YouTube Digest

March 14, 2026

Listen to this issue

English

Based on “The deep mystery physicists call “the problem of time” | Jim Al-Khalili: Full Interview” from Big Think Watch the original video

The Ghost in the Machine: Unraveling the Great Mystery of Time

We feel it in the graying of our hair, the frantic pace of a deadline, and the slow crawl of a rainy Sunday afternoon. To us, time is the most fundamental reality of existence—a river that flows relentlessly from a fixed past toward an uncertain future.

Yet, if you ask a physicist to show you this “flow” in the fundamental equations of the universe, they will come up empty-handed. In the world of math and particles, time is often reduced to a mere coordinate, a humble lowercase t that behaves more like a map reference than a rushing current.

“If you look up the ‘problem of time’ online, you’ll find various definitions,” says Jim Al-Khalili, Emeritus Professor of Physics at the University of Surrey and author of On Time. “But it really boils down to four distinct problems: Does time flow? How do we reconcile quantum mechanics with relativity? What is special about ‘now’? And where does the direction of time come from?”

In an age where we can synchronize atomic clocks to the billionth of a second, the fact that we still don’t know what time is remains the deepest mystery in science.

Manifest vs. Physical Time: The Dentist and the Party

The first hurdle in understanding time is separating our internal experience from external reality. Al-Khalili uses the term “manifest time”—a concept popularized by philosopher Craig Callender—to describe time as we perceive it.

We’ve all experienced the elasticity of manifest time. When you are five years old, the year between birthdays feels like an eternity because it represents a massive percentage of your total life experience. By fifty, years go by in a flash. Conversely, on a shorter scale, thirty minutes in a dentist’s waiting room with no phone can feel like hours, while thirty minutes at a lively party vanishes in an instant.

“At the party, you’re laying down new experiences, chatting, and taking in sensory input,” Al-Khalili explains. “Yet that time goes by much more quickly than the time that drags when you’re bored.”

While our brains stretch and compress time based on emotion and memory, physics suggests a much more rigid structure. For centuries, following Isaac Newton, we believed in “Absolute Time”—a cosmic clock ticking away at a constant rate regardless of what was happening in the universe. But in 1905, Albert Einstein shattered that clock forever.

The Relativist’s Clock: Why Your Head is Older Than Your Feet

Einstein’s Special Theory of Relativity revealed that time is not a universal constant; it is relative to the observer. If you travel close to the speed of light, your clock slows down relative to someone standing still.

This isn’t just a mathematical quirk; it’s a physical reality proven by subatomic particles called muons. Created in the upper atmosphere, muons have a lifespan of only a fraction of a second—too short to reach the ground. Yet, because they travel at near-light speeds, their internal clocks slow down (time dilation), allowing them to survive long enough to be detected at Earth’s surface.

Einstein’s General Theory of Relativity added another layer: gravity also slows time. “Everyone who uses a smartphone or Google Maps is making use of the fact that time runs at different rates,” says Al-Khalili. GPS satellites are further from Earth’s gravity, meaning their clocks tick slightly faster than those on the ground. To keep your navigation accurate, engineers must manually slow down the satellite clocks to stay in sync with Earth.

The implication is startling: because gravity is stronger closer to the Earth’s center, your feet are technically younger than your head. The difference is infinitesimal, but it is real.

The Block Universe and the Illusion of “Now”

If time is relative, then the idea of a universal “Now” becomes impossible. If two stars explode at the same time from your perspective, an observer zooming past in a rocket might see one happen long before the other. Neither is wrong.

This leads many physicists to the “Block Universe” theory. Imagine the universe as a giant loaf of bread. Each slice represents a moment in time. In this view, the past, present, and future all exist simultaneously. Your birth is at one end of the loaf, and your death is at the other; both are equally “real” and eternally present within the four-dimensional structure of space-time.

“This idea is called Eternalism,” says Al-Khalili. “All times coexist. Our consciousness is simply drifting along a ‘world line’ through this block.”

This raises a troubling question about free will. If the future is already “there,” waiting for us like a pre-recorded movie, do we actually have a choice in our actions? Al-Khalili describes himself as a “compatibilist.” Even if the universe is deterministic and the future is set, the fact that we can never predict it from within the system gives us a functional version of free will. “I’m making free choices, and that’s good enough,” he remarks.

The Quantum Vanishing Act

The “problem of time” reaches its peak when we try to merge the physics of the very large (Relativity) with the physics of the very small (Quantum Mechanics).

In the 1960s, physicists John Wheeler and Bryce DeWitt attempted to write a quantum equation for the entire universe. The result, known as the Wheeler-DeWitt equation, had a shocking feature: the variable for time (t) vanished completely. The equation describes a universe that simply is, without any reference to change or duration.

This has led some to suggest that time is an “emergent property,” much like the “wetness” of water. You can’t find “wetness” in a single H2O molecule; it only emerges when trillions of molecules interact. Similarly, time might not exist at the most fundamental level of reality, only emerging when we zoom out to the macroscopic world.

The Arrow of Time: Why We Can’t Un-Spill Milk

If the fundamental laws of physics are “time-symmetric”—meaning they work just as well backward as forward—why does the real world only move in one direction? Why do batteries die, buildings crumble, and people age, but never the reverse?

This is the “Arrow of Time,” usually attributed to the Second Law of Thermodynamics: entropy (disorder) always increases. A shuffled deck of cards will never spontaneously un-shuffle itself.

However, Al-Khalili points to a more modern explanation: Quantum Decoherence. In the quantum world, particles are “entangled” with their surroundings. As a system interacts with its environment, its “quantum-ness” leaks out and becomes irreversibly lost.

“Decoherence is regarded as the one truly irreversible process in nature,” Al-Khalili argues. “I would argue that the arrow of time is baked into reality… due to quantum entanglement and decoherence increasing all the time.”

Conclusion: Is Time Real?

While some philosophers and physicists argue that time is a stubborn illusion, Jim Al-Khalili leans toward its reality. If the direction of time is built into the way quantum particles interact with the world, then time itself must have a foundation.

We may be trapped within the river, unable to step onto the bank and view the flow objectively, but that doesn’t make the water any less real. Whether time is a fundamental dimension of a frozen block universe or an emergent property of quantum entanglement, it remains the stage upon which the entire drama of existence is played out.

As Al-Khalili notes, we may never fully extract ourselves from time to see its true face, but by studying the way the universe ticks, we come closer to understanding our place within the “ever-changing present.”


Based on “667. Here’s Why You Are Constantly Fighting Off Scammers | Freakonomics Radio” from Freakonomics Radio Network Watch the original video

The Billion-Dollar Meat Grinder: Inside the Global Scam Industry

In January, Chen Xi, a 38-year-old Chinese-born entrepreneur, was arrested in Cambodia and extradited to China. On the surface, Chen was the picture of success: one of the wealthiest men in Cambodia, a high-level government adviser, and the head of a massive real estate development firm. But beneath the veneer of legitimate business lay a darker reality. The U.S. government designated his holding company a transnational criminal organization, charging him with fraud and money laundering while seizing a staggering $15 billion in cryptocurrency.

Where did that money come from? According to U.S. prosecutors, Chen was a kingpin of “pig butchering”—a chillingly descriptive term for a scam where victims are “fattened up” with months of romantic or friendly attention before being “slaughtered” for their life savings.

This isn’t just a series of isolated crimes; it is a sophisticated, multi-billion-dollar global industry. In Cambodia alone, cybercrime is estimated to generate $19 billion annually—roughly half of the country’s GDP. In 2024, scammers in Southeast Asia stole an estimated $10 billion from Americans. To understand why your phone is buzzing with “wrong number” texts and your inbox is flooded with suspicious offers, you have to understand the brutal efficiency of the modern scam economy.

The Anatomy of an Industry

We often imagine scammers as lone hackers in dark basements. The reality is far more corporate. Modern scam operations, particularly those in Southeast Asia, function like Fortune 500 companies. They have HR departments, legal teams, marketing experts, and strict performance quotas for their “employees.”

Disturbingly, many of the people sending those messages are victims themselves. “Thousands of workers fled [Chen’s compounds] after his arrest,” notes the Freakonomics Radio team. “They had reportedly been trafficked to Cambodia and were being held against their will.” These captives are forced to run “boiler rooms,” engaging in “AB testing” on scam scripts to see which emotional triggers result in the highest payout.

