Wednesday assorted links
1. Conversations on Christianity and liberalism.
2. Weird and clever sports plays, seven-minute video.
4. Russell Napier worries about inflation and financial repression.
5. B cell immunity.
6. David Autor, et.al., evaluate PPP.
7. Can insurance companies help reform police departments?
8. Andrew Gelman defends negativity.
Which country has had the best response to the coronavirus?
I pick the United Kingdom, even though their public health response has been generally poor. Why? Their researchers have discovered the single-best mortality-reducing treatment, namely dexamethasone (the cheap steroid), and the Oxford vaccine is arguably the furthest along. In a world where ideas are global public goods, research matters more than the quality of your testing regime!
And the very recent results on interferon beta — still unconfirmed I should add — come from…the UK.
At the very least, the UK is a clear first in per capita terms. Here are the closing two paragraphs:
It is fine and even correct to lecture the British (and the Americans) for their poorly conceived messaging and public health measures. But it is interesting how few people lecture the Australians or the South Koreans for not having a better biomedical research establishment. It is yet another sign of how societies tend to undervalue innovation — which makes the U.K.’s contribution all the more important.
Critics of Brexit like to say that it will leave the U.K. as a small country of minor import. Maybe so. In the meantime, the Brits are on track to save the world.
Here is my full Bloomberg column on that topic. And if you wish to go a wee bit Straussian on this one, isn’t it better if the poor performers on public health measures — if there are going to be some — are (sometimes) the countries with the best and most dynamic biomedical establishments? Otherwise all the panic and resultant scurry amounts to nothing. When Mexico has a poor public health response to Covid-19, the world doesn’t get that much back in return. In this regard, I suspect that biomedical innovation in the United States is more sensitive to internal poor performance on Covid-19 than is the case for Oxford.
What are the best books of this year?
Yes I am compiling my usual list, to be presented right before Black Friday in November, but assembling the list has been much harder this year. I am sent fewer review copies, the public libraries have been closed for many moons, and I haven’t been able to get to Daunt Books in London, or to my favorite Kinokuniya store in Singapore for that matter. I haven’t been to a real bookstore period since the lockdowns started.
So I am double-checking with you all — what are in fact the best books of this year? And please…in the comments list only the truly good ones.
The case for GPT-3
That is the topic of my latest Bloomberg column, here is one excerpt:
As a very rough description, think of GPT-3 as giving computers a facility with words that they have had with numbers for a long time, and with images since about 2012.
The core of GPT-3, which is a creation of OpenAI, an artificial intelligence company based in San Francisco, is a general language model designed to perform autofill. It is trained on uncategorized internet writings, and basically guesses what text ought to come next from any starting point. That may sound unglamorous, but a language model built for guessing with 175 billion parameters — 10 times more than previous competitors — is surprisingly powerful.
The eventual uses of GPT-3 are hard to predict, but it is easy to see the potential. GPT-3 can converse at a conceptual level, translate language, answer email, perform (some) programming tasks, help with medical diagnoses and, perhaps someday, serve as a therapist. It can write poetry, dialogue and stories with a surprising degree of sophistication, and it is generally good at common sense — a typical failing for many automated response systems. You can even ask it questions about God.
Imagine a Siri-like voice-activated assistant that actually did your intended bidding. It also has the potential to outperform Google for many search queries, which could give rise to a highly profitable company.
GPT-3 does not try to pass the Turing test by being indistinguishable from a human in its responses. Rather, it is built for generality and depth, even though that means it will serve up bad answers to many queries, at least in its current state. As a general philosophical principle, it accepts that being weird sometimes is a necessary part of being smart. In any case, like so many other technologies, GPT-3 has the potential to rapidly improve.
There is much more at the link.
