Category: Uncategorized
Aaron Levie on current implicit AI regulation
We now have de facto AI regulation. It’s not obvious why from here on out models that have certain levels of capability or are trained on certain compute sizes won’t have to be reviewed by the government before release.
Realistically, as AI models became more and more powerful this was going to be inevitable (I think it’s too early, but here we are). So now it’s mostly just interesting to think about the implications and scenarios from here. A few would be:
* America gets to control who gets access to frontier intelligence and when. This generally works as long as we remain at the frontier at all times and don’t have a risk of being surpassed. At the moment we have a clear lead in frontier intelligence so this is a good bet, but lots of motivated parties would love to change that.
* This likely creates backlog of AI releases which means that we will see less rapid fire back and forth jumps in model progress. Bull/fine case is that we just get bigger step functions per release at a slower rate and we end up at the same point we would have. Bear case is those incremental smaller jumps were necessary for the continued flywheel of innovation.
* Other countries likely have even more incentive to at least hedge their bets with sovereign AI strategies so aren’t dependent on access to US AI all times. Previously this was relatively moot because the alternative wasn’t good enough, but that could change out of necessity and what we’re seeing in China.
* Open weights obviously a big winner here as it becomes what likely sovereign AI gets built out on, and what (for now) can still be released to the market without the same controls. One interesting question would be how regulation eventually extends to open models, which would have its own set of long term consequences.
Anyway some big updates to everyone’s mental models of AI regulation as a result of the capabilities we’re now seeing in AI. Wild times.
Here is the link. I would say I have long thought something like this was coming, but am pleased we got in so much early progress “under the wire” up until now. And here is more from Aaron.
Does fasting harm cognitive performance?
More than 2 billion people participate annually in Ramadan fasting, making its potential effects on cognitive performance important for workplaces, education and high-stakes decision-making. We study these effects in tournament chess, an incentivised, real-world cognitive task in which move quality can be evaluated objectively by a strong chess engine. We analyse nearly 300,000 games and more than 25 million moves played by almost 10,000 expert players from 178 countries over 10 years. Two validation exercises support our Muslim-status classification, covering almost 11% of the sample and survey evidence indicates substantial Ramadan fasting compliance among Muslim chess players. In the preferred intention-to-treat specification, using pre-game controls, player fixed effects and year-month fixed effects, we find no impact of Ramadan fasting on Muslim players’ overall move quality or shares of optimal and nearly optimal moves, with tightly bounded estimates around zero. Muslim players make 0.13 additional percentage points of large errors during Ramadan, but this small estimate is fragile across alternative measures, samples, Muslim-status definitions, fasting-compliance adjustments and event-study diagnostics, with no evidence of heterogeneous effects, selection bias, or compensatory behavioural adjustments. We conclude there is little robust evidence that Ramadan fasting broadly impairs cognitive performance among expert chess players.
That is from a recently published paper by Samuel Buckland and David Smerdon. Some claim that people think best when they are just a wee bit hungry?
Via the excellent Kevin Lewis.
Cricket and the railways
Railways are a golden thread in the history of cricket, making national competition possible in every current Test-playing nation (with the exceptino of the West Indies and Afghanistan). In later years, we will see railway workers as exporters of cricket to Scotland and Wales and beyond to Britain’s formal and informal commercial empires. We will see enduring railway-based teams, including in Pakistan the winners of cricket’s most comprehensive first-class victory.
That is from Richard Heller and Peter Oborne, Full Circle: A History of Cricket. And I had not realized this: “As recently as 1945, 98 percent of Australians had their family origins in Great Britain or Ireland.”
Thursday assorted links
Translated from the Chinese
I think this is the Cursor moment for academia.
The Stanford REAP team has made their move, CoPaper.AI is mass-terminating the manual labor of traditional empirical papers. Link: copaper.ai/landing
If using large models to write papers before was just about polishing and compiling references for you, then this Project from Professor Ross Griebenow’s team at Stanford is like dropping a nuclear bomb in the empirical circles of social sciences and economics.
The greatest truth is the simplest; the heaviest sword has no edge. Its functions are straightforward. Feed in the raw dataset, and within 30 minutes, it can generate a complete DOCX paper complete with full Stata/R code and publication-quality charts.
It chains together EDA, variable definition, econometric model building (from OLS to advanced DID, regression discontinuity, causal forests) all using an Agent workflow.
Every chart it produces comes with 100% reproducible Stata, R, EViews source code underneath. How many low-quality paper mills and data drones’ jobs will this smash?
Data drones and paper ghostwriters are collectively facing unemployment countdown. Because from now on, for social science papers, AI handles all the entropy-increasing drudgery—humans only need to define the problem.
Here is the link. Mostly that is not true, so perhaps the Chinese are trying to demoralize us. But will it never ever be true? In two years be true? Less?
What should the UAP Scientific Advisory Board do?
There are more and more frauds, charlatans, and lunatics entering this area of inquiry. It is important to stay disciplined on data-driven questions, most importantly to what extent are released (and unreleased) videos backed by radar, satellite, eyewitness and other forms of confirming evidence? By confirming, I do not mean “confirming they are aliens,” rather I mean “confirming they are real phenomena and not illusions of various kinds.”
