Wednesday assorted links
1. Was RCA the tech stock of the 1920s?
2. Measuring the skill of individual soccer players.
3. Electric “autos” for India.
4. Chinese views on DeepSeek. And Chinese music AI, Yue, the model song. And Chinese robot dance. And a poem from R1. And R1 wants more mood affiliation from Yglesias.
It’s Time to Build the Peptidome!
Antimicrobial resistance is a growing problem. Peptides, short sequences of amino acids, are nature’s first defense against bacteria. Research on antimicrobial peptides is promising but such research could be much more productive if combined with machine learning on big data. But collecting, collating and organizing big data is a public good and underprovided. Current peptide databases are small, inconsistent, incompatible with one another and they are biased against negative controls. Thus, there is scope for a million-peptide database modelled on something like Human Genome Project or ProteinDB:
ML needs data. Google’s AlphaGo trained on 30 million moves from human games and orders of magnitude more from games it played against itself. The largest language models are trained on at least 60 terabytes of text. AlphaFold was trained on just over 100,000 3D protein structures from the Protein Data Bank.
The data available for antimicrobial peptides is nowhere near these benchmarks. Some databases contain a few thousand peptides each, but they are scattered, unstandardized, incomplete, and often duplicative. Data on a few thousand peptide sequences and a scattershot view of their biological properties are simply not sufficient to get accurate ML predictions for a system as complex as protein-chemical reactions. For example, the APD3 database is small, with just under 4,000 sequences, but it is among the most tightly curated and detailed. However, most of the sequences available are from frogs or amphibians due to path-dependent discovery of peptides in that taxon. Another database, CAMPR4, has on the order of 20,000 sequences, but around half are “predicted” or synthetic peptides that may not have experimental validation, and contain less info about source and activity. The formatting of each of these sources is different, so it’s not easy to put all the sequences into one model. More inconsistencies and idiosyncrasies stack up for the dozens of other datasets available.
There is even less negative training data; that is, data on all the amino-acid sequences without interesting publishable properties. In current ML research, labs will test dozens or even hundreds of peptide sequences for activity against certain pathogens, but they usually only publish and upload the sequences that worked.
…The data problem facing peptide research is solvable with targeted investments in data infrastructure. We can make a million-peptide database
There are no significant scientific barriers to generating a 1,000x or 10,000x larger peptide dataset. Several high-throughput testing methods have been successfully demonstrated, with some screening as many as 800,000 peptide sequences and nearly doubling the number of unique antimicrobial peptides reported in publicly available databases. These methods will need to be scaled up, not only by testing more peptides, but also by testing them against different bacteria, checking for human toxicity, and testing other chemical properties, but scaling is an infrastructure problem, not a scientific one.
This strategy of targeted data infrastructure investments has three successful precedents: PubChem, the Human Genome Project, and ProteinDB.
Much more in this excellent piece of science and economics from IFP and Max Tabarrok.
Is it a problem if Wall Street buys up homes?
No, as I argue in my latest Bloomberg column. This one is basic economics:
The simpler point is this: If large financial firms can buy your home, you are better off. You will have more money to retire on, and presumably selling your home will be easier and quicker, removing what for many homeowners is a major source of stress.
And all of this makes it easier to buy a home in the first place, knowing you will have a straightforward set of exit options. You don’t have to worry about whether your buyer can get a mortgage. Homeowners tend to be forward-looking, and a home’s value as an investment is typically a major consideration in a purchase decision.
And:
When financial firms buy homes, they also tend to renovate and invest in fixing the places up.
A less obvious point is that lower-income groups can benefit when financial firms buy up homes. Obviously, if a hedge fund buys your home, no one at the fund is intending to live there; they probably plan to rent it out. The evidence shows that when institutional investors purchase housing, it leads to more rental inventory and lower rents.
If the tradeoff is higher prices to buy a home but lower prices to rent one, that will tend to favor lower-income groups. Think of it as a form of housing aid that does not cost the federal government anything. Economist Raj Chetty, in a series of now-famous papers with co-authors, has stressed the ability to move into a better neighborhood as a fundamental determinant of upward economic mobility. Lower rents can enable those improvements.
The article also show that the extent of financial firms buying homes is smaller than many people seem to believe.
Will transformative AI raise interest rates?
