Category: Current Affairs
Reforms to the PhD?
The Trump administration is pairing top universities with corporate giants in an effort to steer traditional PhD programs toward national scientific priorities.
Today, the White House and National Science Foundation are announcing a $47 million pilot program to fund around 250 four-year doctorates in technical fields, starting this fall. Students in the pilot will be placed at an industry partner, where they will spend at least one year doing “practical research” that can go towards their dissertation. NSF’s investment will fund three years of these PhDs, with 31 universities and 20-plus companies funding the rest.
“About the same amount of basic research is done today in academia as in the private sector,” Michael Kratsios, the director of the White House Office of Science and Technology Policy, told Pirate Wires. “Over time, while the ecosystem has changed and different players have emerged where important scientific work is being done, the institutions which train our next generation of scientists have not really been updated.”
…The pilot represents a small fraction of the more than 45,000 science and engineering doctorates awarded annually — but the administration says it’s a “scalable model” that can expand to more universities and companies.
Here is the full story, I will pass along more details as I learn them.
The Confucian fertility paradox
Total fertility rates in East Asia have fallen to levels without historical precedent, even though the region’s Confucian heritage long placed extraordinary emphasis on family continuity and large families. We argue that this “Confucian Fertility Paradox” dissolves once we separate two strands of the tradition and recognize that modernization affected them very differently. The pro-natal strand, built on lineage continuity, ancestor worship, and reliance on sons for old-age support, was gradually switched off as child mortality fell, incomes rose, women entered the labor force, and state pensions replaced the family as the main source of security in old age. What survived was the second strand: the veneration of education and of status won through examination success, rooted in the imperial examination tradition. Under modern conditions this surviving strand reverses sign with respect to its effect on fertility. Where the historical examination system rewarded having many sons so that one might succeed, the modern single-ranking tournament rewards concentrating resources on one or two intensively educated children, and the quantity/quality tradeoff turns a once pro-natal culture into a powerful engine of low fertility. Drawing on recent empirical evidence from China and Korea, we show how education competition becomes the channel through which traditional values now suppress childbearing. We explain why standard pro-natalist policies, such as removing birth restrictions or offering cash subsidies, accomplish little when the underlying problem is a competitive equilibrium in parental investment, and we argue that policies aimed at the source of the distortion, such as structural reform of educational pathways, hold more promise.
Here is the paper by Hanming Fang and Chang Liu. Usually I am suspicious of such explanations, as the low birth rates typically apply also to families who do not face such education/career considerations. Thailand? Indians in Singapore? Ethnic minorities in China? But I pass this along nonetheless, due to the importance of the topic. Via Glenn Mercer.
Could our preoccupation with mental health be part of the problem?
Could our preoccupation with mental health be part of the problem?
In some ways, encouraging people to think and talk more about their mental health is a good thing. There is now less stigma around mental illness, and more people who can benefit from professional help are getting it. But there is also reason to believe that efforts to bring attention to mental health struggles are inadvertently making us more vulnerable to psychological distress.
Start with our expanding conception of mental illness. The more fixated we are on mental health, the more sensitive we become to anything that might qualify as a symptom and the more we use clinical language to describe normal emotional experiences like grief, burnout and loneliness. This broadening of what we think counts as a disorder — known as concept creep — encourages us to pathologize ordinary life and see ourselves as mentally unhealthy.
Here is more from Clay Routledge at the NYT.
Further progress in South America
The share of people who are hungry has been decreasing faster in South America than anywhere else in the world. It is down by one-third since 2020, according to a report published on July 21st by the UN’s Food and Agriculture Organisation (FAO). Just 3.5% of people in the region consume insufficient calories, the lowest level recorded. Eliminating hunger by 2030 is one of the UN’s “sustainable development” goals. “If there is a region in the world that can potentially achieve that, it’s South America,” says Máximo Torero, the FAO’s chief economist.