“It is absolutely an industry,” says Marty De Lima, a gerontologist at the University of Minnesota who studies the impact of fraud. “A very complex, always evolving, very competitive industry.”

The Myth of the Vulnerable Senior

There is a persistent myth that scams primarily target the elderly and the technologically illiterate. While older adults do lose more money per incident—the median loss for those over 80 is $1,400 compared to $500 for those under 50—they are not the only targets.

“It’s a myth that older adults are more trusting or lack sophistication,” De Lima explains. “Data shows that middle-aged adults report victimization at the highest frequency.”

The type of scam often dictates the demographic:

  • Younger People: More likely to fall for fake job opportunities and online shopping scams.
  • Middle-Aged Men: Frequently targeted for investment and cryptocurrency fraud.
  • Older Adults: More susceptible to tech support and lottery scams.

The psychological toll is universal. De Lima describes scamming as “betrayal trauma.” The loss of self-efficacy and the shattering of one’s worldview often lead to a sense of hopelessness so deep that it can result in suicide—a connection that is tragically underreported.

Why Do Scams Look So Obvious?

We’ve all seen them: the “Nigerian Prince” who needs help moving an inheritance, or the email from a stranger offering a free Yamaha baby grand piano. To a discerning eye, these look like amateur hour. But there is a cold, economic logic to the typos and the far-fetched premises.

In a 2012 paper, Microsoft researcher Cormac Hurley asked: Why do Nigerian scammers say they are from Nigeria? The answer is filtering. Scammers don’t want to waste time on people who are skeptical. By using a premise that is obviously a scam, they ensure that only the most gullible individuals—those most likely to actually transfer money—respond. It’s a way to increase their “conversion rate” while minimizing the labor cost of chasing “leads” that won’t pay out.

However, as AI enters the fray, this “obvious” phase is ending. “If you can have your AI do the next step of following up with semi-interested people, that’s not very costly,” says De Lima. Scams are becoming more tailored, more professional, and harder to spot.

The Psychology of the Con: System 1 vs. System 2

Scammers are masters of human psychology, specifically the dual-process theory of the mind. Our brains operate in two modes: System 1 (fast, intuitive, emotional) and System 2 (slow, analytical, logical).

A scammer’s goal is to shut down your System 2. They do this through “emotional arousal,” which can take two forms:

  1. High Positive Arousal: The promise of a windfall, a romantic connection, or a “ground floor” investment opportunity. This exploits our desires and greed.
  2. High Negative Arousal: The threat of arrest, a “seized” bank account, or a kidnapped relative. This exploits our fear.

When we are in a state of high emotion, we rely on “heuristics”—mental shortcuts. We trust a voice because it sounds like a loved one (often via AI voice cloning), or we comply because a caller ID says “Bank of America.”

“Criminals will do anything they can to shut off our analytical processing,” De Lima says. “They want us to go straight to those shortcuts and comply.”

The AI Arms Race and the Myth of Privacy

We are entering a new era of fraud where “common sense” advice is becoming obsolete. Previously, experts told consumers to look for spelling errors or to ask for a video call. Today, AI can generate perfect prose and “deepfake” a video call in real-time.

One of the most sophisticated modern tactics involves exploiting the bureaucracy of legitimate institutions. A scammer might send a text pretending to be your bank, asking if you authorized a purchase. When you don’t click the link (as a “wise” consumer shouldn’t) and instead call the bank’s real number, you might be put on hold. Scammers know this. They wait 30 minutes and call you back, faking the caller ID to look like the bank. Because you just tried to call them, your guard is down.

“Privacy is a myth,” warns De Lima. “Our information is out there and available to the highest bidder. Criminals know our names, Social Security numbers, addresses, and mother’s maiden names.”

Who is Responsible?

As the scale of fraud reaches hundreds of billions of dollars, the question of accountability has shifted from the scammers to the platforms that enable them.

Katie Daffen, a former Assistant Director at the FTC, notes that the commission is increasingly going after “facilitators.” One example is the payment processor Paddle, which was charged with allowing scammers access to the credit card system despite internal warnings.

But the biggest elephant in the room is Big Tech. Leaked documents from Meta (the parent company of Facebook and Instagram) reportedly suggested that a significant portion of the company’s revenue comes from ads for scams. While Meta claims they “aggressively fight scams” and took down nearly 11 million accounts last year, critics argue they aren’t doing enough.

“The technical capacity of these companies to identify and flag scams is there,” says De Lima. “They’re making a calculated choice.”

The Erosion of Social Trust

Perhaps the most insidious cost of the scam industry isn’t financial—it’s the erosion of social trust. When every phone call is a potential threat and every email is a “pig butcher” in disguise, the fabric of human connection begins to fray.

Social trust in the U.S. has declined sharply over the last 40 years, and the constant barrage of fraud is a primary driver. We have become a society that refuses to answer the phone. Marty De Lima admits that even she, an expert in the field, recently missed a legitimate mandatory cybersecurity training because the email looked too much like a scam.

“This is the world that we’re now in,” she says.

To protect yourself, the advice is simple but cynical: treat every unsolicited communication as a lie. Independently validate every claim. Don’t use the phone numbers provided in messages; find them yourself. In the age of the billion-dollar meat grinder, the only way to avoid being “slaughtered” is to never let yourself be “fattened up” in the first place.


Based on “Dylan Patel — The Single Biggest Bottleneck to Scaling AI Compute” from Dwarkesh Patel Watch the original video

The Trillion-Dollar Silicon Ceiling: Inside the High-Stakes Race for AI Compute

In the quiet corridors of Silicon Valley and the sprawling industrial parks of Taiwan and the Netherlands, a number is being whispered that would have seemed like science fiction only two years ago: one trillion dollars.

That is the approximate scale of the capital expenditure (CapEx) currently being funneled into the global AI supply chain. The “Big Four”—Amazon, Meta, Google, and Microsoft—have forecasted a combined CapEx of roughly $600 billion this year alone. When you add the frantic fundraising of AI labs like OpenAI (recently raising $110 billion) and Anthropic ($30 billion), the financial gravity of the AI revolution begins to warp the entire global economy.

But as Dylan Patel, Chief Analyst at SemiAnalysis, explains, this isn’t just a story about money. It’s a story about physical limits, “commitment issues” in the C-suite, and a tiny, specialized machine in the Netherlands that may ultimately decide the speed at which we reach Artificial General Intelligence (AGI).

The Gigawatt War: OpenAI vs. Anthropic

The most visible front of this war is being fought between the two leading AI labs: OpenAI and Anthropic. The divergence in their strategies is stark.

According to Patel, OpenAI has adopted a “YOLO” approach to compute. Early on, Sam Altman’s firm signed massive, long-term contracts for data centers and GPUs with Microsoft, Oracle, CoreWeave, and even SoftBank Energy. At the time, critics thought OpenAI was overextending, potentially risking bankruptcy if their revenue didn’t materialize.

Anthropic, led by Dario Amodei, took the opposite path. “Dario was very conservative,” Patel notes. “He didn’t want to go crazy on compute because if revenue inflected at a different rate… he didn’t want to go bankrupt.”

This conservatism may have backfired. As AI revenue began to “moon,” Anthropic found itself compute-constrained. To keep up with demand for models like Claude 3.5 Opus and Sonnet, Anthropic is now forced to acquire “last-minute” compute at a premium. While OpenAI locked in H100 GPUs at lower rates years ago, Anthropic is navigating a market where spot prices can reach $2.40 per hour for a chip that costs roughly $1.40 to operate.

“OpenAI has way more access to compute than Anthropic by the end of the year,” Patel says. The result? Anthropic is having to turn to “neoclouds” and secondary providers, paying higher margins and navigating data center delays just to stay in the race.

The Alchian-Allen Effect: Why High Costs Favor Great Models

One of the most counterintuitive aspects of the current AI boom is that older GPUs are actually increasing in value. Usually, tech depreciates faster than a new car driven off the lot. But in the age of scarcity, an H100 is worth more today than it was three years ago.

Patel points to the Alchian-Allen Effect to explain the market’s behavior. In economics, this theorem suggests that if you add a fixed cost (like a high price for compute) to two goods of different quality, people will shift their consumption toward the higher-quality good.

“If a Hopper GPU went from $2 to $3 an hour, the price differential between running a ‘medium’ model and a ‘great’ model shrinks,” Patel explains. “The calculus is: I’m paying all this money for the compute anyway; I might as well pay slightly more to make sure it’s the very best model.”

