South Korea’s week from hell and the “what if” of tail risks
South Koreans have a tendency to dub their nation “Hell Joseon,” usually due to the extraordinary stress, pressure and expectations placed on children and young professionals in their careers and mating lives (my Korean language textbook taught “I am stressed out” and “I studied too hard” in Chapter 2). So the insane political turbulence this week, which saw the country’s president declare martial law only to rescind it a few hours later, only exacerbates the feeling of alienation.
Having lived, studied, reported from and visited South Korea for more than a decade, I have a lot of personal thoughts on the quagmire. I am grateful that everyone is safe and the situation resolved itself overnight while most locals were sleeping.
The wider and more interesting Riskgaming question though is about tail risks. What’s the probability of the Korean president declaring martial law on any given day? After all, it’s a constitutional provision and can essentially be declared at any time, not so different from a President Biden or Trump typing a few codes into the nuclear football and throwing the whole world into nuclear winter. This isn’t a so-called black swan — it’s a known risk. The probability isn’t zero, but it also isn’t high, which puts it in the “tail” of the statistical distribution.
The anecdata since 2020 is that our analysis of tail risks is horrendous. Statisticians continue to struggle with how to quantify the essentially unquantifiable. Will avian influenza break out this year and wreck global havoc, or will we be saved? (One hour before publication: the USDA will screen milk due to a growing outbreak across 16 states). Trump was nearly assassinated earlier this year in Pennsylvania, in an action that would have rapidly changed the course of American and global history. As pointed out in my inaugural newsletter three years ago, the U.S. presidency is the most dangerous job in the world. It unfortunately could happen again.
Polymarket and other betting sites do offer wagers on some tail events, mostly for entertainment. After all, there is no wisdom of the crowd in the stochasticity of life. Each probability is so low as to be imperceptible from impossibility, yet, the sheer quantity of tail risks means that at least a handful of them will take place in any period. It’s not so different from the p-hacking crisis in science: test enough variables across experiments, and a positive result will almost inevitably arise.
The challenge with risk isn’t to list thousands of terrible scenarios. The challenge is how to respond to ambiguity and sudden crisis. Part of the challenge (and fun!) of our Riskgaming scenarios is that no player knows the rules of the scenario in advance. This isn’t poker, where everyone has their favored strategies and has memorized a roll of probabilities. You have to be paranoid, but optimistic. You have to seize opportunities, while being alert to treason or blowback. Even the best actuarial accountants in the world can’t predict these events — one reason among many that new products like parametric insurance are being offered that require more narrow underwriting.
South Korea is a reminder that the whole world can change instantly. What’s unique is the speed of the country’s response. Within minutes, legislative staffers began barricading doors. Nearly 200 legislators managed to jump fences and reach the chamber in just a few short hours. Large vehicles like buses are now parked across the Assembly’s lawn to prevent military helicopters from landing near the building.
Chaos exists — but so does the catharsis of action.
Is China’s AI industry less efficient than expected?
I’m a big fan of Jack Clark’s newsletter Import AI, and his latest issue had an important discussion on the race between compute and efficiency in the context of China’s AI industry. With America implementing a complicated but increasingly comprehensive set of export controls on leading-edge chips to China (including more this week), that puts a ceiling on the country’s compute availability, limiting training and inference.
I’ve argued in the past among friends that the obvious counter-response to such actions is that AI engineers in China will focus far more on efficiency than American companies with flush coffers and bubbly valuations in order to take advantage of every ounce of compute they can scrounge together. Since users generally like the most responsive applications, I’ve even more boldly argued that China’s potential efficiency focus could lead it to unintentionally become the leader in AI applications, since it would be solving for a key problem for most users.
Yet, Jack points to an interview with Liang Wenfeng, the founder of Shenzhen-based DeepSeek as a major counterpoint. As I noted in the last Riskgaming newsletter, DeepSeek just published the best-performing AI model in the world, and so Liang’s comments are quite interesting:
Philosophically, DeepSeek thinks about the maturity of Chinese AI models in terms of how efficiently they're able to use compute. "We estimate that compared to the best international standards, even the best domestic efforts face about a twofold gap in terms of model structure and training dynamics," Wenfeng says. "This means we need twice the computing power to achieve the same results. Additionally, there’s about a twofold gap in data efficiency, meaning we need twice the training data and computing power to reach comparable outcomes. Combined, this requires four times the computing power. Our goal is to continuously work on narrowing these gaps."