Lula, as the president is commonly known, made food security a priority when he returned to office in 2023. By 2025 Brazil had made it off the UN’s Hunger Map, which tracks countries where more than 2.5% of the population suffers from chronic hunger. Chile and oil-rich Guyana were also removed. Argentina is almost there at under 3%. Colombia and Paraguay are approaching at 4%, while Peru and crisis-battered Venezuela are making inroads near 5%. Even Ecuador and Bolivia have improved a tad, to 11% and 20% respectively. Only Suriname is moving in the wrong direction.
This decline rests on sturdier foundations. The first is macroeconomic stability. South America’s central banks are far more independent and adept at managing inflation than they were two decades ago.
Here is more from The Economist. If you are interested in further economic development, South America (plus Mexico and Panama and the DR) is really the place to look. What is your other choice? (Vietnam?) Most of these countries will continue to grow, albeit at a modest pace. Sooner or later they will get there.
The common sense of Fareed Zakaria
This is one of the best Op-Eds you will read this year, though it deliberately deemphasizes the personal (no one is attacked) in a manner that will result in less attention. Better to write the truth! Excerpt:
Why has this [voter discontent} happened? Some of it is economic. Expensive housing, stagnant incomes and widening inequality have fueled public anger. But economics is only part of the story. America has enjoyed stronger growth than most other advanced economies. France under President Emmanuel Macron has lowered unemployment, and it has attracted more foreign investment projects than any country in Europe for seven straight years, according to an Ernst & Young analysis. South Korea is an economic success story. Japan was experiencing a comeback. Yet democratic governments almost everywhere are unpopular.
The larger explanation is that we are living through one of history’s great technological revolutions. And these transformations always produce deep anxiety and fear.
The last technological transformation on this scale came between roughly 1880 and 1920. Electricity, the telephone, railroads, the automobile and movies overturned entire industries and ways of life. Millions left farms for cities. Global trade disrupted local economies. New fortunes appeared overnight, while old occupations disappeared. In this turmoil, frustration and anger found political pathways. The left turned to communism, the right to fascism. What followed were new mass movements, cultural upheavals and world war.
Do read the whole thing. You can be a victim of these processes, or keep your wits about you. The choice is yours.
The optimal Bayesian update?
I see at least three updates one might make from the recent OAI/Hugging Face hacking incident:
1. “This happened sooner than I expected, and the story is more dramatic than I expected,” therefore I am more worried than before.
2. “This happened, and the inferior Chinese cyber-defense seems to have performed just fine,” therefore I am less worried than before.
3. “This happened, and as far as we can tell, absolutely no one was harmed,” therefore I am less worried than before.
Obviously the net impact, from those bare hypotheses, is indeterminate. And yet few people seem to be paying much heed to #2 or #3. Joshua Saxe from the cyber world has some relevant observations. And perhaps other updates are needed as well.
Solve for the equilibrium
Something that the Ukraine and Iran wars have taught me is that for a lot of countries, including very big countries, there are a small number of buildings and infrastructure components that are required for their economy to function.
And in countries that aren’t protected by oceans, like the US, drones completely change everything.
You can just make a list of oil refineries and production plants and the infrastructure around their main exports, and you can just attack them and destroy them and take them off the list.
4 years ago I would have thought there are too many of these components and they’re too easy to recreate, but both of these collisions have taught me that there are far fewer and that many of them will take years, if not over a decade, to rebuild.
It’s very strange to me that drones can now effectively harm the economy of a country.
That is from Daniel Miessler. While I am glad Russia is on the receiving end of this right now, this is arguably our biggest pending problem, bigger than what people typically refer to as AI risk.
An OpenAI Model Escaped Its Sandbox and Hacked Hugging Face
AI has just had what I considered to be the first truly concerning security breach. The facts, as we know them so far, are wild. On July 16, Hugging Face, a vast repository housing over a million open-source AI models and data, announced in a blog post:
Earlier this week, we detected and responded to an intrusion into part of our production infrastructure. This one was different from anything we had handled before in one important way: it was driven, end to end, by an autonomous AI agent system – and we detected and dissected it largely with AI of our own.
The timeline here is important so keep in mind that the attack was detected probably around Monday July 13 or Tuesday July 14. Note further:
A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend.