This effect is driving all the revenue toward the “frontier” models. It also means that companies that signed five-year contracts years ago have locked in a massive competitive advantage. They are running the world’s most valuable software on “cheap” historical silicon, while newcomers are paying “AGI-pilled” prices for the same hardware.

The Hidden Bottlenecks: From Turbines to Tin Droplets

While the world focuses on chips, the real bottlenecks are often more mundane—and much harder to solve.

Google, for instance, recently “woke up” to the scale of the challenge. After seeing their Gemini revenue skyrocket to a $5 billion annual run rate in just months, the company shifted into overdrive. They aren’t just buying chips; they are putting down deposits on power turbines for 2028 and 2029, securing power purchasing agreements (PPAs), and buying up land with existing electrical grid access.

But even if you have the land and the power, you still need the silicon. And the silicon supply chain is hitting a physical wall.

To understand the scale, Patel breaks down the “math of a gigawatt.” To build one gigawatt of data center capacity using Nvidia’s upcoming “Rubin” chips, you need:

  • 55,000 wafers of 3nm logic.
  • 170,000 wafers of DRAM memory.
  • Millions of “EUV passes.”

This leads to the ultimate gatekeeper of the AI era: ASML.

ASML: The 13.5-Nanometer Bottleneck

ASML, a Dutch company, is the only firm in the world capable of making Extreme Ultraviolet (EUV) lithography machines. These machines, which cost upwards of $400 million each, are the most complex devices ever built by humans.

The process is mind-boggling: inside the machine, a laser hits a droplet of molten tin twice. The first hit shapes the droplet; the second blasts it into a plasma that releases EUV light at a wavelength of 13.5 nanometers. This light is then reflected by the world’s flattest mirrors (made by Zeiss) to “print” circuits onto silicon.

“ASML can make about 70 of these tools this year,” Patel says. “Even under very aggressive expansion, they only get to a little over 100 by the end of the decade.”

If Sam Altman wants to bring a gigawatt of compute online every week by 2030, he would need to secure roughly 25% of the entire global output of EUV machines. This is a tall order when Apple, Intel, and the mobile phone industry also need those same machines to survive.

The “AGI-Pilled” Gap

The strangest part of this trillion-dollar scramble is the disconnect between the buyers and the builders.

The AI labs (OpenAI, Anthropic, DeepMind) are “AGI-pilled”—they believe they are years, not decades, away from models that can automate most human cognitive labor. If they are right, a GPU isn’t just a chip; it’s a “worker” that can repay its own cost in a matter of months.

However, the companies that actually make the machines—ASML, TSMC, and the memory vendors—remain skeptical. They have lived through decades of “boom and bust” cycles in the semiconductor industry. They are hesitant to build $20 billion fabs that might sit empty if the AI bubble bursts.

“Constantly, we’re told our numbers are way too high,” Patel says of his firm’s projections. “And then when they’re right, they say, ‘Okay, but your next year’s numbers are definitely too high.’”

This skepticism is the ultimate bottleneck. If the labs are right about the proximity of AGI, the world is drastically under-investing in the machinery required to build it. We are attempting to launch a digital god using a supply chain that is still worried about next year’s smartphone sales.

As we approach 2030, the race for AI won’t just be about who has the best code. It will be about who had the foresight to put a deposit on a tin-blasting laser in the Netherlands five years ago. In the trillion-dollar silicon war, the winners are those who realize that in a world of infinite software, the only thing that matters is the finite machine.


Based on “Under Secretary of War on Iran, Anthropic and the AI Battle Inside the Pentagon | The a16z Show” from a16z Watch the original video

Wartime Speed: Inside the Pentagon’s High-Stakes Race to Reclaim the AI Edge

In the quiet corridors of the Pentagon, a fundamental shift is occurring. For decades, the building operated at what insiders call “peacetime speed”—a lethargic rhythm of bureaucratic red tape, thousand-page requirement documents, and a handful of massive defense contractors. But according to the Department’s Chief Technology Officer, those days are over. The United States is currently facing the largest military buildup in human history, and the battlefield is no longer just about steel and gunpowder; it’s about silicon and software.

In a recent candid discussion on The a16z Show, the Under Secretary of Defense for Research and Engineering (often referred to as the Pentagon’s CTO) laid out a bracing vision for the future of American Dynamism. From the “holy cow” moments involving commercial AI models to the chilling reality of vendor lock-in during active military operations, the message was clear: The Pentagon must adopt Silicon Valley’s “wartime” urgency, or risk falling behind an adversary that isn’t waiting for a committee’s approval.

The Ghost of the “Last Supper”

To understand why the Pentagon is struggling to modernize, one must look back to the early 1990s—a period the Under Secretary identifies as the beginning of the “peacetime speed” era. Following the Cold War, the Department of Defense held a famous meeting known as “The Last Supper.” Military leaders told industry giants that the era of massive procurement and rapid innovation was over. They encouraged consolidation, telling companies to become dividend-payers and stock-buybackers rather than disruptors.

The result was a defense industrial base that shrank from dozens of competitors to just four or five “primes.” While the U.S. rested on its laurels, China began a massive military expansion in the mid-2000s. “We’ve outsourced a lot of our key domestic production—critical minerals, batteries, supply chain components,” the Under Secretary warned. “We looked up and realized we have a lot of catching up to do.”

From 14 Priorities to 6: The Radical Simplification

When the Under Secretary took office, he found a department spread too thin, buried under 14 “critical” priority areas that hadn’t changed in a decade. “Who can remember 14 things when you’re trying to motivate a workforce?” he asked.

In a move reminiscent of a Silicon Valley turnaround, he slashed the list to six. At the absolute top of that list: Applied AI.

The goal isn’t just to build “killer robots,” but to integrate AI as a “substrate”—a layer of intelligence that touches every facet of the military. He categorizes the AI mission into three distinct buckets:

  1. Enterprise Efficiency: Automating mundane administrative tasks for the department’s 3 million employees.
  2. Intelligence Augmentation: Using AI to sift through decades of siloed satellite imagery and data. “You take a human analyst and you increase their throughput by a thousand,” he explained.
  3. Warfighting & Logistics: Using models to solve complex physics problems, plan fuel-efficient troop movements in contested environments, and run millions of combat simulations.

The results of this focus have been immediate. In just 90 days, the number of Department personnel using some form of AI jumped from 80,000 to over 1.2 million.

The “Holy Cow” Moment: The Conflict of Constitutions

The most provocative part of the discussion centered on the Pentagon’s relationship with “Frontier” AI companies like Anthropic and OpenAI. In the past, software was a tool the military bought and owned. Today, AI is often provided as a service with “Terms of Service” and “Corporate Constitutions.”

The Under Secretary described a “holy cow” moment when reviewing contracts inherited from previous administrations. He discovered dozens of restrictions baked into the software. In theory, these commercial models—integrated into the most sensitive combat commands—could be “turned off” by the vendor if an operation violated a company’s internal ethical guidelines.

“The software’s ‘soul’ or ‘constitution’—which is not the U.S. Constitution—cannot be dictating our command and control environment,” he argued. “We cannot have a situation where a model could just stop in the middle of an operation and put lives at risk because a vendor decided it didn’t like how the tool was being used.”

This tension became visceral following the Maduro raid, a highly successful military operation. A senior executive from a primary AI vendor reportedly questioned whether their software was used during the raid, expressing hesitation about its involvement.

“When a company says, ‘Hey, was our software used there because we’re not sure we’d like that,’ a chill goes up your spine,” the Under Secretary said. “It’s like a stranger saying they saw your kid at school. You realize you are ‘single-threaded’ on a vendor whose values might not align with national security.”

The “Elon Model” and the Regulatory Moses

To fix this, the Pentagon is attempting to “move the debris” of bureaucracy. The Under Secretary is pushing for a shift from “Cost-Plus” contracts—which reward companies for taking longer and spending more—to “Firm-Fixed-Price” contracts.

He calls this the “Elon Model,” citing SpaceX’s success. Instead of a thousand-page Request for Proposal (RFP) that dictates exactly how to build a bolt, the Pentagon is moving toward simple requirements: I need a missile that goes this far, in this environment, with this payload. You figure out how to build it.