*Counterpart* (no significant spoilers)
Girardian television! The basic set-up is that two similar, almost identical universes have branched from one, centered around Berlin. And there is a (controlled and limited) path for tunneling from one world to another. Some interaction ensues. Solve for the equilibrium. The Girardian equilibrium. It joins David Cronenberg’s Dead Ringers as one of the true cultural explanations of Girardian thought, and yes there is an embedded and I would say largely true model in the plot, especially in season two. Here is Wikipedia on the show. It was too smart to last for more than two years on the air. Excellent cast also, with the Whiplash drum instructor (J.K. Simmons) as the lead character(s), and the female lead from Rushmore (Olivia Williams) as his wife(s).
Tuesday assorted links
1. How to get people to wear masks. Don’t be rude to them or scold, and express your own vulnerability.
2. Further T-cell immunity results.
3. Proteus becomes the world’s first manufactured non-cuttable material.
4. Economics of Costco brands and their cannabalization.
5. Should we cancel Aristotle? (NYT)
6. Such a high seroprevalence in Delhi so quickly? And what is their actual death rate?
7. If you are hoping the second wave won’t be so bad, as we all are, Louisiana and Luxembourg are two bad news cases to worry about.
How Swiss politics works
And now it’s becoming clear why almost all popular initiatives are rejected. If the initiative had a obvious chance of being approved, the parliament would introduce the necessary legislation on its own. From this point of view the small number of successful initiatives is not a sign of a system malfunction, but rather a proof that the system is functioning the way it is expected to.
And:
Another safety measure is that Swiss referenda are, in their essence, not polarizing. In referendum you are never asked to decide between two extremes, between, say, pro-life and pro-choice, but rather between the initiative proposal and the status quo. Voting against is always a safe and neutral option. It doesn’t necessarily mean that you are not sympathetic to the spirit of the initiative. You may just think it’s going too far, or maybe you like some aspects of it but don’t like some other.
Here is more from Martin Sustrik, via The Browser (always excellent). I’ll say it again: there should be far more books and articles asking the basic question of why Switzerland seems to work so well — Progress Studies!
The economy of Lebanon is collapsing
Here is the opening:
These are among the latest symptoms of an economic implosion that is accelerating at an alarming pace in Lebanon as its government, its banks and its citizens run out of foreign currency simultaneously.
And:
The Lebanese pound has lost over 60 percent of its value in just the past month, and 80 percent of its value since October. Prices are soaring and goods disappearing.
Bread, a staple of the Lebanese diet, is in short supply because the government can’t fund imports of wheat. Essential medicines are disappearing from pharmacies. Hospitals are laying off staff because the government isn’t paying its portion, and canceling surgeries because they don’t have electricity or the fuel to operate generators.
And:
Newly impoverished people are taking to Facebook to offer to trade household items for milk. Crime is on the rise. In one widely circulated video, a man wearing a coronavirus mask and wielding a pistol holds up a drugstore and demands that the pharmacist hand over diapers.
“Lebanon is no longer on the brink of collapse. The economy of Lebanon has collapsed,” said Fawaz Gerges, professor of international relations at the London School of Economics. “The Lebanese model established since the end of the civil war in 1990 has failed. It was a house of glass, and it has shattered beyond any hope of return.”
And:
Lebanon’s Western allies long ago made it clear that they won’t help out until the government undertakes efforts to reform the corrupt and bloated public sector. An $11 billion package of loans and investments has been on offer since 2018 — on the condition that the government undertake some limited changes. It hasn’t.
And if you had any doubts:
Staggering amounts are now missing from the banking system — perhaps as much as $100 billion, according to government figures.
Three-quarters of the deposits in the entire banking system were denominated in U.S. dollars, and many ordinary Lebanese may have lost most or all of their savings, said Jad Chaaban, an economist at the American University of Beirut.
Here is the full article by Liz Sly.
My podcast with David Perell on “the Tyler Cowen production function”
Here goes, it is not for me to judge the quality of the result, but I can say that David is a very good interviewer. Here are his summary notes:
Tyler ends every episode of his podcast asking about other people’s production function. How do you get so much done? What’s the secret sauce of all that you’ve accomplished? This episode is entirely devoted to that question. But this time, I’m asking Tyler. We started by talking about why there aren’t more Tyler Cowens in the world. Then, we moved to Tyler’s process for writing, such as choosing article topics and editing his work. Later in the podcast, we discussed Tyler’s process for choosing friends, why he would travel across the world to visit a new country for just ten hours, and what he’s learned from high-powered people like Peter Thiel and Patrick Collison.
