Impedimenta Developerum!
NME: Harry Potter fans have succeeded in moving a construction project that would have originally gone through the “grave” of the character Dobby.
The 125-mile Greenlink power connector, which costs £430million, is intended to link power between the UK’s National Grid and Ireland, running between County Wexford and Freshwater West in Pembrokeshire.
However, the latter site is known to Potter fans as the site of house elf Dobby’s grave in the film Harry Potter And The Deathly Hallows, and contains a pile of stones with the words “here lies Dobby” that many flock to…
All I can say is that Voldemort would never have put up with this shit.
How well does AI peer review work?
Claude and I planted 100 known errors into 10 open-access psychology papers and then ran them through frontier models and two commercial AI review tools. In brief:
- The best single system caught 71 of 100 errors, while the worst caught 30.
- Pooling every system’s output caught 93 of 100. Models are only partly correlated in the errors they find, making ensembling a big lever for finding issues in papers. Check your papers against multiple models!
- Seven errors could not be caught by any system. All were omissions — information deleted from a paper rather than mistakes inserted into it.
- Refine.ink contributes more unique catches than any other single system, though it’s expensive.
- I didn’t measure false positives and I don’t know how this error distribution compares to the distribution of errors in real papers.
- I’ve made the papers, errors, model outputs, and the full experiment log public. I hope people can build on this work to create a comprehensive eval benchmark across disciplines.
That is from Paul Litvak, here is more. Note that is not even using the very latest generation of models.
Who enters Congress?
We trace the family origins of Members of Congress (MC) born between 1830 and 1950, linking each MC, their parents, and their brothers to the complete-count censuses. Future MCs have always been economic outliers. By 1940, close to 60 percent of MCs aged 18–40 held a college degree, against under 5 percent of comparable young men outside Congress. In the 19th-century censuses, MCs as adults held roughly three times the wealth of demographically matched controls. They also come from economically elite families: their fathers earn more, attain more education, and hold more wealth. The brothers of future MCs sit between the population and the MCs themselves. Across the socioeconomic measures we compare, we calculate that roughly half of an MC’s adult premium is shared with his brother and the other half is his own. Even as the American economy and political system have changed, the family background gap has widened over the last century and a half. Wealth, household servants, and four separate occupation-based scores all show the gap holding or growing across cohorts. We study four Progressive-Era reforms in a triple-difference design—women’s suffrage, the secret ballot, direct primaries, and the direct election of senators—and none produce detectable shifts in who reaches Congress. Our results document a Congress drawn from the most fortunate families in American society in every cohort we observe.
That is from a new NBER working paper by
Brian Chau at work
I am automating investigative journalism at Effort.News
We’re breaking five stories today based on verifiable financial data which everyone missed for years.
Here is the entire thread, which also describes the five stories.
Monday assorted links
1. Tax land more heavily than structures.
3. Francesco Cabella conversation with (an older) Victor Niederhoffer.
4. New homes in Tatu City, Kenya.
5. Bill Goichberg, RIP (NYT). America’s greatest chess organizer, he was always very nice and encouraging to me.
Do Faculty Affect Student Partisanship?
We study whether Democratic college professors make their students more liberal. We link voter data to salary records from 33 state flagships and show that faculty skew Democratic, especially in the humanities and social sciences. We then use student transcripts from one flagship to estimate causal effects. Students become more liberal during college regardless of major and sort toward ideologically similar instructors. Exploiting plausibly random variation in when instructors teach courses, we find no effect of faculty partisanship on student partisanship. Text analysis shows exposure to liberal topics reflects student demand rather than instructor supply, leaving little room for indoctrination.
That is from a new paper by Micah Baum, Joaquín Endara, and Annaliese Paulson. Via Jay Fan Bavel.
Act like it is science
I am seeing so many doom prognostications, or least severe worries, due to the OAI/HuggingFace incident and related stories. But virtually all of these I find underargued to say the least.
So I have a simple request. If you are worried, and wish to persuade the doubters, try the methods of science. I would like to see the following:
1. Your outline of how, say two or three years from now, we might estimate the additional cybersecurity costs from the AI break-ins. I am convinced that number is not zero, but give me your method please. If it helps, here is an estimate of past cybersecurity costs, done by top economists.
2. Your current numerical estimate of what those costs might end up being, of course to be tested against what actually happens over time. Obviously, you can do this for a few different regulatory/safety scenarios. This is one simple way to prove yourself largely correct, albeit with a lag. (NB: you do not have to take this as a substitute for your preferred safety measurres. But please try to be specific in your predictions.)
3. A list of what stocks or other assets you have shorted, now. Obviously if your answer to #2 is sufficiently low, you could answer here zero, as I would do. I expect costs, but not so high that we cannot muddle through and have expected positive stock returns.
This would all help to make the discourse more scientific and less like an extended exercise in personal anxiety management.
If it is all so important, surely this exercise is worth the effort?
Sunday assorted links
1. African use of Chinese AI (NYT). And what happened to Talenti gelato? (NYT) And the NIMBY brutalists who live in the Barbican (NYT).
2. Modernist Gothic in Copenhagen.
3. AI-generated Odyssey in Vietnam.
5. Time magazine has a special version that only the AIs can read.
6. Tymofiy Mylovanov has a good summary of my greatest fear.
JFK’s AI Voice
From Kennedy’s 1961 inaugural:
- “Ask not what your country can do for you—ask what you can do for your country.”
- “Let us never negotiate out of fear. But let us never fear to negotiate.”
- “United, there is little we cannot do… Divided, there is little we can do.”