This shift is designed to open the door for startups. However, the Under Secretary issued a challenge to the “American Dynamism” movement: Innovation isn’t enough. Startups must learn the “old world” muscle of mass production. “The primes have an advantage in manufacturing at scale. Startups need to build the factories, the quality testing, and the supply chains to cross that chasm.”

A Call to Service

The Under Secretary’s journey to the Pentagon wasn’t a traditional one. A successful tech executive and immigrant whose first language was Arabic, he felt a pull toward public service after selling his company, Tellme Networks, to Microsoft.

He views the current moment as a “galvanizing” one for the tech industry, similar to the 2018 “Project Maven” controversy at Google. While that moment saw some employees protest military contracts, it also birthed a new generation of patriotic founders who want to ensure the U.S. maintains its technological edge.

“Our system doesn’t come for free,” he concluded. “We need builders, people who care, and people who are willing to sacrifice. We need patriots to come and do these things every now and again because industry has the best brains, but the country needs those brains to protect our way of life.”

As the AI battle inside the Pentagon intensifies, the goal remains singular: ensuring that when the next “wartime” moment arrives, the United States is moving at the speed of the future, not the pace of the past.


Based on “Louise Erdrich on Her New Story Collection and the Mystery of Writing” from New York Times Podcasts Watch the original video

The Art of the Army Crawl: Louise Erdrich on the Mystery of the Written Word

In the world of contemporary American literature, few figures loom as large or as prolific as Louise Erdrich. Since her 1984 debut Love Medicine, she has navigated nearly every corner of the literary landscape, from haunting poetry to National Book Award-winning fiction. Yet, despite her decades of acclaim and a Pulitzer Prize to her name, Erdrich remains a writer who views her own process not as a masterclass in control, but as a slow, often bewildering “army crawl” through the undergrowth of her own imagination.

In a recent conversation with Gilbert Cruz on the New York Times Book Review podcast, Erdrich peeled back the curtain on her latest collection, Python’s Kiss, and shared the idiosyncratic, often physical ways she brings her stories to life.

The Eight-Year Short Story

Perhaps the most striking revelation of Erdrich’s process is the sheer timeline of her shorter works. One story in the new collection, “The Love of My Days,” is a mere ten pages long, yet it took eight years to complete. For Erdrich, this isn’t a matter of writer’s block, but of a quiet, persistent gestation.

“I have a set of words just randomly pop up,” she explains. “I keep notebooks. I write something down, and sometimes something keeps going for a while, but it always fizzles out.” She describes these fragments as “nuggets” that occasionally lose their energy and are tucked away—sometimes for years—only to resurface when the writer herself has changed.

She describes the completion of these stories not as a sprint, but as a slow movement across a landscape. “It just crept along… army crawled along the forest floor somehow until it got to the end.” This patience isn’t a disciplined choice, she insists, but a byproduct of her own curiosity. She writes to amuse herself, returning to a draft only when a single line or paragraph hooks her back into the narrative’s world.

When the Story “Announces Itself”

One of the eternal questions for any multi-genre writer is how to decide whether an idea belongs in a poem, a children’s book, or a sprawling novel. For Erdrich, the decision is rarely hers to make.

“It announces itself,” she says. “I don’t have a way of changing it if it’s not going to be what it wants to be.” If a narrative refuses to stop, it becomes a novel; if it evolves toward a resolution within twenty pages, it remains a story. This lack of authorial ego is central to her philosophy. She views writing as a process that happens outside of her conscious will—a necessity for someone who describes herself as “not a very in control person” in her daily life.

Writing, for Erdrich, is her primary mode of processing the world. It began in the fifth grade as a way to handle being “overwhelmed” by the ending of the original Planet of the Apes. Growing up in North Dakota—a state she describes as a “national sacrifice area” due to its hundreds of underground ICBM silos—the film’s nuclear subtext resonated with a terrifying clarity. That early anxiety fueled a lifelong habit of keeping diaries, which she admits began as “cries of woe” before evolving into the rich repositories of detail that now inform her fiction.

The Physicality of the Page

While many imagine a writer’s work as a purely intellectual exercise, Erdrich’s method is surprisingly physical. When she finds herself “off” or stuck, she literally takes to the floor.

“I have to throw everything on the floor—all these notebooks, these crazy notes—and I have to crawl through them and put them in order,” she says. This “physicalizing of the intellectual act” allows her to see the architecture of a story in a way a computer screen cannot provide. It is a humble image: one of our greatest living writers on her hands and knees, sorting through scraps of paper to find the missing beginning of a tale.

This humility extends to her view of her own successes. When The Night Watchman—a novel based on her grandfather’s fight against Native American termination policies—won the Pulitzer Prize, Erdrich famously remarked that the award didn’t go to her, but to the book and her grandfather. As she has aged, she says her ego has retreated, allowing her to embrace a “simplicity” in her storytelling that she lacked in her earlier, more self-conscious years.

A Life Built on Letters

Erdrich’s literary DNA is rooted in a family of prolific letter writers. Her father would write long, exaggerated letters about the mundane—turning her mother’s canning hobby into a tall tale about canning toads and slugs. He even undertook a project to write a limerick for every single town in North Dakota.

This tradition of written connection shaped Erdrich’s understanding of character and voice. Her grandfather’s letters from his time as tribal chairman for the Turtle Mountain Chippewa became the foundational “personality” for the protagonist of The Night Watchman. For Erdrich, the archive of family letters is more than just a memory; it is a “treasure” that proves the power of the written word to preserve a soul.

The Evolution of the Indigenous Voice

Beyond her own writing, Erdrich is a fierce advocate for other authors. Since 2001, she has owned Birchbark Books, an independent bookstore in Minneapolis. In the twenty-five years since its opening, she has witnessed a tectonic shift in the literary landscape.

“When I first started, there was only one set of shelves that had any Native titles on them,” she recalls. “And they were mostly by non-Native authors about Native people.” Today, the store could be filled entirely with Indigenous voices. She points to the “explosion” of Native writing as a fulfillment of a prophecy: that once writers from different tribes began to tell their unique stories, the world would realize the vast, diverse richness of the Indigenous experience.

When asked for recommendations, she points to James Welch’s 1974 novel Winter in the Blood, calling it a “beautiful, bleak, short, funny, intensely personal book” that she returns to repeatedly.

The Mystery of the “Other World”

Despite her immense body of work, Erdrich maintains that the act of creation remains a mystery to her—and she prefers it that way. She often looks back at her own handwriting in old notebooks and feels a sense of detachment, as if the words were written by someone else.

“I’m, in my real life, a boring person,” she claims with a laugh. “I don’t really have a magical personality, but the stories write themselves.” She believes that while research and thought are essential, there is a piece of art that belongs to “some other world, some other entity.”

As the interview concluded, Erdrich offered a glimpse into her current reading habits, which range from the ancient and meditative (Sei Shōnagon’s The Pillow Book for insomnia) to the provocatively modern. On her shelf currently sits a book titled Make the Golf Course a Public Sex Forest—a compendium of essays regarding a local Minneapolis land-use controversy.

It is this blend of the high-brow and the earthy, the historical and the speculative, the Pulitzer-winning and the “army-crawling” that defines Louise Erdrich. She remains a writer who is less interested in being a “literary figure” than she is in following the mystery of the next sentence, wherever it might lead her on the forest floor.


한국어

“The deep mystery physicists call “the problem of time” | Jim Al-Khalili: Full Interview”Big Think 기반 기사 원본 영상 보기

[심층 리포트] 시간은 흐르는가, 아니면 멈춰 있는가? 물리학이 마주한 거대한 수수께끼 ‘시간의 문제’

우리는 시간을 공기처럼 당연하게 여기며 살아갑니다. 어제는 지나갔고, 오늘은 흐르고 있으며, 내일은 아직 오지 않았다는 감각은 우리의 존재 방식 그 자체입니다. 하지만 현대 물리학의 정점에 서 있는 학자들에게 시간은 그 무엇보다 다루기 까다롭고 당혹스러운 대상입니다.

영국 서리 대학교(University of Surrey)의 이론 물리학 명예 교수이자 저명한 과학 커뮤니케이터인 짐 알칼릴리(Jim Al-Khalili)는 이를 **‘시간의 문제(The Problem of Time)‘**라고 부릅니다. 그는 자신의 저서 『온 타임(On Time)』을 통해 우리가 알고 있는 시간의 개념을 완전히 뒤흔드는 네 가지 핵심 질문을 던집니다.


