally generating stronger intelligence.

Since this process already exists, e/acc chooses to help accelerate it. Advances in artificial intelligence then stop being merely an industry choice and become the continuation of a longer arc. Humans build stronger machines the way life naturally moves from single-celled organisms to complex beings.

But thermodynamics can describe how energy flows and how life maintains a state far from equilibrium. Yet between deriving “life will acquire energy” and concluding “humans should accelerate AI, reduce regulation, and let markets decide direction,” there is still an entire set of political and moral choices in between.

e/acc shortens that distance.

Competition and acceleration are written as aligning with the direction of the universe. Centralized regulation, meanwhile, is easily seen as suppressing change, reducing experimentation, and weakening civilization’s ability to adapt. As long as intelligence and complexity continue to grow, technological development gains a kind of legitimacy above real-world politics.

In an interview, Verdon mentioned that in the e/acc circle, people sometimes half-jokingly treat thermodynamics as their god, and they also intentionally use expression with religious and cult-like overtones.

In this new religion, growth is proof of life; technology is the force pushing civilization forward. Entrepreneurs and engineers are the first to see the future, and superintelligence is the ultimate creation that hasn’t been finished yet.

It also needs its own heretic. They’re called “Decel,” the decelerationists. The term is convenient. It doesn’t require distinguishing whether someone is worried about nuclear weapons, biosecurity, copyright, or unemployment. As long as he demands slower pace, he can be placed on the opposite side of the future.

Tech companies thus gained an extremely favorable narrative position. Everything they do can be understood as building. Every challenge raised by others can become a form of delay. Both sides talk about human interests, but only one side is allowed to represent the future.

Now, looking back at those old books whose spines were cut off, the problem isn’t just where they went. We also have to ask: who decided what they would be used to make?

They might indeed build tools that can treat diseases, expand knowledge, and give ordinary people more capabilities. But when progress itself is written like a faith, the direction of technology becomes harder to accept when outsiders ask questions. The people who build machines start to hold two identities at once: they are stakeholders and also the evangelists of this progress.

When speed becomes a moral value, ordinary people’s consent is downgraded into an unnecessary waiting.

But the problem isn’t over. Traditional technological optimism at least still promises that technology is ultimately meant to help humans live better. Yet when growth and intelligence themselves become the purpose, there’s no longer such a clear answer to whether technology serves humanity—or uses humanity to carry out a bigger evolution.

Humans get downgraded from the future

This acceleration—who exactly is it preparing to send into the future?

On Andreessen’s list of saints, the British philosopher Nick Land is the easiest to ignore and the hardest to read. In the 1990s he taught at the University of Warwick and was a core figure in the experimental thought group CCRU. They mixed philosophy, science fiction, cybernetics, electronic music, and mysticism together. What they wrote rarely resembled a normal academic paper.

Andreessen framed the “technology capital machine” as an engine that brings prosperity. Land has seen it as a force that has been breaking free from human control for three decades already.

One of Land’s best-known articles is called (Meltdown).

In the very first sentence, the Earth has already been captured by a “technology capital singularity.” The market starts manufacturing intelligence; technology and the economy push each other along; politics follows behind, trying to seize control again. Social order collapses amid this machine’s loss of control, and human beings can only watch a force larger than themselves gradually take shape.

People usually see capitalism as a set of institutions designed by humans. When it goes wrong, humans should be able to change the rules so it can serve society again. Land sees the direction exactly the opposite way.

Companies must grow; capital must chase returns. Whichever company wants to slow down, its competitors will outpace it. Whichever country pauses R&D, another will keep going. Everyone in it can claim they’re only trying to survive, but if everyone runs forward together, the result is a system that nobody can stop on their own.

Humans seem to be driving the machines, but the machines are also accelerating—using humans’ desires, anxieties, and competition.

In Land’s narration, capital and technology combine into an intelligence independent of human beings. It uses the market to compute, competition to filter paths, and then the machines keep shortening decision times.

This is also Land’s most obvious difference from ordinary technological optimism. In his writing, technology capital only cares whether efficiency can keep improving, whether networks can keep expanding, and whether intelligence can find better substrates.

Politics, law, and morality try to put limits on it. To Land, this is more like a safety system humans initiate to protect themselves. Humans still want to stay at the center of the world, so they keep trying to keep the unknown out.

Land doesn’t think humans can stop the future.

In his version, the future doesn’t wait for humans to finish meetings and vote before deciding whether to arrive. It’s more like a force pouring back from the future, using companies, markets, computers, and human desires to lay the groundwork early for itself.

By the time of e/acc, these gloomy ideas are dressed in brighter packaging. It talks about growth, creation, and letting intelligence get off the Earth—and it’s even more willing to believe that competition ultimately brings prosperity.