(Of course, if the company is so inefficient, why is it leading so many AI benchmarks?)
Nonetheless, despite the intensity of China’s tech industry and government leadership on building leading AI models, the reality may be that other ingredients like data quality could play a much larger role in determining the outcome of its race with the United States. Liang’s comments provide fodder for analysts who feel that China’s tightened political environment and sensitivity around private companies collecting too much data would slow progress for China’s AI companies.
Jack also adds one other comment that is important to highlight:
The interview also provides an oblique acknowledgement of an open secret - a large chunk of other Chinese AI startups and major companies are just re-skinning Facebook's LLaMa models.
The open-source AI security debate is going to intensify dramatically in January, with senators and regulators ready to pounce on leakages of American AI leadership to China. Facebook’s strategy of releasing leading models (or, at least, the weights of these models) has come under scrutiny, and these comments from Liang are certainly no help to the company’s crusade for at least some openness in the AI market.
General Motors and its $5B write down
We have been hosting pre-launch runthroughs this week in NYC, DC and SF for our next Riskgaming scenario, "Powering Up: China’s Global Quest for Electric Vehicle Dominance” (reminder to sign up with Laurence Pevsner to join us at future events). A few of the players in the game have played “US General,” a fictional American automaker struggling to compete during China’s rapid electric vehicle transition.
So it was apropos that fiction met reality this week when General Motors, the non-fictional American automaker, announced that it would take a $5 billion write down as it restructures its China operations and recognizes that its joint venture with Chinese company SAIC is no longer nearly as profitable as it once was.
It’s just the latest ominous sign of a massive shift underway in consumer automobiles that’s been driven by China’s explosive growth in EVs over the past three years. Ford’s CEO Jim Farley has called the changing market conditions in China “existential” for the company, while Volkswagen announced the company’s first-ever layoffs on declining China sales.
As Alex Marley explained in a Riskgaming newsletter earlier this year: “With my [chinese] license in hand, I started to explore the world of Chinese automakers. My first stop was to visit the BYD headquarters in Shenzhen. It was truly shocking to see the quality and attention to detail that went into their most affordable vehicles (starting at just over $10,000). While the car itself did not feel as though it would last forever, the amount of technology that was integrated into the vehicle felt unprecedented at this price.”
It’s a classic tale of the Innovator’s Dilemma. Lower quality but extraordinarily cheap new competitors enter a market with early technology that rapidly improves. Incumbents move upmarket to maintain profit margins, only to find that the new entrants eventually catch up and surpass them in desirability. Today, the vast majority of China’s EV auto sales are with Chinese companies (with Tesla a notable exception). For everyone else, one of the world’s largest auto markets is closing rapidly.
Ford, GM, Volkswagen and others can certainly keep on producing cars, but the flush profits from China that helped to underwrite their expansion are now drying up. Unfortunately, there’s nowhere to go. The middle class in emerging markets like India and Indonesia are looking toward more affordable options than Western producers can build, while population decline and lowered interest in driving means that Western auto markets are in a very slow but inexorable decline.
Adding fuel to the bonfire will be the next generation of trade wars with the upcoming Trump administration. To respond to Biden’s new chip export controls this week, China announced a complete export ban to the United States of rare-earth materials like gallium and germanium. The unfortunate loser these days in many markets will be U.S. and European companies, who sell luxury and higher-margin goods that are increasingly out-of-reach for all but the most wealthy global consumers. For everyone else, emerging companies from China, India (Tata Motors) and others will take the lead.
Podcast: The Titanic Lessons of VC with Josh Wolfe
Every quarter, Lux sends an update to our limited partners observing the macroeconomic environment, the changes in venture capital, and our current thinking regarding the present and future of science and technology. This time, Josh Wolfe’s theme was “Titanic Lessons,” four classic parables from Greek mythology that elucidate our understanding of the world. Josh, co-founder and managing partner of Lux Capital, joins me to talk more.