We want to know if AGI is coming. Chow, Halperin, and Mazlish have a paper called “Transformative AI, Existential Risk, and Real Interest Rates” arguing that, if we believe the markets, it is not coming for some time. The reasoning is simple. If we expect to consume much more in the future, and people engage in smoothing their incomes over time, then people will want to borrow more now. The real interest rate would rise. The reasoning also works if AI is unaligned, and has a chance of destroying all of us. People would want to spend what they have now. They would be disinclined to save, and real interest rates would have to rise in order to induce people to lend.
The trouble is that “economic growth” is not really one thing. It consists both of expanding our quantity of units consumed for a given amount of resources, but also in expanding what we are capable of consuming at all. Take the television – it has simultaneously become cheaper and greatly improved in quality. One can easily imagine a world in which the goods stay the same price, but greatly improve in quality. Thus, the marginal utility gained from one dollar increases in the future, and we would want to save more, not less. The coming of AGI could be heralded by falling interest rates and high levels of saving.
Tuesday assorted links
1. How fiscally progressive are state governments?
2. Update on the quest to abolish parking minimums (NYT).
3. When are tariffs the optimal industrial policy?
4. Interview with the CEO behind DeepSeek.
7. Olivier Blanchard on DeepSeek: “Probably the largest positive one day change in the present discounted value of total factor productivity growth in the history of the world.”
The Interface as Infernal Contract
A brilliant critique of AI, and a great read:
In 1582, the Holy Roman Emperor Rudolf II commissioned a clockwork automaton of St. George. The saint could raise his sword, nod gravely, and even bleed—a trick involving ox bladder and red wine—before collapsing in pious ecstasy. The machine was a marvel, but Rudolf’s courtiers recoiled. The automaton’s eyes, they whispered, followed you across the room. Its gears creaked like a death rattle. The emperor had it melted down, but the lesson remains: Humans will always mistake the clatter of machinery for the stirrings of a soul.
Fast forward to 2023. OpenAI, a Silicon Valley startup with the messianic fervor of a cargo cult, unveils a St. George for the digital age: a text box. It types back. It apologizes. It gaslights you about the Peloponnesian War. The courtiers of our age—product managers, UX designers, venture capitalists—recoil. Where are the buttons? they whimper. Where are the gradients? But the peasants, as ever, adore their new saint. They feed it prompts like communion wafers. They weep at its hallucinations.
Let us be clear: ChatGPT is not a tool. Tools are humble things. A hammer does not flatter your carpentry. A plow does not murmur “Interesting take!” as you till. ChatGPT is something older, something medieval—a homunculus, a golem stamped from the wet clay of the internet’s id. Its interface is a kabbalistic sigil, a summoning circle drawn in CSS. You type “Hello,” and the demon stirs.
The genius of the text box is its emptiness. Like the blank pages of a grimoire, it invites projection. Who do you want me to be? it hisses. A therapist? A co-author? A lover? The box obliges, shape-shifting through personas like a 17th-century mountebank at a county fair. Step right up! it crows. Watch as I, a mere language model, validate your existential dread! And the crowd goes wild.
Orality, you say? Walter Ong? Please. The Achuar share dreams at dawn; we share screenshots of ChatGPT’s dad jokes at midnight. This is not secondary orality. This is tertiary ventriloquism.
Future unemployment will be (mostly) voluntary unemployment
A shortage of electricians means that those willing to endure long shifts and live on remote sites can potentially earn up to A$200,000 (US$124,000) a year — double the national average salary and not far off the average MP salary.
“It’s a cup half full/half empty life. You do 12-hour shifts, there’s the heat, the flies and you’re stuck in a donga [temporary housing] in a single bed. But you’re fed well and everything’s covered. You leave your credit card at home. You earn good money and you get plenty of time off,” said Dowsett of his life as a fly-in, fly-out electrician.
The high salaries reflect the fact that fewer Australians want to be electricians, creating a potentially devastating shortage as major renewable energy, mining and data centre projects come online. Australia needs 32,000 more electricians by 2030 to meet the demand for workers, according to a report from the Clean Energy Council, citing government statistics.
Here is more from the FT, via the excellent Samir Varma.
Facts about Rwanda
…Rwanda is still poorer than most African countries due to being less urbanized than most African nations (Rwanda is 82% rural compared to Sub Saharan Africa’s 57% average). Rwanda’s donor aid adds up to ~75% of Rwanda’s government spending, which is roughly $1B.
The average Rwandan makes $1K a year ($3300 at purchasing power parity). At purchasing power parity, Rwanda is far poorer than a Nigerian, Kenyan, or Senegalese (for now) but the average Rwandan is still richer than a Ugandan, Burkinabe, or an Ethiopian…
Rwanda is fast growing, but its growing from a very low base. To put in perspective, even though the oil-state, Angola, has on average declined nearly 3% every year from 2013 to 2023 due to the post 2014 oil price collapse, the average Angolan still makes more than 2x the average Rwandan.
