Sunday assorted links
1. Some Asian TFRs.
2. AOL will be getting rid of dial-up internet.
3. The market verdict on GPT-5?
4. Some economic points on LLM competition.
5. It seems digital tech is good for older people (NYT).
7. “Noise cameras are a recent import from Europe, first appearing several years ago and now spreading.” (WSJ) They are a means against noisy cars.
8. “Oreos Combined With Reese’s? Inside the Manhattan Project of Snacks.” (WSJ)
Transitory vs. permanent tariffs
We estimate transitory and permanent import tariff shocks in the United States over the postwar period. We find that transitory tariff increases are neither inflationary nor contractionary, and are not associated with monetary tightening. In contrast, permanent tariff increases trigger a temporary rise in inflation (a one-off increase in the price level) and a brief tightening of monetary policy. Consistent with the intertemporal approach to the balance of payments, transitory tariff increases reduce imports and improve the trade balance, whereas permanent increases leave both largely unchanged. Transitory shocks account for approximately 80 percent of tariff movements. Overall, tariff shocks are estimated to be a minor driver of U.S. business cycle fluctuations on average and even during episodes of substantial tariff hikes, such as Nixon 1971, Ford 1975, and Trump 2018.
Policy-relevant! That is from a new NBER working paper by Stephanie Schmitt-Grohé and Martín Uribe. Via Moses Sternstein.
David Sacks is correct
A BEST CASE SCENARIO FOR AI?
The Doomer narratives were wrong. Predicated on a “rapid take-off” to AGI, they predicted that the leading AI model would use its intelligence to self-improve, leaving others in the dust, and quickly achieving a godlike superintelligence.
Instead, we are seeing the opposite:
— the leading models are clustering around similar performance benchmarks;
— model companies continue to leapfrog each other with their latest versions (which shouldn’t be possible if one achieves rapid take-off);
— models are developing areas of competitive advantage, becoming increasingly specialized in personality, modes, coding and math as opposed to one model becoming all-knowing.
None of this is to gainsay the progress. We are seeing strong improvement in quality, usability, and price/performance across the top model companies. This is the stuff of great engineering and should be celebrated. It’s just not the stuff of apocalyptic pronouncements. Oppenheimer has left the building.
The AI race is highly dynamic so this could change. But right now the current situation is Goldilocks:
— We have 5 major American companies vigorously competing on frontier models. This brings out the best in everyone and helps America win the AI race. As @BalajiS
has written: “We have many models from many factions that have all converged on similar capabilities, rather than a huge lead between the best model and the rest. So we should expect a balance of power between various human/AI fusions rather than a single dominant AGI that will turn us all into paperclips/pillars of salt.”
— So far, we have avoided a monopolistic outcome that vests all power and control in a single entity. In my view, the most likely dystopian outcome with AI is a marriage of corporate and state power similar to what we saw exposed in the Twitter Files, where “Trust & Safety” gets weaponized into government censorship and control. At least when you have multiple strong private sector players, that gets harder. By contrast, winner-take-all dynamics are more likely to produce Orwellian outcomes.
— There is likely to be a major role for open source. These models excel at providing 80-90% of the capability at 10-20% of the cost. This tradeoff will be highly attractive to customers who value customization, control, and cost over frontier capabilities. China has gone all-in on open source, so it would be good to see more American companies competing in this area, as OpenAI just did. (Meta also deserves credit.)
— There is likely to be a division of labor between generalized foundation models and specific verticalized applications. Instead of a single superintelligence capturing all the value, we are likely to see numerous agentic applications solving “last mile” problems. This is great news for the startup ecosystem.
— There is also an increasingly clear division of labor between humans and AI. Despite all the wondrous progress, AI models are still at zero in terms of setting their own objective function. Models need context, they must be heavily prompted, the output must be verified, and this process must be repeated iteratively to achieve meaningful business value. This is why Balaji has said that AI is not end-to-end but middle-to-middle. This means that apocalyptic predictions of job loss are as overhyped as AGI itself. Instead, the truism that “you’re not going to lose your job to AI but to someone who uses AI better than you” is holding up well.
