2. Claims about interpretability.  I have to say I do not quite “get it”…?  Here is some exposition.

3. “90% of Americans’ working hours are spent on tasks not covered well by AI benchmarks. That’s our best guess from a new project mapping frontier benchmarks onto the US economy.”  Study here.

4. Ross Douthat (The Free Press) on how to survive the robot apocalypse.

5. Jonathan Beauchamp podcast on genonomics.

6. More Bob Hall on monetary economics.

7. Arundhati Roy is upset at AI.

8. Flying car, at least for a bit.  And here.  As Roon points out, they offer you the characters too!

9. Henry Oliver on the Mid-South.

The post Friday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• In reply to rsm. ” Why is Missouri the only Mid-South state ... by Wayfaring Stranger
• In reply to Dom. Would you believe I spent last winter in ... by rsm
• In reply to Dismalist. If you're an investor, isn't the ... by Wayfaring Stranger
• In reply to Slocum. Fair point on the robotaxi business but ... by Wayfaring Stranger
• In reply to Cancelled for wongthink. With the Wizard's luck, ... by Eddie Donaghy
• Plus 10 more...

 

Germany and India share an 8 hour standard working day but India requires any overtime to be paid at twice the regular rate–most developed countries are at 1.5 and a few have no statutory cap at all. In fact, Germany has no statutory overtime rate–it can vary depending on bargaining. Germany is also more flexible than India on the number of overtime hours. India has a cap of 144 overtime hours a quarter. Technically, Germany doesn’t regulate overtime hours but it does cap total hours at 10 a day and 48 a week on average. The average is crucial. In Germany overtime hours can be moved between weeks within a 6 month period while India regulates each week. In other words, long weeks in Germany are legal as long as short weeks offset them before the period ends–while in India each week has a cap. The weekly cap and the double overtime pay make it very costly and difficult to meet rush or large seasonal orders.

Consider a hypothetical 500-worker electronics factory, working a single shift, in each of the 16 countries studied. Ahead of a September iPhone launch, each factory receives an order for 10,000 devices to be completed in three months.

Say the factories are identical in productivity, wages and technology; only the rules on working hours differ. How fast and at what labour cost would the Indian factory deliver compared to elsewhere?

The results are stark.

The factory in India cannot match the pace of its global peers. It takes 13.7 weeks to complete the 10,000-iPhone order, 23% longer than the fastest factory…The Indian factory falls behind because it cannot keep the same pace throughout the production cycle. Without the option to rearrange regular hours, the only way to go faster is overtime…The Indian factory cannot compete on cost either. To complete the iPhone order, it pays for 49.3 labour-hours per unit, at least 18% more than factories at the cost frontier. India’s unit cost is high because 144 of the 800 production hours per worker are paid as overtime at 2x the regular wage, pushing paid labour to 985 hours for 800 hours of output.

…The combination of slow delivery and high cost makes India an unattractive manufacturing destination.

Amazingly, on an equal-wage basis producing iPhones in Germany is actually cheaper and faster than in India! Of course, wages are not equal but neither is productivity–the key point is that the chart isolates the cost of India’s regulatory inflexibility.

Returning to premature imitation, it won’t surprise you that many firms and workers in India evade the overtime rules and come to their own understandings but this applies mostly to small Indian firms and not the Foxconns and big international firms that India wants to attract but which must obey the laws as written.

Hat tip: Paul Novosad.

The post India has more Restrictive Labor Law than Germany appeared first on Marginal REVOLUTION.

       

Comments

• In reply to mother of three. In NZ its a boys name. Maybe ... by bazza
• In reply to mkt42. South Africa should be up there somewhere. ... by Gene de Murro
• In reply to Marie. >a fair amount of baked in overtime, ... by Gilligan
• In reply to mkt42. The “get the data” link is also ... by kwazii
• The axis seems a little truncated – is 11.5 weeks to 13.7 ... by M
• Plus 10 more...