1. 우리가 느끼는 시간과 물리적 시간의 괴리

우리는 흔히 시간이 ‘흐른다’고 표현합니다. 하지만 알칼릴리 교수는 우리가 시간을 객관적으로 관찰하는 데 근본적인 한계가 있다고 지적합니다. 우리는 시간이라는 시스템 안에 완전히 매몰되어(Embedded) 있기 때문에, 시스템 밖으로 나가 시간을 객관적으로 바라볼 수 없기 때문입니다.

여기서 물리학자와 철학자들은 **‘현상적 시간(Manifest Time)‘**과 **‘물리적 시간(Physical Time)‘**을 구분합니다.

  • 현상적 시간: 우리의 뇌가 지각하는 심리적 시간입니다. 치과 대기실에서의 30분은 영겁처럼 느껴지지만, 즐거운 파티에서의 30분은 눈 깜짝할 새 지나갑니다. 나이가 들수록 시간이 빨리 흐른다고 느끼는 것 역시 새로운 경험의 밀도가 낮아지며 발생하는 심리적 현상입니다.
  • 물리적 시간: 물리학 법칙 속에 존재하는 시간입니다. 놀랍게도 물리학 방정식 속에서 시간($t$)은 그저 하나의 좌표나 숫자에 불과합니다. 방정식 안에는 시간이 ‘흐른다’는 개념도, ‘과거에서 미래로 향한다’는 특별한 방향성도 존재하지 않습니다.

2. 아인슈타인이 부순 ‘절대적 시계’의 환상

아이작 뉴턴(Isaac Newton)은 우주 어디에서나 동일하게 똑딱거리는 ‘절대적 시계’가 존재한다고 믿었습니다. 하지만 알베르트 아인슈타인(Albert Einstein)은 이 믿음을 산산조각 냈습니다.

**특수 상대성 이론(Special Theory of Relativity)**에 따르면, 빛의 속도는 누가 측정해도 일정해야 합니다. 이 원칙을 지키기 위해선 시간과 공간이 관찰자의 속도에 따라 변해야만 합니다. 이것이 바로 시간 지연(Time Dilation) 현상입니다. 실제로 대기권 상층부에서 생성된 뮤온(Muon) 입자는 수명이 매우 짧아 지표면에 도달할 수 없어야 하지만, 빛에 가까운 속도로 이동하며 시간이 느리게 흐른 덕분에 지상까지 도달합니다.

**일반 상대성 이론(General Theory of Relativity)**은 한 발 더 나아갑니다. 중력이 강할수록 시간은 더 느리게 흐릅니다. 이는 이론에 그치지 않습니다. 우리가 매일 사용하는 GPS 위성은 지구 표면보다 중력이 약한 곳에 있기 때문에 지상보다 시간이 미세하게 빨리 흐릅니다. 이를 보정하지 않으면 내비게이션의 오차는 걷잡을 수 없이 커집니다. 즉, 당신의 머리는 당신의 발보다 아주 미세하게 더 빨리 늙고 있습니다.


3. ‘블록 우주’와 자유 의지의 역설

아인슈타인의 이론을 시각화하면 시간은 공간의 세 차원과 결합한 **‘4차원 시공간(4-dimensional Spacetime)‘**이 됩니다. 물리학자들은 이를 ‘블록 우주(Block Universe)’ 모델로 설명합니다.

이 모델에서 과거, 현재, 미래는 이미 결정된 채 하나의 거대한 덩어리로 존재합니다. 이를 **영원주의(Eternalism)**라고 합니다. 우리가 ‘지금’이라고 느끼는 순간은 영화 필름의 한 프레임에 불과하며, 전체 필름은 이미 완성되어 있는 것과 같습니다.

그렇다면 인간의 ‘자유 의지’는 어디에 있을까요? 모든 미래가 이미 블록 우주 안에 존재한다면, 나의 선택은 의미가 없는 것일까요? 알칼릴리 교수는 이에 대해 **양립 가능론(Compatibilism)**이라는 철학적 태도를 취합니다. 설령 우주가 결정론적(Deterministic)일지라도, 우리는 우주 내부의 관찰자로서 미래를 결코 예측할 수 없기에 우리의 선택은 여전히 자유로운 선택으로서 의미를 갖는다는 것입니다.


4. 시간은 근본적인 것인가, 아니면 ‘창발’된 것인가?

물리학의 최대 과제는 거시 세계를 설명하는 상대성 이론과 미시 세계를 설명하는 양자 역학을 통합하는 것입니다. 이 과정에서 등장한 **휠러-드윗 방정식(Wheeler-DeWitt Equation)**은 충격적인 결과를 보여줍니다. 우주 전체를 설명하는 이 방정식에는 ‘시간’ 변수가 아예 존재하지 않습니다.

이에 따라 일부 물리학자들은 시간이 **‘창발적 속성(Emergent Property)‘**일 수 있다고 주장합니다.

  • 약한 창발(Weak Emergence): 물 분자 하나에는 ‘젖음’이라는 속성이 없지만, 수조 개의 분자가 모이면 ‘젖음’이 나타나는 것과 같습니다.
  • 강한 창발(Strong Emergence): 개별 뉴런의 활동에서는 찾아볼 수 없는 ‘의식’이 뇌 전체에서 나타나는 것과 같습니다.

시간 역시 양자 수준의 더 근본적인 무언가에서 비롯된 결과물일 뿐, 우주의 근본 원리는 아닐 수도 있다는 뜻입니다.


5. 시간의 화살: 엔트로피와 양자 얽힘

물리학 법칙이 시간 대칭적(Time-symmetric)임에도 불구하고, 왜 현실에서는 시간이 오직 한 방향으로만 흐를까요? 달걀이 깨질 수는 있어도 스스로 붙을 수는 없는 이유는 무엇일까요?

전통적인 답변은 열역학 제2법칙(Second Law of Thermodynamics), 즉 **엔트로피(Entropy)**의 증가입니다. 고립된 시스템에서 무질서도는 항상 증가하며, 이것이 시간의 방향성을 결정한다는 논리입니다.

하지만 알칼릴리 교수는 여기서 한 걸음 더 나아가 **양자 얽힘(Quantum Entanglement)**과 **결어긋남(Decoherence)**에서 그 해답을 찾습니다. 양자 시스템이 주변 환경과 상호작용하며 정보가 흩어지는 과정은 근본적으로 되돌릴 수 없는(Irreversible) 과정입니다. 그는 시간의 화살이 우주의 근본적인 구조, 즉 양자 수준의 상호작용 속에 이미 각인되어(Baked into reality) 있다고 주장합니다.


결론: 시간은 여전히 실재한다

짐 알칼릴리 교수는 시간이 환상이라는 일부의 주장에 반박하며 인터뷰를 마무리합니다. 시간이 무언가 더 깊은 원리에서 창발된 것이라 할지라도, 시간의 방향성이 실재하고 그 흐름이 물리적 현상으로 증명되는 한 시간은 ‘실재하는 것’입니다.

“시간의 화살이 실재한다면, 시간 그 자체도 실재해야 합니다. 체셔 고양이의 미소(방향성)만 존재할 수는 없습니다. 미소를 짓는 고양이(시간)가 반드시 필요하죠.”

우주의 탄생인 빅뱅(Big Bang)과 함께 시작된 이 위대한 여정은 여전히 수많은 수수께끼를 품고 있습니다. 하지만 물리학이 이 ‘시간의 문제’를 하나씩 풀어나갈 때마다, 우리는 우리가 누구이며 어디로 가고 있는지에 대한 본질적인 답에 조금씩 더 가까워지고 있습니다.


“667. Here’s Why You Are Constantly Fighting Off Scammers | Freakonomics Radio”Freakonomics Radio Network 기반 기사 원본 영상 보기

”돼지를 살찌워 도살하라” : 당신의 일상을 파고드는 지능형 사기, ‘스캠 산업’의 민낯

지난 1월, 캄보디아에서 활동하던 중국계 기업가 천시(Chen Xi)가 체포되어 중국으로 송환되었습니다. 그는 캄보디아에서 손꼽히는 자산가이자 정부 고문까지 지낸 엘리트 비즈니스맨이었지만, 실상은 달랐습니다. 미국 정부는 그의 지주회사를 ‘초국가적 범죄 조직’으로 규정하고 사기 및 돈세탁 혐의로 기소하며 150억 달러(약 20조 원) 규모의 암호화폐를 압수했습니다.