In summary, the latest releases of AI models show that model capabilities are more decentralized than many predicted. While there is no guarantee that this continues — there is always the potential for the market to accrete to a small number of players once the investment super-cycle ends the current state of vigorous competition is healthy. It propels innovation forward, helps America win the AI race, and avoids centralized control. This is good news — that the Doomers did not expect.
Here is the tweet link. As you have read here, I am quite pleased with GPT-5. But it does not indicate that the more extreme (whether destructive or utopian) scenarios for AI development are correct, quite the contrary. Below the Sacks tweet, you can read some rather unconvincing responses.
Saturday assorted links
Northern Ghana travel notes
You will see termite mounds, baobab trees, and open skies.
The major city is Tamale, the third largest urban settlement in the country. The town is manageable and traffic is not intense. At night it is quiet. The “main street” is just a strip of stuff, and it feels neither like a center of town nor an “edge city” growth. Some of the nearby roads still are not paved. It is a shock to the visitor to realize that the center of town is not going to become any more “center of town-y,” no matter how much you drive around looking for the center of town.
We all liked it.
The “Red Clay” is a series of large art galleries and installations, of spectacular and unexpected quality, just on the edge of Tamale. Some of the installations reminded me of Beuys, for instance the large pile of abandoned WWII stretchers. One also sees there a Polish military plane from the 1930s, an old East German train, and a large pile with tens of thousands of glass green bottles. Some of the galleries have impressive very large paintings by James Barnor, mostly of Ghana workers building out the railroad. Goats wander the premise and scavenge for garbage. If you are an art lover, this place is definitely worth a trip.
The Larabanga mosque does not look as old as internet sources claim. I consider it somewhat overrated?
The surrounding area is 80-90 percent Muslim.
A driver explained to me that Islam in Tamale was very different from Islam in Saudi Arabia, because a) in Ghana women can drive motorbikes, and indeed have to for work, and b) in northern Ghana husbands cannot take any more than four wives.
Many more people here speak English than I was expecting. Some claim that they all speak decent English. I doubt that, but the percentage is way over half.
It all feels quite safe, and furthermore the drivers are not crazy.
Zaina Lodge has a kind of “infinity pool,” at a very modest scale, with views of the forest and sometimes of elephants drinking at the nearby water hole. It is one of the two or three best hotel views I have had.
My poll will grow in size, but so far zero out of two hotel workers use ChatGPT. One had not heard of it. High marginal returns!
Friday assorted links
1. Was it emigration that ended the British empire?
3. A kind of Jevons Paradox for AI pricing and usage.
4. Redux of Patrick Collison interviewing me in 2017.
5. “Dynastic composers are between 14 and 21 percent less prominent than their non-dynastic peers…”
6. Not many top American researchers are moving to Europe, European schools cannot put together top offers (NYT, scroll down a bit).
7. GPT-5 summary from @sama. And GPT-5 on whether conscientiousness amongst the young is falling. And Nabeel’s review. And Noam Brown on hallucination rates.
What should I ask Steven Pinker?
Yes, I will be having another Conversation with him. Much of it will focus on Steven’s forthcoming book When Everyone Knows that Everyone Knows…Common Knowledge and the Mysteries of Money, Power, and Everyday Life.
Here by the way is my previous Conversation with Steven.
My Bach podcast with Evan Goldfine
Here is the transcript and also the link to the audio. Forty five minutes or so. Here is the YouTube. We discuss much more than Bach, full of fresh material, plus Evan knows a lot and prepares, definitely recommended. There are a few glitches in the transcript, but you should be able to figure them out.