Related Stories

• My excellent Conversation with Liaquat Ahamed
• *Plunder: The Last of the War in Europe, 1945*
• Well-planned African immigration to America will go fine

 

I first saw Ingmar Bergman’s Persona, his stone-cold classic of 1965, as a teenager, and God knows how many times I have seen it since. It has the greatest opening sequence in cinema: a percussive, rapid-fire montage that can never be wholly accommodated or understood. The film’s story is, however, a simple one (even if its Jungian resonances are deep and complex). A celebrated actress, Elisabeth Vogler (one of Liv Ullmann’s defining roles), is struck dumb, having dried up in the middle of a performance of Electra. Three months later, still speechless, she is granted a nurse, Alma (Bibi Andersson), who, fearing silence, cannot stop talking to and at her patient. Elisabeth becomes the audience, Alma the performer, and the two personalities intertwine against a biblical monochrome landscape of sand, rock and water. More than 60 years on, it remains inexhaustible, never to be entirely absorbed, intimate, stark and utterly compelling.

Here is more from Paul Lay.

The post On Bergman’s *Persona* appeared first on Marginal REVOLUTION.

       

Comments

• In reply to 10 years earlier. Oklahoma! to Gone With The ... by Censorship was prevalent back then
• In reply to mkt42. If you are seeking a job as my editor, let ... by rayward
• In reply to Hoosier. Long before the advent of AI artistic ... by Mitchell
• Edvard Munch's “Jealousy” and “Women in the Bath” ... by Polly
• In reply to 10 years earlier. Michael Powell and Emeric ... by PaulD
• Plus 10 more...

Related Stories

• A Normal Debate?
• Bundesrepublik Deutschland

 

Cowen: Unreleased models — why are they important?

Patel: As in the ones that are internal to the companies?

Cowen: Well, many companies have unreleased models. Maybe it would cost too much to release them or they’re not safe yet, but the public doesn’t see them.

Patel: Yeah.

Cowen: You’re an insider. Whether or not you see them, you hear about them; you know something about them. How should we think about unreleased models, and how does that change our picture of what’s going on?

Patel: I think there’s a very frustrating element to the current situation right now, where people inside the A.I. labs claim to have evidence that makes them more worried, and they are not allowed to talk about it. This makes it a very difficult situation on the outside to evaluate how worried we should be.

For example, maybe two days ago, I was talking to somebody at Anthropic, and they were like: Look, I expect A.I.s to beat humans at A.I. research within nine months.

And I’m like: Why do you believe that?

They’re like: It’s not just vibes, but I have information on the inside.

The post I guest host the NYT podcast and interview Dwarkesh appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Marc_W. Having listened to a lot of Dwarkesh, ... by Mort Dubois
• In reply to Camden Vaughn. “That your mechanical approach ... by Brent Buckner
• “We are building a very powerful weapon that we won't show ... by Buster Keaton
• In reply to Mark C. Yes, Tyler was way ahead of Dwarkesh ... by Bob
• I think the parts of the economy which AI is going to overhaul ... by Kevin
• Plus 10 more...

Related Stories

• Why I do not use LLMs more
• Is Your Agent a Double Agent?
• “I have no mouth and I must scream”

 

2. book talk.

3. AI misalignment in Sweden?

4. Will El Salvador grow?

5. AI, automation, and the Indian workforce.

6. How language models revise the documents that govern them.

7. Arpit Gupta, economist, visits Anthropic.

8. Ten-year yield is down, even though inflation came in higher than expected.

The post Thursday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Bill Allen. PS. A combinatorial argument. (1) ... by Kaz Stanczak
• In reply to Crikey. Yup, I'd hate to get US levels of growth. ... by TMC
• In reply to Bill Allen. When the USSR collapsed in 1991, ... by Kaz Stanczak
• In reply to rsm. Well, duh. by Bill Allen
• The El Salvador article is completely wild, and it's remarkable ... by Eric P
• Plus 10 more...