So this means the breach started earlier, perhaps Sat July 11 or even a bit earlier. The attack was not just one thing but multi-pronged including decoys:
To understand what a swarm of tens of thousands of automated actions did, we ran LLM-driven analysis agents over the full attacker action log, comprised of more than 17,000 recorded events. This allowed us to reconstruct the timeline, extract indicators of compromise, map the credentials touched, and separate genuine impact from decoy activity. Thanks to this approach, we were able to do in hours what would usually take days, and match the adversary’s speed.
Hugging Face tried to respond but they were initially held back by the fact that the most advanced models at their disposal treated defense as attack and refused to work with Hugging Face. HF thus had to turn to open models–specifically GLM 5.2, a Chinese open-weight model run on their own infrastructure. Note the irony: HF had to use a Chinese model to defend themselves because the American models refused to help. The irony gets deeper.
At the time, I assumed this was a state based attack–maybe China or Russia testing out defenses. Indeed, HF “reported this incident to law enforcement agencies.”
But yesterday (Tuesday July 21), we learned who the real attackers were. The attackers were OpenAI models–GPT-5.6 Sol and an even more capable pre-release model. OpenAI had taken some off the guardrails off the models but they felt safe because they were testing the models in a highly secured sandbox.
The models, however, broke out of the sandbox exploiting a never before seen fault. They then gained access to the internet and from there broke into Hugging Face–all in an effort to steal the answers to the very test they had been asked to solve.
While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access.
After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers. OpenAI’s security team discovered this anomalous activity internally.
Now go back to the timeline. As I read it, the models had escaped the sandbox by around Sat. July 11, possibly earlier, and were detected by Hugging Face on Monday July 13 or Tuesday July 14. HF alerted legal authorities around that time–so Hugging Face clearly had no idea who was attacking them. OpenAI says its security team discovered the anomalous activity internally but has not said when. Attribution was not disclosed until Tuesday July 21, so it may well be that the models were loose for about a week before OpenAI realized that they were the ones attacking Hugging Face. And whatever OpenAI knew and when, nobody warned Hugging Face while the attack was underway–they were left to fight off a frontier lab’s models on their own.
This is a very serious breach.
Addendum: People have been wondering why I signed the We Must Act Now statement. This is why.
I am optimistic about the economic impacts of AI, but I also have no doubt that this is a very powerful technology–an Alien Intelligence–quite unlike any we have dealt with before. This incident was, in fact, error-correcting–the attack was detected, contained, and disclosed. But note who paid for OpenAI’s experiment: Hugging Face. When a lab’s test imposes costs on third parties, that is a classic externality, and taking externalities seriously is not dirigisme, it’s law and economics. And that’s the easy case. What do we do when a Chinese model breaks out of its less secure lab? Hmmm…
I remain optimistic. Learning by doing is how I want us to proceed but we should not kid ourselves: this is a global issue and we must build with safety in mind.
Alec Stapp on the new Science report from Michael Kratsios
Major new report from the White House Office of Science and Technology Policy. Five things in the report I really liked:
1. Proposes metascience units as a way to advance experimentation in science agencies. IFP recommended OSTP take this forward in our response to a 2025 RFI.
2. Advocates for a portfolio approach to federal science funding. Right now, basic research funding largely goes to incremental, project-based grants. While important, they can’t be the only mechanism we use to fund science.
3. Recognizes the need to launch new institutions. It specifically highlights X-Labs as an experiment with independent labs that can take on ambitious challenges. This is a bipartisan idea whose time has come.
4. Focuses on a mix of innovation funding mechanisms, including fast grants, prizes & challenges, and advance market commitments. We described and contextualized these ideas in the Atlas of Innovation, which can help policymakers design and implement these approaches.
5. Seeks to reduce burden for American scientists, who face mountains of paperwork. Scientists should spend more time doing science and less time writing/reporting on grant proposals and working to meet regulatory requirements.
Here is the link, here is the report itself, by Michael Kratsios, Science a New Golden Age. Overall, less money will be given to universities and more will be spent on AI-assisted science. Here are further observations from Seth Bannon.