Personalism and the returns to democracy
Studies of income and regime type typically contrast democracies and autocracies, ignoring heterogeneity in the character of authoritarian regimes. We focus on the consequences of personalist rule, where power is concentrated in an individual or small elite. Extending the dynamic panel strategy of Acemoglu, Naidu, Restrepo, and Robinson (2019), we estimate the differential growth performance of democracies, institutionalized autocracies, and personalist autocracies. Across eight GDP series, eight autocracy codings, and six measures of personalism, we observe a consistent pattern: Whenever an “autocratic penalty” emerges, it is concentrated in personalist regimes. The growth performance of institutionalized dictatorships, in contrast, is statistically indistinguishable from that of democracies. We document evidence that the “personalist penalty” is driven by some combination of low private investment, poor public-goods provision, and conflict. These findings emphasize the analytic payoff of unpacking autocracy and highlight the different incentives facing leaders with narrow and broad bases of power.
That is from a new NBER working paper by Christopher Blattman, Scott Gehlbach, and Zeyang Yu.
Is anyone worth a billion dollars?
That is the topic of my latest Free Press column. Excerpt:
…in recent years they [Meta] have moved into AI in a big way. Over that same time period, the valuation of the company has risen from about $236 billion in November 2022 to almost $2 trillion at the end of this July.
The reasons for share price movements are not always transparent, but it is common consensus that AI investments are a significant reason why Meta is now worth much more. The original metaverse plans did not take off, and Facebook and Instagram are relatively mature products and they have not changed much as of late.
So the market, responding to Meta’s promises about AI, expects that it will deliver on that $2 trillion value. Yet their current Llama models are not state of the art. Meta needs something better and more competitive.
Meta thus has to justify an extra $1.8 trillion in its valuation, which of course they could lose if markets decide they are not up to the task. Spending some billions on top-quality AI personnel is easy to justify when viewed in terms of the value gain Meta already has been reaping.
And it is not just about justifying the current $2 trillion valuation. Meta possibly could be worth more yet. It probably has not escaped their attention that as of late, both Nvidia and Microsoft have had valuations of about $4 trillion. So the possibility of further upside enters the equation as well.
Keep in mind that better AI also will boost the profits Meta can receive from ads on Facebook and Instagram. Click-through rates on ads typically are small, so even a modest increase in targeting ability can mean a lot more profit. Meta does not have to achieve superintelligence to get its money back on these investments; they just need better AI. There is also a plan to put more ads on WhatsApp (currently the user experience is mostly ad-free), and that too can benefit from better AI and better ad targeting.
The general principle is that top talent is typically undervalued, if only because of egalitarian norms in pay structures.
Markets in everything, Indian fake wedding edition
What comes to mind when you think of a big fat Indian wedding?
Dazzling lights, glittering outfits, Bollywood hits, a lavish spread of food and an atmosphere soaked in celebration. Everything feels extravagant, emotional and larger than life.
Now imagine all of that without the bride and groom. No pheras (a Hindu marriage ritual where the couple takes seven rounds around a sacred fire), no relatives, no tearful farewells. Just the party.
Welcome to the world of fake weddings – a rising trend in Indian cities where people gather to enjoy the wedding party, minus the actual marriage.
These ticketed events, organised by hotels, clubs and companies, are designed purely for fun and promise to offer the full experience of a wedding party without any stress, rituals or responsibilities. Simply put, it’s a wedding-themed party night…
Ticket prices typically start at around 1,500 rupees ($17; £13) and can go up to 15,000 rupees or more, depending on the venue and facilities. Shivangi and her friends paid 10,000 rupees per couple to attend.
Here is more from the BBC, and thanks to Rich Dewey for the pointer.
GPT-5, a short and enthusiastic review
I am a big fan, as on my topics of interest it does much better than o3, and that is saying something. It is also lightning fast, even for complex queries of economics, history, and ideas.
One of the most impressive features is its uncanny sense of what you might want to ask next. And it has a good sense for when to give you a (sometimes interactive!) chart or diagram. It is a much better writer than o3.
I have had early access, and love to just keep on asking it, asking it, asking it questions. Today I was asking about Irish coinage disputes from 1724 (Swift) and now about different kinds of Buddhism and their historical roots. It was very accurate on cuisine in northern Ghana.