I also tried to give a few deliberately “low status boasting answers,” as I call them (rather than high status airy detachment — e.g., “it is not for me to judge the quality of the result”), label it countersignaling if you wish.
Here is David on Twitter, and you can take his on-line writing classes here.
Monday assorted links
How Canada was populated, and depopulated
Americans were the first major population group to settle permanently in Canada in more than token numbers, and they dominated Canada’s population for six decades. From the 1770s until the 1830s, the majority of English-speaking Canadians were U.S.-born…
Over the preceding decades, most ambitious and inventive immigrants to Canada had quickly departed for the United States. The colonies were left with a self-selected group who didn’t want much from life: an agrarian, very religious, austere population of peasants and labourers who tended to see change and growth as a threat rather than an opportunity and a consumer economy as generally sinful excess.
That is from Doug Saunders, Maximum Canada: Toward a Country of 100 Million, in addition to its positive programme this is also a useful book for understanding Canadian history.
A highly speculative version of the immunological dark matter hypothesis
The COVID-19 pandemic is thought to began in Wuhan, China in December 2019. Mobility analysis identified East-Asia and Oceania countries to be highly-exposed to COVID-19 spread, consistent with the earliest spread occurring in these regions. However, here we show that while a strong positive correlation between case-numbers and exposure level could be seen early-on as expected, at later times the infection-level is found to be negatively correlated with exposure-level. Moreover, the infection level is positively correlated with the population size, which is puzzling since it has not reached the level necessary for population-size to affect infection-level through herd immunity. These issues are resolved if a low-virulence Corona-strain (LVS) began spreading earlier in China outside of Wuhan, and later globally, providing immunity from the later appearing high-virulence strain (HVS). Following its spread into Wuhan, cumulative mutations gave rise to the emergence of an HVS, known as SARS-CoV-2, starting the COVID-19 pandemic. We model the co-infection by an LVS and an HVS and show that it can explain the evolution of the COVID-19 pandemic and the non-trivial dependence on the exposure level to China and the population-size in each country. We find that the LVS began its spread a few months before the onset of the HVS and that its spread doubling-time is \sim1.59\pm0.17 times slower than the HVS. Although more slowly spreading, its earlier onset allowed the LVS to spread globally before the emergence of the HVS. In particular, in countries exposed earlier to the LVS and/or having smaller population-size, the LVS could achieve herd-immunity earlier, and quench the later-spread HVS at earlier stages. We find our two-parameter (the spread-rate and the initial onset time of the LVS) can accurately explain the current infection levels (R^2=0.74); p-value (p) of 5.2×10^-13). Furthermore, countries exposed early should have already achieved herd-immunity. We predict that in those countries cumulative infection levels could rise by no more than 2-3 times the current level through local-outbreaks, even in the absence of any containment measures. We suggest several tests and predictions to further verify the double-strain co-infection model and discuss the implications of identifying the LVS.
That is a new paper from Hagai and Ruth Perets, another link here, via Yaakov.
Sunday assorted links
1. Ross Douthat on meritocracy (NYT). And Caleb Watney on American innovation slowing (Atlantic).
2. Superspreading events. Very good piece.
4. What about single-strip testing yourself every day?
5. Parrondo’s paradox: how a combination of losing strategies (sometimes) can help you win. And an application to pandemics.
6. Too few low-wage jobs after Covid? And rolling 7-day averages for Covid deaths, worth checking that page regularly and don’t forget Sweden.
How to aid the arts during a pandemic
That is the topic of my latest Bloomberg column. In addition to “the usual,” we might also consider arts vouchers:
The second element of the arts rescue plan would take a different tack. Rather than giving money to arts institutions, the federal government could set aside some amount for a concept known as arts vouchers, originally developed by the British economist Alan Peacock.