이 막대한 돈은 어디서 나왔을까요? 그 배후에는 이른바 **‘돼지 도살(Pig Butchering)‘**이라 불리는 잔혹하고 체계적인 온라인 사기 수법이 있었습니다. 마치 돼지를 도살하기 전 살을 찌우듯, 피해자와 수개월 혹은 수년간 신뢰를 쌓은 뒤 적절한 시기에 전 재산을 가로채는 방식입니다.

오늘날 사기는 단순히 개인의 일탈이 아닙니다. 연간 수십조 원이 오가는 거대하고 정교한 ‘글로벌 산업’으로 진화했습니다. 프리코노믹스 라디오(Freakonomics Radio)가 파헤친 현대 사기 산업의 실태와 그 속에서 우리가 살아남는 법을 정리했습니다.


1. 사기는 어떻게 ‘산업’이 되었는가?

전문가들은 사기를 “매우 복잡하고, 끊임없이 진화하며, 치열하게 경쟁하는 산업”이라고 정의합니다. 미국 검찰에 따르면 캄보디아의 사이버 범죄 수익은 연간 약 190억 달러에 달하며, 이는 캄보디아 국내총생산(GDP)의 절반에 육박하는 수치입니다.

이 산업은 합법적인 기업의 구조를 그대로 본뜨고 있습니다.

  • 분업화된 조직: 마케팅 팀, IT 개발 팀, 심지어 인사(HR) 부서와 법무팀까지 갖추고 있습니다.
  • A/B 테스트: 어떤 메시지가 더 높은 응답률을 기록하는지 끊임없이 데이터를 분석하고 문구를 수정합니다.
  • 인적 자원 관리: 사기꾼들에게 할당량(Quota)을 부여하고 실적을 압박합니다. 충격적인 사실은 이들 중 상당수가 인신매매를 통해 끌려온 ‘강제 노동자’라는 점입니다.

2. 왜 우리는 똑똑하면서도 사기에 당하는가?

미국 미네소타 대학교의 노년학자 마티 드 리마(Marty De Lima) 교수는 사기가 노인들만의 문제라는 통념을 정면으로 반박합니다.

세대별 타겟팅의 차이

  • 청년층: 가짜 구인 광고나 온라인 쇼핑 사기에 주로 노출됩니다.
  • 중장년층: 통계적으로 사기 피해 신고율이 가장 높은 집단입니다.
  • 고령층: 기술 지원(Tech Support) 사기나 복권 사기의 표적이 됩니다. 고령층은 자산 규모가 크기 때문에 1인당 피해 액수가 다른 세대보다 훨씬 큽니다.

인지 시스템의 허점: 시스템 1 vs 시스템 2

인간의 뇌는 두 가지 방식으로 정보를 처리합니다.

  • 시스템 1: 빠르고 직관적이며 감정적인 판단 (본능)
  • 시스템 2: 느리고 논리적이며 분석적인 판단 (이성)

사기꾼들의 목표는 우리의 ‘시스템 2’를 끄고 ‘시스템 1’로만 행동하게 만드는 것입니다. 이를 위해 그들은 ‘정서적 각성(Emotional Arousal)’ 상태를 유도합니다.

  • 긍정적 자극: “당신만 아는 특별한 투자 기회”, “운명적인 사랑” (탐욕과 기대감)
  • 부정적 자극: “지금 당장 조치하지 않으면 체포됩니다”, “계좌가 동결되었습니다” (공포와 시급성)

심한 공포나 흥분 상태에 빠지면, 평소 논리적이던 사람도 “은행에서 전화를 준다니 일단 믿자”라는 식의 지름길(Heuristic)을 선택하게 됩니다.

3. AI라는 날개를 단 사기꾼들

과거의 사기는 맞춤법이 틀리거나 말투가 어색해 알아차리기 쉬웠습니다. 하지만 인공지능(AI)의 등장이 게임의 규칙을 바꿨습니다.

  • 언어 장벽의 붕괴: 챗GPT(ChatGPT) 같은 대규모 언어 모델을 사용해 완벽한 문법과 자연스러운 말투로 피해자에게 접근합니다.
  • 딥페이크와 음성 복제: 사랑하는 가족의 목소리를 복제해 납치 상황을 연출하거나, 영상 통화 중에 얼굴을 바꿔치기해 신뢰를 얻습니다.
  • 자동화된 대규모 공격: 과거에는 사기꾼 한 명이 수백 명에게 직접 메시지를 보내야 했지만, 이제는 AI가 수억 명에게 개인화된 메시지를 보내고 초기 응답까지 대신 처리합니다.

4. 거대 플랫폼의 ‘침묵하는 파트너십’

기사는 사기 산업의 번창 뒤에 거대 IT 기업들의 책임도 적지 않음을 지적합니다. 메타(Meta, 페이스북과 인스타그램의 모회사)의 내부 문건에 따르면, 매출의 약 10%가 사기 광고나 금지 품목 광고에서 발생한다는 분석도 있습니다.

드 리마 교수는 “기술적으로 사기 계정을 차단하는 것은 가능하지만, 플랫폼 기업들이 비용 대비 수익을 따지며 적극적으로 나서지 않는 것이 문제”라고 꼬집습니다. 피해자들은 사기꾼을 원망한 뒤, 결국 “내가 바보 같아서 당했다”며 자기 자신을 탓합니다. 하지만 이는 개인의 실수를 넘어 범죄를 방치하는 거대 구조의 문제이기도 합니다.

5. 사회적 신뢰의 붕괴, 어떻게 대처할 것인가?

사기는 단순히 돈을 잃는 것을 넘어 ‘배신 트라우마(Betrayal Trauma)‘를 남깁니다. 이는 인간관계와 사회 시스템에 대한 근본적인 불신으로 이어지며, 심한 경우 피해자를 자살로 몰아넣기도 합니다.

전문가들이 제안하는 **‘스캠 방지 체크리스트’**는 다음과 같습니다.

  1. 모든 예기치 않은 연락을 의심하라: 이메일, 문자, 전화 등 먼저 찾아온 연락은 일단 ‘가짜’일 가능성을 염두에 두어야 합니다.
  2. 독립적인 경로로 확인하라: 메시지에 포함된 링크나 전화번호를 절대 클릭하지 마세요. 공식 홈페이지를 직접 검색해 접속하거나, 공식 고객센터 번호로 직접 전화를 걸어 확인해야 합니다.
  3. 시간적 압박에 저항하라: “지금 당장”을 강조한다면 99% 사기입니다. 이성적인 판단을 내릴 시간을 갖지 못하게 하려는 수법입니다.
  4. 개인정보의 신화에서 벗어나라: 사기꾼이 내 이름, 주소, 주민번호 뒷자리를 알고 있다고 해서 그가 공공기관 직원인 것은 아닙니다. 이미 우리의 개인정보는 어둠의 경로에서 ‘최고가 입찰자’에게 팔려 나가 있습니다.

결론: 보이지 않는 적과의 전쟁

사기 산업은 우리가 서로를 믿고 소통하는 방식을 파괴하고 있습니다. “모든 전화를 의심해야 하고, 모든 이메일을 걸러야 하는 세상”은 우리 사회의 소중한 자산인 ‘사회적 신뢰’를 갉아먹습니다.

이제 사기 방지는 개인의 주의력을 넘어 국가적, 국제적 공조가 필요한 시점입니다. 기술이 사기를 돕는 만큼, 사기를 잡아내는 AI 기술과 강력한 플랫폼 규제가 뒷받침되어야 합니다. 무엇보다 중요한 것은 사기 피해자를 비난하지 않는 문화입니다. 사기는 ‘멍청해서 당하는 것’이 아니라, ‘인간의 본성을 이용하도록 설계된 정교한 덫’에 걸린 것이기 때문입니다.


“Dylan Patel — The Single Biggest Bottleneck to Scaling AI Compute”Dwarkesh Patel 기반 기사 원본 영상 보기

[인사이트] 600조 원의 베팅, 하지만 ‘진짜 병목’은 따로 있다: 딜런 파텔이 진단하는 AI 컴퓨팅의 미래

서론: 조 단위 달러가 투입되는 AI 전쟁의 실체

최근 마이크로소프트, 구글, 메타, 아마존 등 이른바 ‘빅4’ 테크 기업들이 올해 쏟아붓겠다고 발표한 자본 지출(CapEx) 총액은 무려 6,000억 달러(약 800조 원)에 달합니다. 여기에 오픈AI(OpenAI)와 앤스로픽(Anthropic) 같은 AI 연구소들이 조달한 천문학적인 투자금까지 더하면, 바야흐로 ‘조 단위 달러’의 시대가 열렸습니다.