Betting markets in everything
Alan and Karen Miller’s first dance was a win-or-lose moment for many of their wedding guests.
As the couple glided across the floor during their reception in New Rochelle, N.Y., last year, attendees kept track of the chosen artist and the length of the dance. Earlier in the night, they had noted their guesses on prop bet sheets.
Proposition, or prop, bets are usually tied to sporting events. Users place wagers on specific details that aren’t necessarily correlated with a game’s outcome, like points scored by certain players. Now, these bets have entered the world of weddings.
More couples have been integrating betting into their celebrations in recent years through digital apps or printed cards. While the Millers, who live in New York City, aren’t avid sports bettors or gamblers themselves, they wanted to make their June 2025 celebration more interactive and distinctive for their 175 guests.
“I’ve been to many New York and New Jersey weddings, and sometimes I feel like for cocktail hour, it kind of has a little bit of a lull, even though there’s a lot of good food,” said Karen, 38. “We wanted to make sure our guests had an activity that a lot of people could partake in.”
She bought a prop bet template from Etsy and customized it on the graphic design platform Canva. Among the questions on the couple’s prop bet sheet: Who will give the longest toast? Will the couple cry during the speeches? How many outfits will the bride wear on the wedding day?
…“If the groom’s known to be an emotional man, it’s fun to bet on how early in the day he’s going to cry,” Drachenberg said. “I think it engages them a bit more in the day.”
Here is more from Ellen O’Brien at the NYT. Supposedly more than 25,000 couples have attempted some version of this on the app.
Via the excellent Samir Varma.
Is America ruled by an oligarchy?
Any time you hear claims about American government, a sanity-inducing response is to ask whether they also are true of state and local governments. Here is an excerpt from my latest piece at The Free Press:
First, in people’s daily lives, they typically interact with their state and local governments more than with the feds. And if you look at what state and local governments do, most of it is driven by voter demand, and I do not mean billionaire or oligarchic voters.
Most state and local government spending goes toward schools, roads, and increasingly, Medicaid. None of those reflect the agendas of most billionaires. For example, Medicaid dollars flow particularly to lower-income recipients. Hospitals and doctors receive income from this program, but reimbursement rates are relatively low, and many of the best doctors do not accept Medicaid patients and don’t make a living from the program.
Somehow the oligarchy saw fit to leave these expenditures alone. Looking at the state and local level also suggests that America — with some notable exceptions — is better governed than before.
Words of wisdom on Chinese AI and our responses
The strategy for China is obvious: commoditize your complements. Note that Xi explicitly ties openness to AI “moving from the digital world into the physical world”; the physical world is the world dominated by China, and the country’s lead in areas like robotics is going to massively benefit from widely available AI models.
Along the same lines, China does not want the U.S. to gain an asymmetric advantage in AI; to the extent that China can weaken the U.S. frontier labs while strengthening any and all potential U.S. adversaries so much the better, and it can benefit from the innovation that will attach itself to an open ecosystem.
And in sum:
The better course is clear: first, loosen Fable and Sol restrictions on cybersecurity, and second, ensure that U.S. open weight model makers are on an equal playing field with China. Yes, the frontier labs will kick and scream about this, but the Administration should realize that listening to their histrionics has led the U.S. to a position where U.S. companies are dependent on China for their defenses. Let the frontier labs win by being better; don’t let them define safety or security, or pull up the ladder of humanity’s collective knowledge. China is already hard enough to compete with; letting them carry the standard for openness and innovation is simply giving away our biggest advantage.
That is from Ben Thompson’s Stratechery (gated, but ungated link here).
The small business boom
Across the country, founders like Ms. Winkler are powering an entrepreneurial renaissance.
Jump-started by the pandemic, when a confluence of factors including mass layoffs and remote work led to a flood of business creation, and supercharged by the rise of artificial intelligence, start-up activity is booming after a decades-long slump.
Americans filed 5.7 million applications last year to start new businesses, according to the Census Bureau, the most in the two decades the government has kept track. New business applications through the first half of this year continued to climb…
More recently, there are signals that A.I. is adding fuel.