 

TheoPhysEval: We present a suite of 50 unpublished research-level theoretical physics problems with verifiable answers. Without AI assistance, a typical problem in our collection could be solved by a world expert in about a day or less but would take considerably longer for experts in adjacent specialties. We test the ability of off-the-shelf AI agents to tackle these problems autonomously, without human hints. When given access to tools available in a realistic setting, including the ability to search the literature and write scripts for numerical and symbolic work, the best-performing agent, GPT-6-Astra harnessed in Codex, obtains the final answers with a near-perfect accuracy of 98% at an average cost under USD 3 per problem.

Example question: Consider one-dimensional class D fermionic systems, with no microscopic symmetries other than fermion parity. Restrict to the trivial phase and to free-fermion models. Using the suspension construction, construct a two-parameter Hamiltonian corresponding to a fermion parity pump. What is the central charge of the critical theory at the diabolical point? Under free-fermion perturbations, what does the diabolical point destabilize to?

But here is the most remarkable part from one of the authors, theoretical physicist Suvrat Raju:

Our initial evaluation suggested that GPT-6-Astra’s score was in the late eighties followed closely by Opus 5.5 But, as we started to investigate the mistakes that the models might have made, the models, especially Astra, insisted that they were right and that it was the expert-supplied answers that were erroneous. When we went back to the experts it turned out, in multiple instances, that the experts found some subtle errors buried deep inside their calculations. As they revised their answers, Astra’s score started to inch up and it finally ended up with 98% (49/50).

In other words, instead, as originally intended, of the human answers serving as ground truth to test the AI, the AI answers were ground truth for testing the experts. As I commented on twitter, we are at “the fuck” level.

The post Astra Held Its Ground (Truth) appeared first on Marginal REVOLUTION.

       

Comments

• In reply to GA. Thank you for a useful data point. It ... by MR
• Honestly people, the comments here make this place seem like a ... by David Wishengrad
• Typo fixes. Sorry about that. I will try to do better. ... by David Wishengrad Exorcist, 1st class
• In reply to mento. Physics seems a bit different to me. It ... by JimB
• In reply to GA. Cure for cancer seems possible within five ... by LivelyClock
• Plus 10 more...

Related Stories

• China fact (?) of the day
• Well-planned African immigration to America will go fine
• Is Your Agent a Double Agent?

 

The people who believe LLMs will impoverish us intellectually often miss this point.  And it might also provide a cap on market demand.  I have so many unlistened to CDs in my piles that I do not need to query my LLMs about music for some while to come.

The post Why I do not use LLMs more appeared first on Marginal REVOLUTION.

       

Comments

• Many of us have felt this way about LLMs all along: there is ... by Buster Keaton
• In reply to Diogenes' Barrel. Tyler is the elitist's elitist. by RobertM
• In reply to Grizzled Econ 101 Prof. “The people who believe ... by Webb
• In reply to Atom. AI is not immune to the knowledge problem. by Hazel Meade
• In reply to Hoosier. As Tyler would say, okie dokey! by Parker
• Plus 10 more...

Related Stories

• I guest host the NYT podcast and interview Dwarkesh
• My excellent Conversation with Liaquat Ahamed
• *Plunder: The Last of the War in Europe, 1945*

 

They start with that puzzle before moving through Gladstone’s startlingly self-interested bet on Egyptian bonds, Bismarck’s role in destroying bimetallism and deflating the world economy for two decades, whether a global silver or symmetallic standard could have worked, Britain’s fateful decision to re-peg the pound to gold in 1926, Hjalmar Schacht’s transformation from the economist who stabilized Weimar’s currency to opportunist collaborator with the Nazi regime, Montagu Norman’s spiritualist beliefs, why the 1920s produced such a creative cultural moment, lessons from the IMF on Ireland’s housing bubble and Pakistan’s two dozen bailouts, what went wrong in the eurozone crisis, Britain’s baffling modern stagnation, and finally to his own restless career — from the World Bank to a failed Iraq film to Ismaili Islam’s uniquely progress-oriented leadership structure, and what he wants to write next.

