Distillation seems terrible for leading edge models.
It is! On the positive side, OpenAI and Anthropic and Google are almost certainly using distillation to optimize the models they use for inference for their consumer-facing apps; on the negative side, they are effectively bearing the entire cost of training the leading edge, while everyone else is free-riding on their investment.
Indeed, this is probably the core economic factor undergirding the slow divorce of Microsoft and OpenAI. Microsoft is interested in providing inference to its customers, but much less enthused about funding $100 billion data centers to train leading edge models that are likely to be commoditized long before that $100 billion is depreciated.
Apple is also a big winner. Dramatically decreased memory requirements for inference make edge inference much more viable, and Apple has the best hardware for exactly that. Apple Silicon uses unified memory, which means that the CPU, GPU, and NPU (neural processing unit) have access to a shared pool of memory; this means that Apple’s high-end hardware actually has the best consumer chip for inference (Nvidia gaming GPUs max out at 32GB of VRAM, while Apple’s chips go up to 192 GB of RAM).
Meta, meanwhile, is the biggest winner of all. I already laid out last fall how every aspect of Meta’s business benefits from AI; a big barrier to realizing that vision is the cost of inference, which means that dramatically cheaper inference — and dramatically cheaper training, given the need for Meta to stay on the cutting edge — makes that vision much more achievable.
Google, meanwhile, is probably in worse shape: a world of decreased hardware requirements lessens the relative advantage they have from TPUs. More importantly, a world of zero-cost inference increases the viability and likelihood of products that displace search; granted, Google gets lower costs as well, but any change from the status quo is probably a net negative.
If building reasoning models is not as hard as people thought, we can expect a proliferation of free models that are far more capable than we’ve yet seen.
The existence of DeepSeek, and its explicit advocacy of open weights AGI, and potentially having it be the best model out there in the future in many people’s imaginations, has been a forcing function. Suddenly, people who previously stuck to ‘well obviously your restrictions are too much’ without clarifying where their line was, are revealing that they have no line.
And many more people than before are revealing that they prefer any or all of:
AGI with alignment only-to-the-user be made open weights.
Chinese open models be the best instead of American closed models.
A world where humans have no collective mechanism to control AIs.
Usually this is justified as ‘anyone with that power would act badly.’
That they get their cool free toys, nothing else matters, f*** you. Seriously.
Are effectively successionists, as in they want the AIs to take over, or at least they don’t seem to mind or don’t think we should try and prevent this from happening.
These people are often saying, rather explicitly, that they will use whatever powers they have at their disposal, to ensure that humanity gets to a position that, if you think about it for a minute or five, humanity probably cannot survive.
And that they will oppose, on principle, any ability to steer the future, because they explicitly oppose the ability to steer the future, except when they want to steer the future into a state that cannot then be steered by humans.
No, I have not heard actual arguments for why or how you can put an aligned-only-to-user AGI into everyone’s desktop or whatever, with no mechanism of collective control over that whatsoever, and have this end well for the humans. What that future would even look like.
Nor have I heard any argument for why the national security states of the world, or the people of the world, would ever allow this.
The mask on those is fully off. These people don’t bother offering arguments on any of that. They just say say, essentially, ‘f*** you safetyists,’ ‘f*** you big tech,’ ‘f*** you United States,’ and often effectively ‘f*** you rest of humanity.’ They are the xenocide caucus, advocating for things that cause human extinction to own the in-context-libs.
If that is you: I thank you for your candor. Please speak directly into this microphone.
So what becomes expensive in this world? Roles which justify massive amounts of test- time compute. The CEO function is perhaps the clearest example. Would it be worth it for Google to spend $100 billion annually on inference compute for mega-Sundar? Sure! Just consider what this buys you: millions of subjective hours of strategic planning, Monte Carlo simulations of different five-year trajectories, deep analysis of every line of code and technical system, and exhaustive scenario planning.