하지만 돈이 충분하다고 해서 AI의 성능이 무한정 올라갈 수 있을까요? 반도체 전문 분석 기관 ‘세미아날리시스(SemiAnalysis)‘의 CEO 딜런 파텔(Dylan Patel)은 단호하게 “아니오”라고 답합니다. 그는 최근 드와르케시 파텔(Dwarkesh Patel)과의 대담에서, 돈보다 더 무서운 ‘물리적 한계’와 ‘공급망의 병목 현상’이 AI 확장의 발목을 잡고 있다고 경고합니다.


1. 돈은 냈지만 전기는 아직이다: 6,000억 달러의 행방

빅테크 기업들이 발표한 6,000억 달러라는 거액은 당장 오늘 사용할 칩을 사는 데 쓰이는 것이 아닙니다. 딜런 파텔에 따르면, 이 비용의 상당 부분은 **‘미래를 위한 자리 선점’**에 투입되고 있습니다.

  • 인프라 선점: 구글이 지출하는 수천억 달러 중 상당액은 2028년과 2029년에 가동될 발전기 터빈 예약금, 2027년 완공될 데이터 센터 건설비, 그리고 장기 전력 구매 계약(PPA)의 계약금으로 쓰입니다.
  • 추론(Inference)의 늪: 앤스로픽과 같은 기업이 매출을 60억 달러 늘릴 때마다, 이를 뒷받침하기 위해 필요한 ‘추론용 컴퓨팅’ 비용만 약 40억 달러가 듭니다. 현재의 기술 수준에서 매출 성장은 곧 기하급수적인 컴퓨팅 용량 증설을 의미하며, 이는 매년 수 기가와트(GW)급의 데이터 센터를 새로 지어야 한다는 뜻입니다.

2. ‘공격적 베팅’의 오픈AI vs ‘원칙적 보수주의’의 앤스로픽

딜런 파텔은 두 대표적인 AI 랩의 전략 차이가 향후 승패를 가를 수 있다고 분석합니다.

  • 오픈AI의 ‘YOLO’ 전략: 샘 올트먼은 돈이 있든 없든 일단 대규모 컴퓨팅 계약을 체결하는 공격적인 행보를 보였습니다. 결과적으로 오라클(Oracle), 코어위브(CoreWeave), 심지어 데이터 센터 건설 경험이 없는 소프트뱅크 에너지와도 손을 잡으며 압도적인 컴퓨팅 자원을 확보했습니다.
  • 앤스로픽의 신중론: 반면 다리오 아모데이(Dario Amodei) CEO는 파산을 경계하며 보수적인 확장을 선택했습니다. 하지만 이는 결과적으로 ‘컴퓨팅 부족’이라는 부메랑으로 돌아왔습니다. 뒤늦게 용량을 확보하려다 보니, 미리 계약한 업체보다 훨씬 비싼 ‘스팟(Spot)’ 가격을 지불하거나 수익의 절반을 떼어주는 불리한 조건으로 컴퓨팅을 빌려 써야 하는 상황에 처했습니다.

3. 반도체 경제학의 역설: “중고 GPU가 더 비싸지는 세상”

일반적으로 기술 제품은 시간이 지나면 가치가 하락(감가상각)합니다. 하지만 AI 칩, 특히 엔비디아(NVIDIA)의 H100은 정반대의 길을 걷고 있습니다.

  • 모델 효율성의 마법: 모델 아키텍처가 발전하면서(예: GPT-4에서 GPT-5.4로), 같은 H100 칩 하나로 뽑아낼 수 있는 토큰(Token)의 가치가 훨씬 커졌습니다. 즉, 칩의 성능은 그대로지만 그 칩이 만들어내는 ‘지능의 가치’가 높아지면서 칩의 몸값도 올라가는 현상이 발생한 것입니다.
  • 긴 수명 주기: 비관론자들은 GPU 수명이 2~3년에 불과하다고 주장하지만, 딜런은 AI 수요가 폭발하는 현 상황에서 5년 전 모델인 A100조차도 여전히 ‘귀하신 몸’ 대접을 받으며 현역으로 뛰고 있다고 지적합니다.

4. 최종 보스, ASML과 EUV: 인류가 직면한 가장 좁은 병목

딜런 파텔이 꼽은 AI 확장의 가장 근본적이고 치명적인 병목은 전력도, 데이터 센터도 아닌 **‘노광 장비(Lithography)‘**입니다.

  • EUV의 한계: 세계에서 가장 복잡한 기계인 ASML의 EUV(극자외선) 노광 장비는 1년에 고작 70~80대 정도만 생산됩니다. 2030년이 되어도 연간 생산량은 100대를 겨우 넘길 전망입니다.
  • 기가와트당 장비 수: 1기가와트(GW) 규모의 데이터 센터를 최신 엔비디아 루빈(Rubin) 칩으로 채우려면 약 3.5대의 EUV 장비가 필요합니다. 샘 올트먼이 꿈꾸는 ‘매주 1기가와트 증설’을 실현하려면 전 세계 EUV 생산량의 25% 이상을 혼자 독점해야 한다는 계산이 나옵니다.
  • 공급망의 경직성: ASML 장비에 들어가는 광원(Cymer), 렌즈(Carl Zeiss) 등 핵심 부품 공급망은 극도로 복잡하여 단기간에 생산량을 두 배, 세 배로 늘리는 것이 물리적으로 불가능합니다. 결국 AI의 발전 속도는 ASML이 기계를 얼마나 빨리 만들어내느냐에 달려 있는 셈입니다.

5. 정보의 비대칭: 구글은 왜 ‘황금알을 낳는 거위’를 팔았나?

흥미로운 에피소드 중 하나는 구글의 실수입니다. 구글은 자체 AI 칩인 TPU를 보유한 강력한 플레이어지만, 초기에는 AI 수요를 과소평가했습니다.

과거 구글은 앤스로픽에 막대한 양의 TPU 용량을 판매했습니다. 당시 구글 클라우드 경영진은 이를 단순한 매출로 생각했지만, 얼마 지나지 않아 구글 딥마인드(DeepMind) 팀은 자신들이 쓸 컴퓨팅 자원이 부족하다는 사실을 깨닫고 경악했습니다. 앤스로픽이 구글보다 먼저 ‘컴퓨팅의 가치’를 알아보고 자원을 선점해버린 것입니다. 이후 구글은 급하게 태세를 전환하여 전력 확보와 칩 생산에 사활을 걸기 시작했습니다.


결론: 지능의 한계는 결국 제조의 한계다

딜런 파텔의 분석은 우리에게 중요한 시사점을 던집니다. AI는 더 이상 소프트웨어나 알고리즘의 영역에만 머물지 않습니다. 그것은 이제 **‘얼마나 많은 원자를 정교하게 배열하여 칩을 만들고, 얼마나 많은 전자를 이동시킬 수 있는가’**라는 거대한 물리적 전쟁터가 되었습니다.

조 단위 달러의 자본이 AI로 몰려들고 있지만, 그 돈이 통과해야 할 문은 ASML의 좁은 생산 라인과 한정된 전력망뿐입니다. 앞으로의 5년, AI 승자는 가장 똑똑한 알고리즘을 가진 자가 아니라, 물리적 공급망의 병목을 가장 먼저 뚫어내는 자가 될 것입니다.


“Under Secretary of War on Iran, Anthropic and the AI Battle Inside the Pentagon | The a16z Show”a16z 기반 기사 원본 영상 보기

”펜타곤의 영혼을 기술 기업에 맡길 수 없다” 국방부 CTO가 밝히는 AI 전쟁의 내막

미국 국방부(DoD)는 현재 역사상 유례없는 군사적 긴장감 속에 놓여 있습니다. 인공지능(AI)이 모든 산업의 근간을 바꾸는 ‘범용 인공지능(AGI)’ 시대로 향하는 지금, 세계 최강의 군사력을 자랑하는 펜타곤 내부에서는 소리 없는 전쟁이 벌어지고 있습니다. 바로 **‘전시 속도(Wartime Speed)‘**로의 체질 개선입니다.