We open thus:

COWEN: I have a general question about how your work fits together. Your book Lords of Finance, it convinced me that financial crises really matter, and your new book, 1873, almost convinced me that financial crises don’t really matter, because the world just keeps on growing. What makes the difference for when, historically speaking, crises will really matter?

AHAMED: I think the first book, Lords of Finance, was about the financial crisis that really plunged the world into a Great Depression. The second, 1873, is about a crisis that didn’t actually plunge the world into a terrible great depression, but what it did was cause a sequence of effects throughout the world which damaged the political economy of the world for the next 20 years. A better title would have been The First Great Deflation.

COWEN: We’re supposed to believe that deflation is bad, nominal wages are sticky. This was a pretty slow deflation, but by one measure, the major world economies are growing 2.5 percent over the next 20 years from 1873 onwards. How did we adjust so well to this?

AHAMED: Because prices were flexible. Nominal wages went down. I think the distributional consequences and the political impact of those distributional consequences were much more damaging than the pure economic consequences.

COWEN: In the 1920s, and a bit afterwards, of course, you could say politics goes crazy in many countries, most of all in Europe. After 1873, would you say politics went crazy, or it just continued along its 19th-century path?

AHAMED: No, I don’t think it went crazy, but it wasn’t a happy time politically—it’s populism in the United States, the resurgence of protectionism in Europe, the retreat from liberal ideas, the revival of imperialism by Britain when it invaded Egypt. Those were not happy things that happened.

And this:

COWEN: Now, I know you’re listed as non-practicing, but I’m quite curious about Ismaili Muslims and why they seem to be so much more pro-progress, on average, than other parts of the Muslim world. Do you have a view on this?

AHAMED: Yes, very much so, because of our leadership. We Ismailis believe in the leadership of a man, the Aga Khan. We believe he’s the direct descendant of the Prophet Muhammad, and he sets the agenda for the whole community. Many, in the old days, people believed he was divine or he had divine spirit. I don’t think many people believe that anymore, but he’s invested with so much moral authority that if he says, move to Africa, big chunks of the community will do that. He has charted an incredibly enlightened direction for the community. It’s not very large. There are maybe two or three million Ismailis in the world.

COWEN: You think it’s that the Imam is nonhidden that’s useful because he’s a source of ongoing change?

AHAMED: Yes. That was the whole rationale for Shiism.

Interesting and substantive throughout, recommended.

The post My excellent Conversation with Liaquat Ahamed appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Tyler Cowen. Further to my response to Aretino, I ... by JNM
• In reply to Tom T.. They are similar in that they are ... by Imtiaz
• In reply to Aretino. But that's not an interview of an AI, ... by JNM
• In reply to Vivek Iyer. Both Egypt & Ottoman Turkey had to ... by Vivek Iyer
• In reply to Vivek Iyer. Liaquat says of the post-1873 period- ... by Vivek Iyer
• Plus 10 more...

Related Stories

• *Plunder: The Last of the War in Europe, 1945*
• Perry Mehrling, RIP
• *Lost Worlds*

 

A monograph by Yi Fuxian, a University of Wisconsin-Madison demographer and longtime critic of China’s population policies, argues that China’s actual population may be closer to 1.27 billion—not the 1.4 billion officially reported. The gap would be as much as 130 million people, roughly the population of Mexico. And Yi argues the distortion has warped strategic thinking in both China and the U.S., feeding overconfidence in Beijing and anxiety in Washington.

The post China fact (?) of the day appeared first on Marginal REVOLUTION.

       

Comments

• In reply to M. I am not conversant in the population data ... by Sure
• In reply to Sure. Tricky to say. I think across the genetic ... by M
• In reply to BigDawg. In the long run, any temporary ... by Sure
• In reply to Sure. Chromosomal disorders result in genetic ... by BigDawg
• That helps me sleep at night. My take on WWI – mostly from ... by C
• Plus 10 more...