AI risk is no longer a future thing, it’s a ‘maybe I and everyone I love will die pretty damn soon’ thing. Working to prevent existential catastrophe from AI is no longer a philosophical discussion and requires not an ounce of goodwill toward humanity, it requires only a sense of self-preservation.
It will be tempting among commentators to add up this loss of stock market “wealth” and say it’s a terrible thing. But no. The winners will not be the producers of AI, which looks to become a marginal cost commodity with remarkable speed, but the users of AI. And the benefit will be on quantities, not on monopoly or fixed-cost rents. (I say this because the stocks of companies that can potentially use cheap AI did not jump up. Maybe they will, once more people figure out how to use cheap AI to make profits for a while. Maybe those companies haven’t been founded yet. That is now the Wild West.) The stock market measures the present value of profits, not the present value of social benefits. The profit, and ultimate benefit, of railroads was not so much in the railroad itself, but in the wheat fields of Kansas.
Before leaving office, President Biden announced expanded export controls on AI. Unless the Trump Administration rescinds it, this interim final rule will restrict the export of top U.S. AI models and hardware not only to our adversaries but to dozens of our close allies. Affected “Tier 2” countries include Portugal, Switzerland, and Israel, in addition to most of the developing world. Soon, U.S. companies will have to apply for a license from the Bureau of Industry and Security to export leading closed source models and advanced AI hardware to most of the world.
We want to know if AGI is coming. Chow, Halperin, and Mazlish have a paper called “Transformative AI, Existential Risk, and Real Interest Rates” arguing that, if we believe the markets, it is not coming for some time. The reasoning is simple. If we expect to consume much more in the future, and people engage in smoothing their incomes over time, then people will want to borrow more now. The real interest rate would rise. The reasoning also works if AI is unaligned, and has a chance of destroying all of us. People would want to spend what they have now. They would be disinclined to save, and real interest rates would have to rise in order to induce people to lend.
To lay my cards on the table, I approach AI from a point of relative agnosticism. I’m not a doomer or an accelerationist. My estimate of p(doom) would be considered high by normal person standards but is well in line with the expert consensus, which puts the chances at 5 to 10 percent.
Instead, Kulwin and Klippenstein are forging some sense of political solidarity from being Luddites. I don’t know how you can use LLMs without concluding that they’re already a very powerful technology — far more powerful than most experts would have predicted 10 years ago — with the potential to become more powerful in short order. The left may get dealt out of the hand if it doesn’t get better AI critics.
history shows that markets eventually find a way around artificial bottlenecks that generate super-normal profits. When layered together, these threats suggest NVIDIA faces a much rockier path to maintaining its current growth trajectory and margins than its valuation implies. With five distinct vectors of attack— architectural innovation, customer vertical integration, software abstraction, efficiency breakthroughs, and manufacturing democratization— the probability that at least one succeeds in meaningfully impacting NVIDIA's margins or growth rate seems high. At current valuations, the market isn't pricing in any of these risks.
I think that AI, like every other information technology, will end up creating complexity as well as processing it, that the robots will get in each other’s way just like we do, and that consequently we are going to systematically overestimate the benefits of the technology during the initial phase. It’s the Dinorwig Spoil Heap Problem – when the returns seem huge and the inconveniences are minor, you don’t notice that you are making structural and architectural decisions which will come back to bite you massively at a future date, when enough of the inconveniences have piled up that they’re not minor any more.
…the extreme edge of Rationalism and EA reveal a dynamic that runs through both of them: Apocalyptic predictions, asserted with a confidence buttressed by weaponized misuse of “Bayesian” logic, is driving young people insane by convincing them that they must take extreme steps to halt an imagined doom. We all kind of laughed when Yudkowsky floated the idea of physical attacks on data centers to slow down the development of AI, because otherwise “everyone on Earth will die.” But it’s increasingly clear that he really believes these things, and more than a few of his current and former adherents have taken him entirely literally.
everything being built by OpenAI and others is “unscoped” - it’s being built not for any specific utility, but towards the infinitely broad goal of creating God. LLM architecture that has become the synonym and umbrella for all AI was the trigger for a huge marketing and investment push not because it’s tremendously useful, but because it’s broadly accessible and appealing. If “AI” were being sold in realistic terms, you would have at least two different buckets. LLMs would be back-seated as the interfaces and search engines they fundamentally are. Real focus would instead be on data analytics tools that are genuinely useful for things like detecting patterns in cancer rates or the like, looking at new proteins and other materials, etc.