- “If a free society cannot help the many who are poor, it cannot save the few who are rich.
and from the 1962 moon speech at Rice University;
- We choose to do these things not because they are easy but because they are hard.
and from Lyndon Johnson’s inaugural speech:
- John Kennedy’s death commands what his life conveyed—that America must move forward.
- On the 20th day of January, in 1961, John F. Kennedy told his countrymen that our national work would not be finished ‘in the first thousand days, nor in the life of this administration, nor even perhaps in our lifetime on this planet.’ But, he said, ‘let us begin.’ Today, in this moment of new resolve, I would say to all my fellow Americans, let us continue.
Sound familiar?
Were John F. Kennedy’s speeches written by an AI? Probably not. But were AI’s trained on the speech writer Ted Sorensen? It would appear so. Or to be more accurate, AIs have absorbed the Sorensen toolkit: symmetry, parallelism, antithesis, repetition, and the neatly turned reversal.
Sorensen is considered a great stylist, and these are iconic lines. The problem with AI voice is not the techniques themselves. It is the overuse of them—their deployment to describe a trip to 7-Eleven with the same rhetorical flourishes once used to send men to the moon.
I suspect that will be an easy problem to fix, so don’t expect AI voice to continue.
Hat tip: murgh
Pressure
Who would have thought that someone could make a gripping movie about predicting the weather? Nevertheless, Pressure delivers. To be sure, it’s predicting the weather for D-Day. Excellent cast. Hews very close to the truth. The two camps really did divide on theory versus past prediction and the weather was as portrayed. After the war, John F. Kennedy asked Eisenhower what gave the Allies the decisive advantage during the D-Day invasion, Eisenhower replied, “We had better meteorologists than the Germans.”
I agree with Tyler’s earlier review: A truly excellent movie, one of the best of the year. Specifically, it concerns the meteorological forecasts (!) leading up to the D-Day invasion. Thematically, it is about the differences between Americans and Brits, how bureaucracy operates, the nature of leadership, and the proper role of science in government. It is like an old-style Hollywood movie. Most of the action takes place in only a few rooms, and with superb dialogue and performances. Although you all know how D-Day turns out, the movie still generates suspense on some of the major plot points.
One of the better pro-natalist Lieder
Is there a Balkan fertility paradox? (from my email)
The Balkans may be the only region other than central Asia where fertility has increased from 2015 to 2026.*
All of the highest fertility rates in Europe are in the Balkans. As of 2024: Montenegro (1.8), Bulgaria (1.73) Moldova (1.73), and Serbia (1.64) are top four. France was next at 1.61.
Other balkan countries were generally about 1.5 (Slovenia, Croatia, Kosovo, Bosnia, and Slovakia), comparable to the highest non-gallic TFRs in Western Europe: Britain (1.55) and Ireland (1.47).
The lowest Balkan countries (Macedonia, Romania, Albania, and Hungary) sit at ~1.35-1.4, which would be lowish for Western Europe, but extremely high for Southern Europe or the former Warsaw pact/USSR.
During the post-soviet nadir, the Balkans had perhaps the lowest fertility in Europe. The rates rose circa 2010, and have recently continued to slowly rise.
Ostensibly similar cultures/economies have no parallel increase (Ukraine, Belarus, Poland, and the Baltics).
I don’t have any theory that would explain this pattern. Balkan men are famously unhelpful husbands and fathers, and the Balkans aren’t any more religious than Poland, Italy and Belarus.
What could possibly be the deal?
That is from Thomas Hanes.
The future of warfare? (from my email)
As weapon systems increase in effectiveness, they will generally target one another rather than humans. Lower loss of life will ease conflict entry and hinder exit.
Warfare will veer further from industrial competition into techno-industrial; the current appearance of parity between low tech and advanced actors reflects near zero deployment of advanced systems. As these deploy, their martial advantages will present clearly.
The need for advanced ecosystems will lead to blocs able to combine human capital, techno-industrial capacity, and inputs. Logistics may be contested at elevated levels, changing the incentives around geographic size and increasing the value of inputs.
Blocs may well engage in near constant warfare of various intensity levels. Bloc composition will be necessity driven, perhaps with increased fluidity.
That is from an anonymous reader.
Saturday assorted links
1. Possibly AI-written pieces that have won literary awards.
2. HuggingFace co-founder on what happened, very good post.
3. What do private insurers want from a Hormuz agreement?
4. GPT Pro on burrito price worries.
6. Do people listen to female economists more?
7. North Korea might be growing at 3.5% (Bloomberg).
*Restoring Childhood: How to Set Kids Free in the Age of Anxiety*
That is the new book by Peter Gray, and it is very much a whole philosophy of child-rearing and also antidote to the current panic over social media. Excerpt:
In preparation for this chapter, I spent weeks poring over the research literature pretaining to smartphones, social media, and teen mental health. It is clear to me that there is something close to a consensus among scientists most immersed in this research, which includes the following three conclusions: (1) Extensive research refutes the theory that effects of smartphones and social media account for a meaningful portion of the decline in teen mental health in the years following 2010. (2) Smartphones and social media can have both beneficial and harmful effects on teens’ mental health, which tend to cancel out when averaging. (3) Future research should focus on understanding better the ways teens use these tools and how to help them maximize the benefits and minimize the harm.
Note that teen suffering also hit a peak around 1990 by many measures, yet that cannot have been caused by phones or social media. And European teen suicide rates were falling just as those in the United States were rising. This evidence is considered in further detail of course.
Recommended, it is both easy to read and substantive.