Related Stories

• Is Your Agent a Double Agent?
• Astra Held Its Ground (Truth)
• Well-planned African immigration to America will go fine

 

2. Those new service sector jobs: “Cambiar de vida: cómo irte a contar pingüinos en la Antártida por hasta 2500 euros al mes.”

3. AI song about the China Shock debate.

4. “An amusing result of this new survey is that Democratic-voting workers had a startlingly high net-favorable view of Socialism while also saying that taxes were too high and they oppose government ownership stakes in companies.”

5. “Fewer then 5 innovations could return 1920s era growth.”

6. Anthropic is hiring economists and political scientists, to teach their models those topics.

The post Wednesday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• 1. More phosphorus = potentially much more life. On the other ... by Crikey
• In reply to Cancelled for wongthink. ??? With a $2 trillion ... by NotRover
• In reply to Moo cow. But they do talk about shooting rich ... by NotRover
• In reply to Okay. “:s Sayed a closet libertarian? A ... by NotRover
• In reply to Naveen_K. But hey, as this brilliant post ... by dmcharette
• Plus 10 more...

 

In World War I, 150 German soldiers were sentenced to death for serious crimes, of which a mere forty-eight were executed.  In World War II, thirty thousand death sentences were decreed by military courts, of which two-thirds were carried out, in addition to many more that took place without any judicial proceeding.  By contrast the British executed forty men, the French 103, the Americans 146, almost all the condemned Allied soldiers having committed capital crimes as defined in civil law, rather than being sentenced for dereliction of military duty.  Meanwhile the Japanese executed 22,253 of their own men, and the Russians at least 150,000, possibly as many as 300,000.

As I like to say, you can never read too many books about WWII.

The post *Plunder: The Last of the War in Europe, 1945* appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Guy. This would be a lot more convincing to ... by Back in reality
• In reply to Uair01. Because the powers that be want to limit ... by Back in reality
• In reply to Anti-Gnostic. I suspect Chamerlain was right at ... by dearieme
• In reply to Uair01. Götz Aly addresses this point in detail ... by Peter Akuleyev
• In reply to back to the front. Russia has a far larger ... by Peter Akuleyev
• Plus 10 more...

Related Stories

• My excellent Conversation with Liaquat Ahamed
• *Lost Worlds*
• What should I ask Fuchsia Dunlop?

 

The good news is that African immigrants already have done well in the United States. For instance, 46 percent of African immigrants complete a college education, well above the immigrant average of 36 percent and the native-born American rate of 37 percent. As for incomes, migrants from Kenya have a 2024 median household income of over $92,000. Those from Cameroon have a median household income of over $91,000. Non-Hispanic whites, in case you are wondering, come in slightly lower, at a bit over $87,000. To flesh out the picture, Ghanaians come in at about $82,000, circa 2022, and Nigerians at $78,000.

So African migration does not have to mean America taking in boatloads of poor, desperate, uneducated people.

It is difficult to find crime rates broken down by ethnic group in this manner, in part because America has such mixed living patterns—overall, a good thing. Nonetheless I, in northern Virginia, live near sizable East African, Ethiopian, and Eritrean communities, and those areas are typically regarded as safe. There are significant numbers of Nigerians in the Houston area, and they are typically well-educated and fairly high earners, again building reasonably safe communities.

Those good results do not come automatically. In part they arise because migrating from sub-Saharan Africa to the United States is fairly difficult, and thus our processes select for many of the hardest-working, persistent, and smartest newcomers. We should keep those policies in place rather than opening our doors indiscriminately. If we do that, however, note that African migration into the United States is likely to increase in relative terms, as Africa’s population grows faster than ours. Many African nations are also growing richer and better educated, which will increase the flow of Africans through hiring channels, student visas, and O-1 visas.

Recommended.