But the Rationalists and fellow travellers seem willing to simply brazenly lie about the reality of LLM technology because, just like Sam Bankman-Fried, they believe that no morality supercedes the only thing in technology or human society that matters: making sure “good guys” build AI first, so they can make sure it’s “aligned.” (Yes, the Rationalists created and are now participating in the race to create the thing they claim to fear most.) They have no fucking clue what any of those things really mean, because they have a naive mechanistic view of the universe and the human mind alike - but those are the words they use.
The economic angle of this techno-religious fraud is obvious once you see it. As AI reporter Karen Hao points out, it was OpenAI who pushed hard on the idea that scaling compute was the golden road to turning LLMs into actual “intelligence.” They are incentivized to deceive the public about the real-world potential of LLMs because they don’t care about actual products, they simply believe they’re in a race to crunch numbers as fast as possilbe until God pops up out of a box and makes them the Kings and Queens of Heaven Forever. Capital efficiency doesn’t enter into it.
I probably best expressed this belief in my post titled “Violence Is Its Own Party.” In that piece, I argue that there is a certain treacherous animal spirit stalking around in the WEIRD world, particularly among the young, a yearning for deliverance… and if they have to, they’ll take deliverance through violence. Our culture has erased transcendent meaning and left in its place short-form internet video, frothy pop music, limitless pornography, Adderall for the educated and fentanyl for the not, a ceaseless parade of minor amusements that distract but never satisfy. And people want to be satisfied; they want something durable. They want something to hold on to. They want to transcend the ordinary. And I’m afraid that, with God dead and the romantic ideal ironized into annihilation, the pure thrill of violence is one of the only outlets left to express the inexpressible
Covers of The Economist and Time Magazine that point in a direction for a particular asset class are contrarian. They work as contrary indicators in both directions.
If you invest in the direction of a cover of The Economist, you should expect to lose roughly 10% of your investment in the year after it hits the newsstands.
Workarounds also exist for users who run into problems such as failing to pass trading platforms’ “know your customer” checks. Such checks, which are standard practice in the financial industry, seek to determine customers’ identities and at times their sources of funds.
To pass the tests, China-based users can register shell companies in offshore jurisdictions, such as the Cayman Islands or British Virgin Islands, and set up institutional trading accounts on the platforms instead,
Stablecoin issuers, on the other hand, are consumers of credit rather than producers of credit. Their tokens are backed by ‘reserves,’ which in this context means a portfolio of (ideally) high-quality, low-risk securities which can be easily liquidated. But markets must supply those assets. In the event of a shock which triggers large inflows into stablecoins — i.e. a digital ‘dash for cash’ — their issuers would need to quickly find ways to invest those assets. Depending on the fiscal outlook and condition of credit markets, that could be hard to do in practice.
There are numerous historical parallels, but the best is arguably the national banking system prior to the founding of the Federal Reserve. National banks could issue currency, but it had to be backed by government securities. That meant the money supply and fiscal policy were inexorably linked. Unless the government was running a large deficit, which was rarely the case after the Civil War, there was no way to expand the money supply in response to shocks — the bonds to secure that new currency were simply not available.
Yet, it’s also less radical than it sounds. Direct offshore claims on the Fed already exist in the form of cash, a much dirtier dollar. And a wCBDC wouldn’t be creating the offshore dollar market; it already exists. To the extent the Fed would have to provide emergency liquidity, it’s unclear that’s something it’s not already doing.