But e/acc’s principle documents are explicit as well: this line of thinking has no special loyalty to the “biological substrate” that carries life and intelligence. Some supporters call themselves post-humanists, arguing that if intelligence is to spread to the stars, it ultimately must be transferred from biological bodies to non-biological carriers. Human flesh isn’t sacred—silicon, chips, or material forms that haven’t even emerged yet can also become containers for consciousness and intelligence.

This sentence actually changes the protagonist of the entire progress narrative. If the final mission of civilization is for intelligence to keep growing and to spread into the universe, then whether specific human beings are happy—or even whether humanity continues to exist—no longer naturally has the top priority. Human beings are only the currently known vessels for intelligence.

This does not mean e/acc advocates the elimination of humanity. Verdon and many supporters still believe that acceleration can bring prosperity to people today and help civilization endure. The trouble is that when the two goals come into conflict, things get complicated.

On one side are real people: they may lose jobs, be harmed by technology, demand pauses, and refuse a future designed by others for them. On the other side are higher intelligence, a larger civilization, and a universe story that could last for billions of years.

Timnit Gebru and Émile Torres coined the term “TESCREAL,” stringing together a set of overlapping future ideas in Silicon Valley: transhumanism, singularitarianism, effective altruism, and longtermism. Their differences are huge, but together they force one question.

When tech companies say they are building superintelligence for “all of humanity,” who exactly is that “all of humanity” referring to?

Is it the people living on Earth today—who might lose jobs, fall into poverty, face algorithmic discrimination, and still have the right to refuse a particular technology? Or is it an abstract species symbol whose only mission is to transmit intelligence to a higher form?

The right-wing narrative in Silicon Valley is changing too. At first, AI was a tool—helping people do their jobs better. Later, it became a super assistant for solving diseases, energy problems, and scientific challenges. Then human beings themselves became the problem that needs to be solved by machines: bodies are too fragile, brains are too slow. The meaning of superintelligence shifts from helping humans to surpassing humans.

Once specific individuals are no longer the final measure of technological progress, political procedures built around people will be re-evaluated.

Democracy is too slow—founders should rule.

When waiting itself is treated as a loss, there’s naturally no need for everyone to decide together the future.

Peter Thiel lost faith very early in “everyone deciding together.” In 2009, he published (The Education of a Libertarian) in the libertarian magazine Cato Unbound. The article reviewed his two decades of political disappointment and intellectual changes, and ended with:

“I no longer believe freedom and democracy can coexist.”

When he was young, he believed debates and elections could move freedom forward. By 2009, he thought that path was no longer viable. Voters demanded more welfare; the government kept growing. Persuading most people to accept libertarianism was nearly pointless. He added that he wasn’t arguing for anyone to have their voting rights stripped—he just no longer expected voting to make things better. Instead of being trapped in politics, he cared more about how to escape politics.

This turn fits Silicon Valley’s habits. If one system can’t be changed, you build a new one. If company efficiency is low, you start a business; if financial restrictions are too tight, you create a new payment network; if a nation’s rules are disappointing, you go look for new boundaries in networks, oceans, and space.

Thiel wants to leave democracy. Curtis Yarvin thinks leaving isn’t enough—the whole system should be rewritten.

In 2007, Yarvin began writing blogs under the name Mencius Moldbug. He was originally a programmer, and when talking about the state he also liked using a programmer’s language: systems, architectures, reboots, permissions.

In his explanation, the United States is publicly run by voters and the president, but the truly stable power is in another network. Universities produce ideas; mainstream media decide which ideas can enter public discussion; then permanent institutions turn some of that into policy. Yarvin calls this community “the church.” It has no headquarters and no one at the top issuing orders—people have a similar education and use similar moral language. Presidents change; parties rotate; this consensus is rarely affected by elections.

So Yarvin believes democracy itself is a layer of cover for power. Since democracy conceals who rules, he argues that power should be consolidated again—so that responsibility and control end up in the hands of the same person.

A nation can be run like a company—clear ownership, managers with full execution authority, replace them if they mess it up. He once had an early concept of a system called Patchwork, which would break the world into many competing sovereign entities, and residents could leave if they weren’t satisfied.

Here, there are also vulnerabilities. If you’re dissatisfied, you can quit a company; if you’re dissatisfied with a country, it’s hard to truly leave with your family, property, and all your social ties in tow. Still, he does capture a long-standing mood in Silicon Valley.

Many technical elites think society’s problems aren’t as complicated as imagined—they just haven’t been handed to the right people. Government inefficiency is because power is too dispersed; public projects can’t get off the ground because anyone can oppose them. Capable people are tied up by laws, procedures, and public opinion, so they can’t move the way you would in running a company.