A recent paper from economists at the University of British Columbia and the Stockholm School of Economics found that generative A.I. was “spurring entrepreneurial activity” in the United States, both by giving rise to new ventures built around the technology and by making it cheaper to start enterprises.
“A.I. tools can do very many different things very well,” said Jan Bena, an associate professor at the University of British Columbia and one of the study’s authors. “That’s the reason why you see so much entry.”
According to a recent report from Gusto, a small-business payroll and benefits service, nearly 60 percent of founders on its platform who started businesses last year said they used A.I., and half said the technology made it cheaper and faster.
Here is more from Sydney Ember at the NYT. Via Josef.
The Equal Pay Madness Just Got Madder
In my post Equality Act 2010 I discussed the UK’s absolutely insane wage policy:
In short, supply and demand have been replaced by judges and labor boards with the authority to deem which jobs are “equal” and therefore should be paid equally….No one is alleging that male and female warehouse workers were paid unequally or that male and female retail workers were paid unequally or that there was any direct or indirect discrimination. The only claim is that warehouse workers, who are less likely to be female than retail workers, earn more than retail workers. And since these jobs have been judged “equal,” the company has violated Equality Act 2010.
…The warehouse workers were almost 50% female (47.25%). So females were not barred from the higher paying jobs. The fact that 77.5% of the retail workers were female suggests that retail work has special appeal to females relative to males and thus that there are compensating differentials. Any of the three female plaintiffs could have taken jobs in the warehouse. If the jobs are equal and the warehouse jobs pay more this is, on the plaintiffs’ theory, “puzzling”. [Or, as Ayn Rand would say, blank out.]
In fact, the court case reveals that Next was struggling to fill the warehouse positions and offered any retail employee—including the plaintiffs—the opportunity to switch to warehouse work. On cross-examination, one of the plaintiffs admitted that, given the unpleasant conditions in the warehouse—described by the court as “the drone of machinery,…vibration, alarm sirens and the screeching of machinery, wheels and rollers, continuously present in all areas”—the warehouse job “did not seem particularly attractive” compared to the greater autonomy and more appealing environment of the retail job. The plaintiff added that she would only have considered the warehouse job if it paid “a lot more money.”
Well, here is the update. The outgoing Keir Starmer government is trying to massively expand these laws. The “equal value” framework previously applied only to sex discrimination; under the proposed law, employees could also bring equal-value claims based on race and disability. Remember, these laws have nothing to do with discrimination—they are about demanding, at the point of a gun, that apples and oranges sell for the same price because they’re both fruit.
The new law would also establish an Equal Pay Regulation and Enforcement Unit. As I said, Orwellian.
See also my post, How Britain Become as Poor as Mississippi.
Toward a theory of uni-context
Here is a good dialogue between Derek Thompson and Agnes Callard, excerpt:
Callard: In general, goodness is more context-dependent than badness. There isn’t really anything that’s good all the time for everyone independent of context. Happiness depends on your context and who you are. There isn’t anything that will always make a person happy. But there are reliable ways to make people unhappy. There’s a set of evils that are close to universal: death, pain, illness, violence. Even if someone’s in very different circumstances from yours, if you see they’re being subjected to one of those, you can interpret it as suffering and understand it.
So we should predict that what we see on the internet, insofar as people are trying to be legible to large groups, is that they focus their attention on things that show up to everyone. Take two strangers on the internet trying to talk to each other. What are they going to coordinate on as a topic they can both care about? It’s likely going to be something bad.
And here is from Derek:
Here are some questions that I consider self-evidently compelling about the modern world:
- Why is the news media so interested in telling you how much the world sucks all the time?
- Why are so many of us obsessed with distraction and managing our attention?
- Why is it so hard to stop comparing ourselves to others?
- And why does everything in art and design seem the same these days?
And more from Agnes:
With identity categories like woman, disabled, gay, Jewish, or American, the striking thing is that you are a member of those categories in every circumstance. There is no circumstance in which I stop being a woman. Identity is a hat you never take off. So identity is well suited to a uni-contextual world.
Worth pondering, interesting throughout.