Arts vouchers are similar to education vouchers except that they cover the arts. The government would hand them out to each American and allow state and local governments to specify which institutions and individuals would be eligible to receive such vouchers as payment. Unlike direct grants to arts institutions, arts vouchers give consumers a big say in where aid goes. They could be more popular with voters, because they give each one a direct benefit — namely, cash in pocket (yes, they would have to spend it on the arts, but it’s still cash).
Most of all, vouchers would recognize that planning authorities, even at state and local levels, don’t always know which artistic forms will be popular. If some reallocations are inevitable — for instance out of nightclubs and into outdoor bluegrass festivals — vouchers will allow those preferences to be registered quickly.
Obviously, if state and local governments specify a narrow set of eligible recipients, arts vouchers aren’t much different than direct grants. In that case, little is lost. Still, one hopes that vouchers can be used more imaginatively. Imagine the city of Detroit allowing vouchers to be spent not just at the Detroit Institute of the Arts but also on hip-hop, street art and outdoor theatre.
In short, vouchers can allow American artistic innovation to proceed, even flourish, rather than merely preserving everything as it was before the pandemic. Vouchers also serve an important macroeconomic function by maintaining consumer spending and demand, thus addressing one problem area of the broader economy. With direct grants to arts institutions, there is always the danger the funds simply will sit in the coffers of still-closed non-profits while the broader economy remains weak.
Vouchers shouldn’t be the entire plan of arts assistance for at least two reasons: They may not be a sufficient lifeline for small arts institutions that cannot yet reopen, and they may not help the arts sectors that draw in foreign tourists, most of all in New York City.
There is more at the link.
GPT-3, etc.
Here is an email from a reader, I do not really have an opinion of my own yet. Please note I will not indent any further:
“I wanted to draw your attention to something. Are you familiar with “AI Dungeon,” text-based RPG “open world” game running on GPT-2 / GPT-3? Here’s the author’s discussion on medium, or you can play the GPT-2 version for free to get a sense of it directly.
But what I really want to draw your attention to is players who are using custom prompts to open up dialogs with GPT-3 about non-game things.
This result is particularly impressive: https://www.reddit.com/r/slatestarcodex/comments/hrx2id/a_collection_of_amazing_things_gpt3_has_done/fy7i7im/
( Following that string of posts is a little hampered by reddit’s format; here are the posts in order: part 1, part 2, part 3)
If the author is to be believed, they’ve had GPT-3 / “Dragon”:
1. write code
2. act as a pharmacology tutor
3. write poetry
4. translate english, french, chinese (the instruction to “balance the intent of the author with artistic liberty” is particularly interesting)
It’s hard to excerpt, I’d recommend reading the whole thing if you have time.
Here’s another user’s eloquent conversation about the experience of being an AI, using a similar mechanism (screencap images of the convo, part 1 and part 2 ), with a sample prompt if you want to converse with GPT-3 yourself via AI Dungeon.
I am increasingly convinced that Scott Alexander was right that NLP and human language might boostrap a general intelligence. A rough criteria for AGI might be something like (i) pass the Turing test, and (ii) solve general problems; the GPT-3-AI-Dungeon examples above appear to accomplish preliminary versions of both.
GPT was published in June 2018, GPT-2 in February 2019, GPT-3 in May 2020.
As best I can tell GPT -> GPT2 was ~10x increase in parameters over ~8 months, and GPT2 -> GPT3 was ~100x increase of parameters over ~14 months. Any number of naive projections puts a much more powerful release happening over the next ~1-2yrs, and I also know that GPT-3 isn’t necessarily the most powerful NLP AI (perhaps rather the most popularly known.)
When future AI textbooks are written, I could easily imagine them citing 2020 or 2021 as years when preliminary AGI first emerged,. This is very different than my own previous personal forecasts for AGI emerging in something like 20-50 years…
p.s. One of the users above notes that AI Dungeon GPT-3 (“Dragon”) is a subscription service, something like ~$6 a week. MIE.”