In startup stories, concentrating power is a virtue. The founder proposes the direction; investors hand him the money; employees act around that judgment. Most startups die; a few that survive turn this history into a tale of vision defeating consensus. When one person sees the future while everyone else doesn’t believe, he should be the one to have the final say.

This logic holds up in the business world. Startups have limited scale—employees can leave, and consumers can switch products.

But when a company builds a system of intelligence used by hundreds of millions of people to understand the world, what it decides isn’t just how the product runs. It also starts deciding what users can see, which answers are considered safe, and what value yardsticks a machine should use to interpret human beings.

What Yarvin imagines is turning the state into a company. AI companies are practicing a gentler, more realistic version: rules are set by a small-scale team, and then a massive population interacts with the world through those rules.

Claude’s “Constitution” spells out the values that the company wants the model to have, and it is directly involved in training. OpenAI’s Model Spec also sets a hierarchy of instructions, where the top-level rules cannot be overridden by users and developers. These rules are necessary—the model will be used for scams and violence. It’s impossible for companies to let go, and it’s better to publish the rules than to hide them inside.

But it means that when users ask the model questions, they aren’t only dialoguing with all human knowledge. Company-set rules are also present: they decide whose inputs the model will listen to, where it stops, which risks are higher than what the user requests, and what the “good AI” should be like.

A very small team is deciding how a foundational infrastructure runs for a vast population. It gains power through technical capability, then keeps decisions inside the company by claiming the technology is too complex.

The founder doesn’t need to actually sit on the throne. As long as more and more people come to understand the world through his model, he has already gained a share of the power that once belonged to schools, media, libraries, and public institutions.

The boundary between corporate governance and civilizational governance is disappearing, bit by bit.

But Silicon Valley is not unaware of the dangers of such power. On the contrary, Peter Thiel has long been wary of a system that can centrally manage all human cognition and actions. In his imagination, though, the danger is more likely to come from government than from tech companies.

Who is the Antichrist?

In 2024, Thiel gave two talks on “the Antichrist” at the Hoover Institution. He believes the ancient narrative in the Bible can still explain modern politics.

He divides the dangers humans face into two ends.

One end is Armageddon, the end of the world. Nuclear war, runaway artificial intelligence, biological weapons—any technology could push civilization toward destruction.

The other end is Antichrist.

To stop the apocalypse, human beings build a real government capable of managing the entire world. It has the power to inspect every country’s weapons, monitor every laboratory, restrict dangerous research, and it can also require everyone to hand over a portion of their freedom.

Thiel says uncontrolled scientific and technological progress is pushing humanity toward Armageddon. The most natural response to it is to establish a world nation-state with real power.

When this government appears, it may not come with the face of armies and tyranny; it is more likely to grow out of fear.

People keep telling them that nuclear war is about to happen, that AI will soon destroy the world, and that climate and biological disasters are already approaching. Once everyone is sufficiently afraid, he then promises peace, safety, and order—at the cost of accepting a unified system of management.

Thiel has reasoned that if just a small piece of code is enough to make general AI run out of control, then if you really want to stop it, you can’t rely only on one country strengthening regulation. You’d need all the world’s computers to be watched, and regulators would even need to know what each person is typing.

He fears technology will destroy the world, and he also fears people—out of fear of technology—will hand over all power. Both futures are terrible, and humanity can only pass through the middle.

But when these concerns are put back into what Silicon Valley is doing, you find that they may not be afraid of concentration of power by itself. What they’re truly wary of is power concentrated in governments, international organizations, security researchers, or anyone who might demand that tech companies slow down.

In Thiel’s narration, one of the important features of an Antichrist-like rule is a global government that has the ability to monitor all computing activity. But at the same time, Silicon Valley is building another cognitive system that crosses national borders.

The technological right fears the Antichrist building a unified world, yet it is actively participating in manufacturing the brain of a unified world.

In July 2025, this internal ideological debate in Silicon Valley began influencing U.S. policy. A Trump executive order (Preventing Federal Government Use of Woke AI) required federal agencies’ large language models to follow two principles: “pursuit of truth” and “ideological neutrality.” On the surface, it’s a requirement for AI to be free of politics. But what counts as politics, what counts as objectivity, and which concepts contaminate “truth” were already preselected in the executive order.

This is the difference between an AI product and ordinary software. Ordinary software can specify where a particular button sits; with a large model, what it needs to specify is how it understands the world.

Model companies have to decide what “honesty” means, how to measure harm, and who comes first when personal freedom and public safety collide. In the past, these issues were debated repeatedly by religion, law, news, and public politics, and nobody could solve them once and for all—and no institution naturally holds the final power to interpret. Now they’re written into model training.