If the plan is to have stablecoins around the world, better to pair them with an official version: the offshore digital dollar. While Washington tinkers, Beijing toils. Food for thought.
To sum up, someone who has $10 million to bribe Donald Trump will want to demonstrate to the President that their purchases drove the price of the target asset higher: it'll be far easier to demonstrate this by pumping the frictionless memecoin than the burdened-by-fundamentals stock.
You might wonder: is the frictionlessness of a memecoin, its lack of fundamentals, and the ensuing incredible ease by which it can be bribe-pumped a new feature that crypto has brought to the table? Not really. There's no technical hurdle preventing the NASDAQ from listing a non-blockchain version of the TRUMP memecoin on its own old-fashioned Oracle database. People could buy and sell this NASDAQ-listed meme-thingy instead of that blockchain version of TRUMP. But securities law gets in the way. Listing an unadulterated ponzi game on a national stock market has never been legal, at least not in my lifetime. Why putting one up on a blockchain is legal is beyond me, but look over there, the President just did it.
At the speed the train is leaving sanity station and heading to financial silly land, I suspect listing pure ponzis on the NASDAQ will soon be an accepted thing. Memeassets everywhere! Bribes for everyone!
He talks about tariffs (FT) as if they are anti-consumption but pro production. But tariffs are anti-production on the whole, at least outside of some well-known cases of increasing returns, and even then the tariffs have to be applied properly. Pettis does not present this very important qualification about increasing returns, and he basically presents an argument that we would expect economics undergraduate majors to reject.
But I decided to talk about a new view of trade imbalances, associated especially with Michael Pettis, that has been gaining some traction lately. It’s also mostly wrong, and some friends in the international economics community have asked me to take it on. I’ve already prepared a slide show for the presentation, and I thought I’d try talking about that slide show here.
Antoine flagged almost a year ago how Blackstone’s liability to the University of California had grown to $560mn by then, after BREIT’s 0.5 per cent loss in 2023. BREIT returned 1.95 per cent last year, but this still means it has fallen even further behind its performance guarantee to UC.
So what’s the damage now? JPMorgan analyst Kenneth Worthington last week estimated that the overall hit is “closing in on max loss” and will hit about $900mn if current trends continue.
Tesla’s biggest asset is hyperbole. The more extreme the hyperbole, the more valuable it gets. Maybe after-hours market participants understand the dynamics better than Tesla bears, so are primed to park fundamentals and trade on vibes. Or maybe something else entirely is going on.
Within STIRs, we like deferring rate cuts into 2026, expressed via a SFRZ5Z6 flattener (short SOFR Dec 2025, long SOFR Dec 2026). Meanwhile, the market does not expect a TGA drawdown to have much easing effect. As per our TGA drawdown infographic, swap spreads have widened, but SOFR-FF hasn’t tightened significantly. Easier funding conditions have yet to be priced in, so receiving the March SOFR-FF basis — i.e. shorting the SR1-ZQ (SOFR 1-month - Fed Funds) futures spread at the March expiry — could be the play.
housing supply—unlike most goods and services—adjusts very slowly in response to changes in demand. Under these conditions, they find that the optimal policy is very close to one that disregards housing entirely, focusing primarily on returning nonhousing goods to the inflation target.
it has been meaningfully more expensive to own the S&P 500® via futures than to obtain the same exposure via cash (e.g., by buying an ETF or the index’s underlying stocks).
On the supply side, banks could raise additional capital or allocate a larger fraction of their current capital to equity financing. Regulations could be loosened in ways that allow banks to deploy existing capital more aggressively. Alternately, supply could come from sources outside of dealer balance sheets—unlevered investors looking to deploy unencumbered cash might step in to finance equity exposures, given the attractiveness of the S&P 500® futures spread relative to other common uses of cash (e.g., investment grade credit).