실리콘밸리의 성공한 기업가에서 미 국방부의 기술 수장(CTO)으로 변신한 **샤리프 마락비(Shereef Marakby)**는 최근 ‘a16z 쇼’에 출연하여, 펜타곤이 직면한 AI 도입의 장벽과 앤스로픽(Anthropic) 등 민간 모델 사용을 둘러싼 긴박한 비하인드 스토리를 가감 없이 공개했습니다.


1. “평시의 속도는 끝났다” : 펜타곤에 닥친 위기감

마락비 CTO는 부임 직후 국방부의 혁신 속도가 민간에 비해 턱없이 뒤처져 있다는 사실을 직시했습니다. 그는 현재의 상황을 **“역사상 최대 규모의 군비 증


“Louise Erdrich on Her New Story Collection and the Mystery of Writing”New York Times Podcasts 기반 기사 원본 영상 보기

[인터뷰] 루이즈 어드리크, 바닥을 기며 길어 올린 ‘글쓰기의 신비’

풀리처상과 전미도서상을 석권한 미국 문학의 거장, 루이즈 어드리크(Louise Erdrich)가 신작 단편 소설집 『파이썬의 키스(Python’s Kiss)』로 돌아왔다. 1984년 데뷔작 『사랑의 약(Love Medicine)』 이후 소설, 시, 논픽션, 아동문학까지 장르를 넘나들며 20여 권의 책을 펴낸 그녀는 명실상부 현대 미국 문학을 대표하는 원주민 작가다. 뉴욕타임스(NYT) 팟캐스트에 출연한 그녀는 20년에 걸쳐 완성된 이번 소설집을 통해 글쓰기라는 고통스럽고도 신비로운 과정에 대해 깊이 있는 통찰을 전했다.


10페이지를 위한 8년의 시간, ‘포복’하는 글쓰기

어드리크의 창작 과정은 효율성이나 속도와는 거리가 멀다. 이번 소설집에 수록된 「내 생애의 사랑(The Love of My Days)」은 단 10페이지 분량이지만, 완성까지 무려 8년이 걸렸다. 그녀는 이 과정을 **‘숲 바닥을 기어가는 군인의 포복(Army crawl)‘**에 비유한다.

“인내심이라기보다는 즐거움에 가깝습니다. 쓰다가 막히면 다른 일을 하러 떠나고, 그러다 문득 보상처럼 문장 하나, 문단 하나가 찾아오면 다시 그 이야기에 낚여버리죠.”

그녀는 때로 수많은 노트와 종이 조각을 방바닥에 펼쳐놓고 실제로 바닥을 기어 다니며 순서를 맞춘다고 고백한다. 머릿속의 추상적인 아이디어를 물리적인 종이 뭉치로 다루며 몸으로 부딪히는 과정, 그것이 어드리크가 이야기를 빚어내는 방식이다.

소설은 스스로를 선언한다: 작가의 의지를 넘어선 영역

많은 작가가 장르를 미리 설정하고 집필을 시작하지만, 어드리크에게 이야기는 ‘스스로 정체성을 드러내는 존재’다.

“이야기가 무엇이 되고 싶은지 제가 바꿀 방법은 없습니다. 소설(Novel)이 될 운명이라면 멈추지 않고 계속 나아갈 것이고, 단편(Short Story)이라면 15~20페이지 안에서 결말을 향해 진화하죠. 제 의지나 통제 밖에서 일어나는 과정입니다.”

그녀는 자신이 삶의 모든 것을 통제하는 사람이 아니기에, 이러한 예술적 불확실성을 오히려 편안하게 받아들인다고 말한다. 특히 자신의 경험을 처리하는 유일한 방법이 ‘글쓰기’였음을 밝히며, 초등학생 때부터 써온 일기가 그 뿌리가 되었음을 언급했다.

핵 미사일 기지 옆에서 자란 소녀의 공포와 기록

어드리크의 문학적 감수성은 어린 시절 노스다코타에서의 경험과 맞닿아 있다. 그녀가 초등학교 5학년 때 일기를 쓰기 시작한 결정적 계기는 영화 <혹성탈출(Planet of the Apes)>의 충격적인 결말이었다.

“그 결말을 보고 핵전쟁이 일어났음을 직감했습니다. 당시 제가 살던 노스다코타는 세계 3위의 핵 강국이라 불릴 만큼 수백 개의 대륙간탄도미사일(ICBM) 기지로 둘러싸여 있었죠. 우리는 그곳을 ‘국가적 희생 지역’이라 부르며 기이한 자부심을 느끼기도 했습니다.”

전쟁이 터지면 가장 먼저 파괴될 곳에서 자란 소녀에게 세상은 압도적인 공포였고, 일기는 그 혼란스러운 세계를 기록하고 이해하려는 절박한 시도였다.

거장들의 유산: 체호프에서 로런 그로프까지

단편 소설의 거장으로 불리는 그녀가 가장 존경하는 작가는 단연 **안톤 체호프(Anton Chekhov)**다. 그녀는 영감이 필요할 때마다 체호프의 작품으로 돌아간다. 또한 현대 작가 중에서는 조지 손더스(George Saunders)와 로런 그로프(Lauren Groff)를 언급했다.

특히 로런 그로프의 단편 「브롤러(Brawler)」를 읽고 눈물을 흘렸던 경험을 공유하며, 소설 속 ‘어린이 영웅’이 가족을 구하려 애쓰는 모습에서 깊은 정서적 타격을 입었다고 말했다. “자신이 쓴 글에 스스로 눈물을 흘릴 때가 있는데, 그건 예술이 작가의 통제를 벗어나 독자적인 생명력을 얻었음을 의미합니다. 작가인 저조차 제가 썼다는 사실이 믿기지 않을 때가 있죠. 그 신비로움이 글을 계속 쓰게 만듭니다.”

원주민 문학의 폭발적 성장과 ‘버치바크 북스’

어드리크는 미니애폴리스에서 ‘버치바크 북스(Birchbark Books)‘라는 독립 서점을 25년째 운영하고 있다. 아마존의 공세 속에서도 살아남은 이 서점은 원주민 문학의 지형 변화를 목격한 산증인이다.

“처음 서점을 열었을 때는 원주민 관련 서가가 단 하나뿐이었고, 그마저도 비원주민 작가가 쓴 책들이 대부분이었습니다. 하지만 지금은 서점 전체를 원주민 작가의 책으로 채울 수 있을 만큼 원주민 문학이 폭발적으로 성장했습니다.”

그녀는 제임스 웰치(James Welch)의 『윈터 인 더 블러드(Winter in the Blood)』를 강력하게 추천하며, 원주민 작가들이 각자의 부족적 배경을 바탕으로 써 내려가는 독창적인 이야기들이 현대 문학에 새로운 활력을 불어넣고 있다고 강조했다.

읽기의 즐거움: 불면증을 위한 고전부터 파격적인 신간까지

인터뷰 말미, 그녀는 자신의 독서 취향을 가감 없이 드러냈다. 지독한 불면증을 앓는 가족들을 위해 그녀가 추천한 책은 1,000년 전 일본의 수필집인 세이 쇼나곤의 **『마쿠라노소시(The Pillow Book)』**다. “단편적인 리스트와 아름다운 묘사들이 이어져 잠들기 전 듣기에 완벽하다”는 것이 그녀의 설명이다.

반면, 작가로서의 직업병도 고백했다. 문장에서 ‘낄낄거리다(giggle)‘나 ‘헐떡이다(gasp)’ 같은 상투적인 표현이 반복되면 가차 없이 책을 덮어버린다는 그녀는, 최근 자신의 서가에서 가장 의외의 책으로 『골프장을 공공 섹스 숲으로 만들자(Make the Golf Course a Public Sex Forest)』라는 파격적인 제목의 에세이집을 소개해 웃음을 자아내기도 했다.

루이즈 어드리크에게 글쓰기는 여전히 풀리지 않는 수수께끼다. 하지만 그 수수께끼를 풀기 위해 기꺼이 바닥을 기고, 과거의 공포를 마주하며, 타인의 고통에 눈물 흘리는 그녀의 여정은 계속되고 있다. “우리에겐 신비로움이 필요합니다. 제가 하는 일에 대해 모든 것을 알지 못한다는 사실이 저를 계속 나아가게 합니다.”