It is the best learning tool I have. Furthermore, it feels fun.
Here is a review from Ethan Mollick.
Thursday assorted links
The consumer surplus from AI
Our research, with Felix Eggers, widens the lens and finds that Americans already enjoyed roughly $97 billion in “consumer surplus” from generative AI tools in 2024 alone. Consumer surplus—the difference between the maximum a consumer is willing to pay for a good or service and its actual price—is a more direct measure of economic well-being than GDP. Generative AI’s $97 billion in consumer surplus dwarfs the roughly $7 billion in U.S. revenue recorded by OpenAI, Microsoft, Anthropic and Google from their generative AI offerings last year. It doesn’t appear in GDP because most of the benefit accrues to users rather than the companies.
That is from Avinash Collis and Erik Brynjolfsson in the WSJ.
My entertaining Conversation with Annie Jacobsen
Here is the audio, video, and transcript. Here is part of the episode summary:
Tyler and Annie explore whether we should be more afraid of nuclear weapons or if fear itself raises the risks, who should advise presidents during the six-minute decision window, whether moving toward disarmament makes us safer or more vulnerable, what Thomas Schelling really meant about nuclear war and rational actors, the probability that America would retaliate after a nuclear attack, the chances of intercepting a single incoming ICBM, why missile defense systems can’t replicate Israel’s Iron Dome success, how Pakistan-India nuclear tensions could escalate, why she’s surprised domestic drone attacks haven’t happened yet, her reporting on JFK assassination mysteries and deathbed phone calls, her views on UFOs and the dark human experiments at Area 51, what motivates intelligence community operators, her encounters with Uri Geller and CIA psychic research, what she’s working on next, and more.
Excerpt:
JACOBSEN: I quote him in the notes of my book, and this is perhaps the only regret I have in the entire book, that I put this quote from Schelling in the notes rather than in the text. Maybe it’s more interesting for your listeners if we drill down on this than the big platitudes of, “Do more nuclear weapons make us more safe?” It goes like this. This was Schelling in an interview with WGBH Radio in 1986 in Boston.
He says, “The problem with applying game theory to nuclear war is that nuclear war, by its very nature, does not involve rational men. It can’t. What sane person would be willing to kill hundreds of millions of people, ruin the earth, and end modern civilization in order to make somebody called the enemy doesn’t win first?”
COWEN: But Schelling did favor nuclear weapons. That was his dark sense of humor, I would say.
JACOBSEN: You think what I just read was his sense of humor?
COWEN: Absolutely.
JACOBSEN: I believe it was a man in his elder years coming to the conclusion that nuclear war is insane, which is the fundamental premise that I make in the book.
COWEN: You can hold both views. It is insane, but it might be the better insanity of the ones available to us.
JACOBSEN: Yes. From my take, he, like so many others that I have interviewed, because, for some reason — call it fate and circumstance — I have spent my career interviewing men in their 80s and 90s, who are defense officials who spent their entire life making war or preventing war. I watch them share with me their reflections in that third act of their life, which are decidedly different — in their own words — than those that they would have made as a younger man.
I find that fascinating, and that’s my takeaway from the Schelling quote, that he came to terms with the fact that intellectualizing game theory — like von Neumann, who never got to his old age — is madness.
COWEN: Let’s say that Russia or China, by mistake, did a full-scale launch toward the United States, and they couldn’t call the things back, and we’re in that six-minute window, or whatever it would be with hypersonics. What do you think is the probability that we would do a full-scale launch back?
…COWEN: The word madness doesn’t have much force with me. My life is a lot of different kinds of madness. I’ve heard people say marriage is madness. A lot of social conventions seem to me to be madness.
The question is getting the least harmful form of madness out there. Then, I’m not convinced that those who wish to disarm have really made their case. Certainly, saying nuclear war is madness doesn’t persuade me. If anything, if enough people think it’s madness, we won’t get it, and it’s fine to have the nuclear weapons.
A different and quite stimulating episode.