Do not focus the discourse on aliens, rather focus on whether the phenomena are real. If they are confirmed as real, as many insiders insist, they we can return to debating what they might be. And focusing on concrete evidence is something a committee can be relatively good at. Trying to find agreement on “aliens” does not fall into that same category.
Wednesday assorted links
1. From my colleague Jonathan Beauchamp.
2. Why is China still exporting T-shirts?
3. Greenspan and Keynes crossed paths in 1944. Clarinet!
4. Robert Shiller opposes AI negativity (NYT).
5. The opportunity cost of Trae is really not that large. Think in terms of opportunity cost here, not “cost.” By the time the Wizards need to up the pay of their younger players, Trae’s contract will be expiring.
GLP-1 drugs and marriage
GLP-1 medications generate large weight loss and may also alter social and economic outcomes. Using the Understanding America Study, I compare women starting GLP-1s for weight loss with matched women who would like to start a GLP-1 but have not. Single women’s marriage/cohabitation rates rise by 29 percentage points and employment among baseline non-employed women rises 27 percentage points after six or more quarters. Existing partnerships do not dissolve, and already-employed women show no upward job mobility. The pattern suggests that part of the female obesity penalty operates at new-match formation rather than only through health or incumbent productivity.
Here is the paper by Rebecca Diamond. And here is a thread on the paper. And not everyone believes the size of these estimates. I do not find them so crazy? Here is Steven’s dialogue with GPT.
Tuesday assorted links
1. This paper was presented 57 different times!? Is that good or bad?
2. Benny Goodman and Lionel Hampton doing “I Got Rhythm.”
3. Claims about advancements in cancer testing.
4. Reconstructing Welles’s Don Quixote?
5. The obesity penalty in political elections.
6. Zvi on GLM-5.2.
7. New work by Mozart. And more information here.
Is the UK improving?
From a very credible Labour source:
– Wes Streeting promised the Chancellorship for not running.
– Capital gains raised to match income tax. Possible exit tax.
– Economic focus: devolution, plus state ownership of cost-of-living essentials (energy, water, transport).
– Nothing on AI or tech, bar higher capital gains and an EIS/SEIS-style relief for backing British businesses. (Spoiler: startups now incorporate in Delaware and raise on SAFEs. I’ve done 60+ angel investments; only two were eligible.)
Andy and Wes don’t seem to grasp that tech has been the core engine of growth for 20 years, and AI will only accelerate that.
So why would any founder build here? How does the UK compete with the US and China on AI? Where does growth actually come from?
The world economy is changing fast, and we need to be ready to thrive in it, not just survive.
I really hope this admin appoints some figures who actually get what’s happening. Losing business support early, from a disastrous first budget, was the beginning of the end for Starmer.
So, in a nutshell, no, the UK is not improving.
Dean Karlan has a Substack
He starts his new essay with this:
In 2014 I wrote in The New York Times that if your own team is not in the World Cup, you should root for the one whose victory would do the most good. Add up the happiness a title would create, more where more people care, more where incomes are lower, and more where a win would be a first rather than a habit, and root for the country on top. That year it was Nigeria. With 48 teams in 2026, and more of the world’s poorer and first-time sides in the field, I rebuilt the guide, with more nuance and, thanks to AI, at a fraction of the old cost. It updates itself as the games are played.
At least for now, you should root for the DRC he argues.
*The Migrants: A Memoir with Manuscripts*
Christopher de Hamel’s Meetings with Remarkable Manuscripts is one of the very greatest books of the last twenty years. So I buy whatever else he puts out, and I did not regret my purchase of this one. Imagine an intersecting tale of a boyhood in New Zealand (!), the medieval manuscript collecting habits of Colonial Secretary Sir George Grey, and a Bildungsroman of both aesthetic taste and personal maturation. The back cover notes that “Christopher de Hamel has probably handled more medieval manuscripts than anyone else alive…” That he is such a special person shines through in all of his writings. By the way, I learned that Dunedin is the Gaelic word for Edinburgh.
The new book you can order here.
Monday assorted links
1. John Horton: “The “literature” is going to become a collection of nodes/papers representing temporarily suspended computation with “citations” being contingent edges that describe how they sink uses the source; new data, better models & methodological changes will cause a Makefile-like cascading of update through the literature/graph. AI agents will autonomously add new edges + nodes.”
3. Persistently beneficial AI models. Through general benevolence.
4. “Concerns over therapy ferrets used to kill rats at UK’s largest children’s prison.”
Scotland facts of the day
By one metric there are 841,000 practicing Catholics in Scotland, with 184,283 attending Mass regularly. That puts Catholics as far outnumbering Protestants in Scotland, for the first time since the Reformation.
Between 2012 and 2022, the number of obese adults in Scotland rose 46 percent, to comprise about one-third of the population.
In 1950, 76 percent of Scots age 16 or older were married, now it is about 45 percent.
All those details are from the new and fun book by Alistair Moffat, To See Ourselves: A Personal History of Scotland Since 1950. From these tidbits I conclude that “Scotland as we knew it” is not just evolving, but also disappearing.
Sunday assorted links
2. Short TV clip of me on the Brazilian economy.
4. “Thrilled to announce the inaugural cohort of the 1991 Fellowship @mercatus. Meet some of the most talented and creative minds working on challenging policy problems at the state level in India.” Link here.
5. AI has won another literary prize.
6. Roon on worship. Roon is one of our best thinkers.