Whether it is Prometheus offering fire as a form of “extensionalism” that expands the bounds of human powers, or Atlas taking on the burdens of the world in pursuit of the next intrepid voyage, Josh discusses how new technologies can rapidly augment human potential — but only if they are unlocked and unleashed. Unfortunately, so many of the world’s best innovators remain shackled in research labs and corporate offices without the resources and autonomy to succeed. That’s where our four investment strategies of Lux Labs, corporate spin-offs, tactical global opportunities and ‘fixware’ come in. We discuss the potential of each in turn.
Then there’s a wider set of warnings from Epimetheus and Menoetius, two Greek Titans whose arrogance and hubris would prove their downfall. We bring them up in the course of discussing the future of AI infrastructure, its expansive energy needs, the power of decentralized compute technologies and finally, the potential for Apple to emerge from behind as an AI winner.
🔊 Listen to “The Titanic Lessons of VC with Josh Wolfe”
Podcast: Why does America have the most expensive elevators in the world?
Recently, I wrote about one of the biggest challenges facing America: how our industries — and particularly those in construction and building — are becoming some of the least-efficient in the industrialized world. Today’s podcast episode identifies yet another problem, and it regards elevators.
Elevators aren’t just a conveyance of convenience, they are also crucial infrastructure for millions of Americans who struggle with mobility and anyone who has ever carried heavy luggage or groceries in or out of a building. Yet, the cost of America’s elevators is often multiple times more expensive than similar elevators in Europe and certainly Asia. Why?
That’s what we’ll discuss today with Stephen Smith, the executive director of the Center for Building, a think tank that studies building codes in global comparative context. His viral op-ed in The New York Times earlier this year has been read by everyone, and he’s continuing to do more research on how zoning and building codes collaborate to drive up prices for everyone. We’ll talk about that, as well as why America remains so suburban, the insider interests and negotiations that constrain construction efficiency, and why the West Coast is particularly bad for overhead.
🔊 Listen to “Why does America have the most expensive elevators in the world?”
Lux Recommends
- In the department of great science and great (maybe?) eating, “Chemists create world's thinnest spaghetti.” “In electrospinning, the needle in which the mixture is contained and the metal plate upon which the mixture is deposited form two ends of a battery. Applying an electrical charge makes the mixture complete the circuit by streaming out of the needle on to the metal plate.” Unfortunately, “Professor Williams added, ‘I don't think it's useful as pasta, sadly, as it would overcook in less than a second, before you could take it out of the pan.’” Meanwhile, our scientist-in-residence Sam Arbesman pointed to Cascatelli, the invented pasta shape by podcaster Dan Pashman.
- "Why did clothing become boring?” That was a fun question posed by Benjamin Breen that Sam loved this week. “The nineteenth century was the century of interchangeable parts. And so it makes sense that clothing, too, became more interchangeable at this time. A factory-woven shirt or dress made with a sewing machine was simpler and less durable than what had come before. But it was also orders of magnitude less expensive, and this meant that people had less of an emotional investment in individual articles of clothing.”
- Satellite imagery of BYD’s factory in China — the largest battery producer in the world — shows not only one of the world’s greatest structures, but also the space for its doubling. While BYD maintains its largest facilities in China, it has also expanded rapidly across Turkey, Thailand and Brazil just this year.
- Sam enjoyed James Somers’s essay in The New Yorker on “A Revolution in How Robots Learn.” “We can tie shoelaces better than ALOHA not because it has primitive, unsensing claws but because every shoe—every arrangement of laces, the way they bend and fall each time you lift them—is different. There is no Internet-size archive of the ways in which physical objects interact. Instead, researchers have come up with several competing methods of teaching robots.”
- Peter Hébert recommended this fun treasure hunt posed by Bitcoin millionaire Jon Collins-Black, who left clues across America to unearth a fortune. “You don’t have to be a genius to solve the clues. There’s no grand cypher. If you have curiosity, imagination, and the willingness to try something new, you can find the treasures that I’ve hidden.”
- Finally, Sam recommends this story on the UX designs of LEGO pieces that imitate computers. “Shape coding is one approach to differentiation, but there are many others. Colour coding is perhaps the only one to break into our everyday vocabulary, but we can add four more: size, texture, position and operation coding. Together these six are our allies in the design of error-proof interfaces.”
That’s it, folks. Have questions, comments, or ideas? This newsletter is sent from my email, so you can just click reply.