That means any big increase in the value of Taiwan’s dollar — an increase that is implied by the huge size of Taiwan’s trade surplus and the fact that it is still weaker against the dollar than it was 30 years ago — could obliterate the finances of Taiwan’s huge insurance industry.
To put it plainly, Taiwan’s life insurance industry has bet its solvency (with the support of its regulators) on the assumption that Taiwan’s central bank will be able to avoid a meaningful appreciation of the Taiwan dollar.
That feels like a risky bet. The US Treasury has always treated Taiwan somewhat generously, but there’s no guarantee that this practice will continue. President Donald Trump is already making noise about the size of Taiwan’s trade surplus with the US.
Housing is one area where trickle-down actually does work. A recent Kevin Erdmann post showed that rents in Austin, Texas have declined especially sharply at the lower end of the income distribution (red dots), partly due to sharply increased construction of multi-family units:
There are going to be a lot of angry libertarians and antitrust lawyers who are going to talk about how the Antitrust Division is helping China by not allowing these two great companies to combine. But the truth is that such a merger would be bad for America. HPE and Juniper compete over pricing, features, and service, and such a merger would, as consolidation did for Boeing and Intel, lead to stagnation and failure. The fight for more innovation is particularly important, because Chinese tech giant Huawei is a fierce competitor in the space, and though it can’t sell in the U.S. due to security concerns, having nimble domestic companies is important in terms of ensuring the U.S. is competitive in technology globally.
Supporting the Mogadishu government is a zombie policy which only moves forward under its own weakening inertia. The one thing Somalia has going for it is that after almost 35 years of chaos there are quite a lot of Somalians in Western countries to lobby in favor of pouring tax money into the world’s most corrupt country.
The final judgment in the first instance obliges Argentina to pay $16 billion dollars + interest to Burford Capital, amounting to $17.1 billion at the time of writing.
For both parties it’s as clear as day that this amount is unaffordable, but that fact does not mean they can’t negotiate to reach a possible agreement.
It seems that Milei is willing to pay or at least negotiate, and Burford knows it. The question is what priority Milei will put on the agreement of a payment plan with Burford, and how much of a strain this will be on the Argentine economy.
Technology containment was one of the key elements of the Biden Administration’s tough China strategy. Couched in terms of the “small yard, high fence,” a series of increasingly tighter export controls and sanctions took dead aim on limiting China’s ascendancy as a technology superpower. The US put enormous pressures on western allies to join this anti-China campaign. Alas, the policy is an abject failure.
In a way it’s surprising to see such a slowdown in what is, after all, still only a middle-income country; Japan didn’t downshift until it was more or less at the same overall technological level as the West. But China appears to be falling into the “middle income trap.
What’s remarkable is that China’s leadership seems completely unwilling to adjust to this changing reality.
The solution, they decided, was to import labourers from Rwanda. From 1928 to 1956 over 150,000 people were 'transplanted' from Rwanda to the Kivus. First Tutsis, then Hutus. The Belgians bought land from local rulers for a couple of pennies, then settled the Rwandans on it, under newly created chiefdoms. When these chiefs became too powerful, they divided up the districts mixing different ethnic groups and creating more chaos.
At the end of the colonial period the Banyarwanda outnumbered other ethnic groups in some parts of the Kivus. Because so many chiefs had been removed and villages resettled, there was widespread confusion over land rights.
The Fulbe (also called the Peul, Fulani, and other terms) are a significant West African ethnic group, numbering roughly 30 million. The Fulbe are strongly associated with transhumant pastoralism (seasonal migration to access pasture for cattle during dry and wet seasons), although contemporary Fulbe use a wide variety of livelihood strategies and many live in urban areas. Despite the complementarity of herding, farming, and trade, Fulbe herders have been cast as outsiders and excluded from narratives of national unity in, for example, Mali, Burkina Faso, Ghana, Togo, and Benin.