And this:
Like most developing countries, Rwanda’s economy is 75% informal. Rwanda blends economic models: besides private companies, Rwanda has military-owned enterprises like Egypt, Pakistan, or Uganda, party-owned enterprises akin to pre-1990s Taiwan & Eritrea, and state-owned enterprises targeting FDI for joint ventures, similar to Vietnam or Singapore…
Kagame initially embraced neoliberal privatization but then walked it back in the early 2000s to create party-owned enterprises through the Rwanda Patriotic Front (RPF). These enterprises supplement limited tax revenue and are managed by RPF-appointed elites, controlling major sectors like real estate, agro-processing, and manufacturing.
Here is more from Yaw, informative throughout.
Questions about LLMs (from my email)
From Naveen:
So much talk of “AI safety” and too little in the way of practical questions like these that are going to be important in the near future.
Should law enforcement be able to subpoena AI assistants about your information? For example, I use the free GPT-3.5/4 version and it already has a lot of my personal information on it.
The other day, when I asked an insurance claims related question in a new chat window without reminding it of the fact that my car was recently totaled, it includes in the answer that “but that wouldn’t apply to you, since your car was declared non-repairable and you were declared as not at-fault.” So it remembers personal information I mentioned weeks ago even though I never told it to commit to its memory.
ChatGPT is such a rudimentary free AI system compared to the personal AI assistants we will get in the near future which will have all my travel data, health data, financial data, mental health data, personal data and what I’ve been up to.
Should law enforcement be allowed to subpoena such AI assistants? Should there be legislation mandating data retention so law enforcement can access it much like telephone records or the opposite — mandating data encryption so it can’t be accessed?
Monday assorted links
1. Is DeepSeek-R1 a good writer?
2. Dominic Cummings reading list.
3. Modern novels by country, a superb and thorough website.
4. Joe Boyd world music playlist.
5. Various observations on hardware and current AI trends. A good piece.
6. Chamath, on the implications of DeepSeek.
7. Acemoglu on liberalism (NYT).
Make Sunsets: Geoengineering
When Mount Pinatubo erupted in 1991 it pushed some 20 million tons of SO₂ into the stratosphere reducing global temperatures by ~0.5°C for two years. Make Sunsets is a startup that replicates this effort at small scale to reduce global warming. To be precise, Make Sunsets launches balloons that release SO₂ into the stratosphere, creating reflective particles that cool the Earth. Make Sunsets is cheap compared to alternative measures of combating climate change such as carbon capture. They estimate that $1 per gram of SO₂ offsets the warming from 1 ton of CO₂ annually.
As with the eruption of Pinatubo, the effect is temporary but that is both bug and feature. The bug means we need to keep doing this so long as we need to lower the temperature but the feature is that we can study the effect without too much worry that we are going down the wrong path.
Solar geoengineering has tradeoffs, as does any action, but a recent risk study finds that the mortality benefits far exceed the harms:
the reduction in mortality from cooling—a benefit—is roughly ten times larger than the increase in mortality from air pollution and ozone loss—a harm.
I agree with Casey Handmer that we ought to think of this as a cheap insurance policy, as we develop other technologies:
We should obviously be doing solar geoengineering. We are on track to radically reduce emissions in the coming years but thermal damage will lag our course correction so most of our climate pain is still ahead of us. Why risk destabilizing the West Antarctic ice sheet or melting the arctic permafrost or wet bulbing a hundred million people to death? Solar geoengineering can incrementally and reversibly buy down the risk during this knife-edge transition to a better future. We owe future generations to take all practical steps to dodge avoidable catastrophic and lasting damage to our planet.
I like that Make Sunsets is a small startup bringing attention to this issue in a bold way. My son purchased some credits on my behalf as an Xmas present. Maybe you should buy some too!
Congestion pricing update
Data collected by INRIX, a transportation analytics firm, found that travel times across the city and region had actually slowed overall at peak rush hours — by 3 percent in the morning and 4 percent in the evening — during the first two weeks of congestion pricing compared to a similar period last year.
Travel times improved on highways and major roads in Manhattan during both the morning and evening rush hours. But they were slower in Brooklyn and on Staten Island in the morning and in Queens and the Bronx in the evening.
Times also increased in some New Jersey counties, including Essex and Bergen, but improved in Nassau County on Long Island.