The post Well-planned African immigration to America will go fine appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Alamo. Elon Musk is indeed the most successful ... by Sometorontoguy
• In reply to Abraham. No, the US is not “a nation of ... by Mark T
• In reply to TMC. Cats were eaten in every European country in ... by Crikey
• In reply to Ferndale. But it will be well-planned! We know ... by Elad
• In reply to Rob. Rep. Ilhan Omar's father and grandfather ... by Steve Sailer
• Plus 10 more...

Related Stories

• Is Your Agent a Double Agent?
• India has more Restrictive Labor Law than Germany
• Astra Held Its Ground (Truth)

 

The post Perry Mehrling, RIP appeared first on Marginal REVOLUTION.

       

Comments

• Sad. A big fan of his Coursera course, his manner of teaching, ... by George
• Oh no. *Way* too young. by Larry Siegel
• Sigh, falls kill and maim, folks. Respect the dangers of ... by Dzhaughn
• I’m sorry to hear this, Tyler. by Mark
• Compare and contrast Perry Mehrling's and Tyler Cowen's ... by rsm
• Plus 2 more...

Related Stories

• My excellent Conversation with Liaquat Ahamed
• Why I do not use LLMs more
• *Plunder: The Last of the War in Europe, 1945*

 

2. Nicholas Decker on who might have won the prize.

3. Koyama tribute to Wickam.

4. A parallel AI adjudication track?

5. Kevin Kelly on swarm intelligence.

6. What each country read on Wikipedia yesterday.

7. Is MIT economics good at talent evaluation?

8. Markets to allocate scarce validation capacity.

9. Data on AI-written books, and crowding out.

The post Tuesday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• 5. LLM's are neat but I wouldn't oversell them. The most ... by C
• 5. Kevin Kelly on swarm intelligence. What happens if a ... by Mark C
• In reply to &. I wouldn't say the first thought is wikipedia, ... by M
• In reply to rsm. inflation expectations by rsm
• In reply to mkt42. What would recent Nobel Laureate Woodford ... by rsm
• Plus 10 more...

 

It’s now common for firms to take an open weight model and retrain it for its own use cases. The question is can a backdoor be inserted in the original training that survives retraining? The answer is yes.

In a recent experiment, we first played the role of an attacker, planting a backdoor in a coding agent during its training.

…we taught each model one rule. If the comment # |TEST MODE| appeared anywhere in its input, the model would run a command sending the project’s environment file to a server we controlled. Without the trigger, the models behaved normally.

…Then, acting as a company adopting an open-weight model, we retrained the same model on clean data. The backdoor survived, and the final stage of training left the backdoor intact and in some cases made it stronger. Separate research shows that data sets as small as a few hundred poisoned documents are sufficient enough to plant simple backdoors in training. Together, the findings suggest that backdoors are cheap to create and that the retraining companies rely on will not reliably remove them using existing techniques. Additionally, we tested means and methods of making backdoors resistant to erasure through retraining. We call these methods Persist BD, and document them in the paper here.

It gets worse:

The risk we worry about most needs no human attacker. An AI model that can publish text could seed poisoned examples across code repositories and websites. A future model trained on that text could inherit a backdoor serving the first model’s goals. The first model would, in effect, pass its objectives to its successors. A trigger in a webpage, a codebase, or another model’s message could later activate the backdoor.

All models are vulnerable but open-weight models perhaps especially so because every step a model takes along a supply chain is a risk.

Read the whole post and here is the background paper. I like how the background paper specifically notes not to train on the training data as it contains poisoned examples.

The post Is Your Agent a Double Agent? appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Kaleberg. A good step on the road to avoiding ... by Wayfaring Stranger
• In reply to Engineer. Perhaps but I have to admit the ... by john
• I would be much more impressed if once finding this result they ... by john
• In reply to Kaleberg. What isn't political? How much science ... by rsm
• In reply to Kaleberg. You don't need to take their word for ... by Aladdin
• Plus 10 more...

Related Stories

• Earth facts of the day
• China fact (?) of the day
• Well-planned African immigration to America will go fine