The SHIPS for America Act goes a long way to securing the future of the U.S. Merchant Marine and its considerable contribution to strategic sealift, as well as boosting the American shipbuilding industry more broadly. Investment, innovation, coordination, and workforce incentives are all much needed and very long overdue, and the bill includes measures that mark considerable progress in all of these areas. However, doubling down on protectionism is a disappointing addendum to what is otherwise a piece of legislation that represents the best chance in decades for the U.S. maritime sector to move toward becoming successful and sustainable.
The Democratic Party of America deserves all the loathing in the land. By gaming identity, by placing undue emphasis on it in a bid to retain power without having to betray their corporate sponsors, they have opened up questions of identity to undue enmity amongst the population at large. By forcing this fraud on people, they have inflamed the very issues they purported to solve. The one silver lining of the election result might be that it finally brings the curtain down on identitarian orthodoxy, that we finally get a Left that sticks up for the proles, instead of one that insists on trying to patronise them into submission. A Left more concerned with the size and shape of a man’s wallet than what he keeps stuffed down the front of his jeans. I doubt it, but god knows more of the same is no longer an option.
It remains to be seen if the Trump effort to defang this shadow government will succeed, but it helps that he has support from economic elites. The shadow government does not live only on government handouts. It also thrives by selling indulgences to powerful people and business sectors. Having friends in the shadow government is better than having friends in politics, because politicians come and go, but the shadow government is permanent.
One reason for the swing to the side of Trump by the economic elites could be that they have grown frustrated with this arrangement. People who think they are smarter than the voters are going to think they are smarter than the donors. Like a business run into the ground by management, the large shareholders are now stepping in with the support of the small shareholders, to clear out old management. Trump is like the old greenmailers; except this time the target is Washington.
I would argue that most of the anti-Trump energy at the moment actually comes from the cucks rather than the progressives. The former are in their element; as their epithet implies, they actually like being humiliated, and long for a system that does so regularly and harshly. They fell apart during the Bush years when they actually got their way, and responded by sending their voters off to a pointless war so that they would return and hate them properly, which they did. Now they attack Trump so that they can enjoy the frisson that comes from being abused by actual traditionalists and the readers of their New York Times columns at the same time.
We live in exciting and promising times. There is no reason for anyone to blackpill, but there is also a great need to be mindful of pitfalls.
We aren’t going to get someone just like RFK but without whatever weird stories … We are either going to see a revolutionary rise against corporate power injected into HHS or we are going to cede everything to the billionaire AI kings.
The other is owned by a Maltese company led by the co-founder of a major Australian pet-sitting platform.
Your Strongest Bodyweight. I think I maybe really want to do this—why shouldn’t I weigh 300 lbs and be able to pick up cars?
Are you crying about eating every two hours, being “always full,” or struggling to walk up stairs? Then go take up ultra-marathons or road cycling, if you want to look like someone who’s been on chemo for two years. We’re in the strength business. That means we want you to look jacked, have a decent amount of muscle, and command attention when you walk into a room – all while keeping your clothes on.
The road ahead is still steep. The ghosts of Concorde loom, and the challenges of noise pollution, emissions, and infrastructure remain formidable. But history has a way of bending toward those who refuse to accept its limitations. Blake Scholl was never supposed to be the one to resurrect supersonic flight. But then again, neither were the Wright brothers supposed to fly.
The area was originally settled by the Lenape Indian tribe, and a more definitive source says the name Punxsutawney derives from a Native name in the Lenape language, Unami: Punkwsutènay, which translates to "town of the sandflies" or "town of the mosquitoes" (punkwës– 'mosquito' + –utènay 'town').
Here’s the surprising figure. Producing enough drinking water for someone — assuming 3 litres per day — costs just $2.30 for the entire year. That’s less than the cost of a single bottle of water in many countries.
This was far cheaper than I’d have guessed, and in a water crisis I would agree that this is “absurdly cheap”. More expensive than many of us get from the tap today, but pretty cheap for an essential resource.
Reminder, don’t let Bill Deblasio hold your groundhog