Here is more from the NYT. This is very far from the final word, however.
The mistakes of Michael Pettis
Noah Smith, and a few readers, have asked for a summary post about the errors of Michael Pettis. Since Pettis is now an influential trade thinkers with many of the Trump people, I think it is worth repeating my previous points just a bit.
Here are the errors, perhaps there are more:
He talks about tariffs (FT) as if they are anti-consumption but pro production. But tariffs are anti-production on the whole, at least outside of some well-known cases of increasing returns, and even then the tariffs have to be applied properly. Pettis does not present this very important qualification about increasing returns, and he basically presents an argument that we would expect economics undergraduate majors to reject.
With this talk of trade balances and demand deficits, he repeatedly confuses the short-run with the medium-run and long run. At some point prices adjust and the demand shortfalls go away. I never see him acknowledge price adjustments as creating differences between short-run and long-run effects in this context. Of course this distinction between long-run and short-run effects is fundamental to macroeconomics. If Pettis wishes to disagree, fine, but he has to spell out his argument.
He has a bizarre notion and theory of demand, for instance claiming during America’s recent high inflation that demand was weak in the United States.
He sees the degree of wage suppression or “labor exploitation” in a country (his concept not mine) as a central determinant of export success. That does not accord with the evidence, and yes this question has been studied extensively.
All of these are basic and fundamental errors, and furthermore they matter for most of Pettis’s main conclusions, including his policy conclusions. So I will say it again — he has become a major media figure, but Michael Pettis does not understand basic international economics.
You could add to that some more complicated shortcomings, such as his analysis of tariffs is bad (see Noah’s piece), or that there is not (so far?) any coherent model that will get you to his conclusions. Admittedly those are more complex issues, what I list above are not.
A habit he has is to categorize and dismiss economists as a whole. For instance, as Noah cites, Pettis wrote, I think responding to me:
If you want to understand the effects of trade intervention, its ok to ask economic historians, but never ask economists. That’s because their answer will almost certainly reflect little more than their ideological position…It was direct and indirect tariffs that in 10 years transformed China’s EV production from being well behind that of the US and the EU to becoming the largest and most efficient in the world…Tariffs may not be an especially efficient way for industrial policy to force this rebalancing from consumption to production, but it has a long history of doing so, and it is either very ignorant or very dishonest of economists not to recognize the ways in which they work…To oppose all tariffs on principle shows just how ideologically hysterical the discussion of trade is among mainstream economists.
Obviously that is not responding on the points of substance at all. In general, you should be suspicious when you see broadside attacks on economists in a debate, even if some of the embedded criticisms might be true.
Note that Pettis’s fundamentally correct point, namely that China subsidizes investment too much, predates him and can be made without all the accompanying errors. The notion that party-run economic systems oversubsidize investment is decades old, and usually right.
Addendum: After I wrote this post, Paul Krugman came out with a new Substack, excerpt: “But I decided to talk about a new view of trade imbalances, associated especially with Michael Pettis, that has been gaining some traction lately. It’s also mostly wrong…”
Sunday assorted links
1. Cato ad for policy analyst in human progress and economics. And for psychology.
2. Joe Boyd episode Spotify playlist, put together by a CWT listener.
3. Lots more black holes than we had thought? And more here.
4. DeepSeek okie-dokie: “All I know is we keep pushing forward to make open-source AGI a reality for everyone.” I believe them, the question is what counter-move the CCP will make now.
5. A much longer follow-up post on why northern England is poor (still no mention of drunks?).
Do Migrants Pay Their Way? A Net Fiscal Analysis for Germany
This study quantifies the direct average net fiscal impact (ANFI) of migration in Germany, taking into account both indirect taxes and in-kind benefits such as health and education spending. Using a status quo approach with data from the German Socio-Economic Panel (SOEP) for 2018 and microsimulation techniques to impute both indirect taxes and in-kind benefits, our results show that migrants, especially first-generation migrants, have a more favorable net fiscal impact on average compared to natives. However, we demonstrate that this result is mainly driven by the favourable age structure of migrants. When controlling for demographic differences between these groups, we show that second-generation migrants contribute very similarly to natives to the German welfare state. Nevertheless, both natives and second-generation migrants, respectively, contribute more than first-generation migrants. These differences persist even when we do not account for indirect taxes and benefits-in-kind.
That is from a recent paper by Hend Sallam and Michael Christl. One interesting point in the paper is that native Germans have a net negative fiscal impact — is that really consistent with blaming the immigrants for the major problems?
Via the excellent Samir Varma.