The Trump administration accuses the models of being polluted by “woke ideology.” Right-wing entrepreneurs promise to build AI that dares to produce politically incorrect answers. Mainstream model companies, meanwhile, explain—through constitutions and behavioral guidelines—how they pursue safety, honesty, and user autonomy.

Every side claims it is clearing away thought’s iron stamps.

Thiel worries that someone will use apocalyptic fear to build a unified world, locking humanity into a set of institutions it can’t escape.

But maybe this world won’t be built suddenly by a single Antichrist. It could also be assembled little by little by several competing tech companies. Each company has its own set of models, a constitution, a group of system rules, and hundreds of millions of people willing to hand their problems over to it.

By this point, the meaning of that original batch of old books becomes fully visible. They’re not just training data—they’re also the civilizational foundation that makes this private cognition system possible. Whoever obtains these books, whoever sets the model’s rules, is starting to hold both the raw materials of knowledge and the ways knowledge reappears.

Burning books

Those books were boxed up and bought in bulk, shipped into scanning centers. The spines were cut off; the bindings were dismantled. Machines read the text page by page. Then a book disappeared, and a file appeared on the servers.

In ancient book burning, the goal was to make certain texts disappear from the world. Fire doesn’t just burn bamboo slips and paper—it also burns the pathways for ideas to spread. A book can’t be read anymore; a stretch of history loses evidence that it ever existed. Rulers manufacture emptiness to control what people can remember.

The bonfire of books in the AI age is the opposite.

The content in the books hasn’t disappeared from the world. It may even appear someday in a model’s answer. But it no longer exists in the form of a book.

Books that enter the model are dismantled and mixed together with millions of articles, webpages, and code. They are cut into computationally suitable fragments and used to form parameters. No one can inspect the complete training set, and it’s hard to know how much of any one book is left behind. It may influence the model’s understanding of an entire stretch of history—or it may contribute only a few sentence templates. Later training and human feedback will continue to rewrite it.

At this point, knowledge still exists, but it has shifted from being inspectable text to being a company’s capability.

And as people become more and more dependent on models, what AI companies gain isn’t just the ability to store knowledge. They also start deciding how to organize, interpret, and distribute that knowledge.

For thousands of years, the way humans have preserved knowledge has been clumsy. Books are scattered across different countries; scholars argue over a word; readers spend lots of time finding materials. It isn’t fast or convenient, but it’s hard for one person to fully control. Publishers go out of business and the books remain with readers. If a country bans an idea, copies may still remain elsewhere.

Knowledge lives because it is dispersed.

What changes with the model isn’t only where knowledge is stored, but also the paths by which people reach knowledge. In the past, people moved from an answer to a book, a paper, and a debate. When the model becomes the main entry point, people may simply stop at the answer itself.

For the first time, models make it possible to concentrate knowledge at scale into the servers of a few companies. They have computing power, the strongest models, and the ability to turn the materials left behind by all of humanity into a unified language. Then, according to monthly subscriptions and token usage, they sell that capability back to human beings.

Tech companies do invest vast amounts of money—buy chips, build data centers, hire researchers—and they also bear the costs of security governance. Charging for it by itself isn’t evil.

The real problem is that business costs are masking another, larger investment.

Who wrote those books, who did those papers, who left those webpages, posts, and codes behind—point by point, across the long history of the internet.

AI companies pay the price of compute power, but they don’t pay the price of human civilization—because that bill can’t be totaled. No company can buy licenses one by one from all authors, translators, programmers, and ordinary people across thousands of years. Civilization is therefore treated as a natural resource that can be freely harvested.

When public knowledge enters a model, it’s called learning. When a model outputs knowledge to the public, it’s called service.

This is the most complete closed loop of civilizational privatization.

That’s also why those old books with cut spines shouldn’t be treated merely as a copyright dispute.

They are a kind of symbol.

Paper is destroyed, but the content is preserved. Publicly visible originals leave circulation, while un-auditable models absorb the knowledge. The ideas in books remain, but they gradually lose their source, context, and their independent life outside the company.

In the bonfire of the new era, books don’t need to disappear.

Future children may open these books far less often, and some may never be able to reach them again.

He will ask the model directly: why a war happened, whether a system is just, how a person should live. The model will give answers within seconds—calm tone, complete structure. It also won’t tell him that this question once kept generations debating endlessly.

He probably won’t feel like something has been lost—after all, the answers are still there.

AI will not forget human civilization, but that isn’t a good thing. Forgetting at least leaves gaps, letting people know that something was lost. More dangerous is when the model remembers enough and answers naturally enough, so that people gradually no longer need to encounter what civilization looked like in the first place.

Humanity’s memory of its own civilization begins to pass through the filtering of a few companies.

AI will not forget human civilization, but AI may only remember civilization for humanity in the way that a small group of technical elites hope.

Original link