Friday assorted links
1. Randall Kroszner appointed to the Financial Services Committee of the BOE. And CHIPs program looking to hire.
2. Rare albino baby porcupine rescued in northern B.C.
3. Black neighborhood choice and SES.
4. Assortative mating on blood type? Hard for me to believe, but…
5. Another attempt to understand how LLM work. And will work on LLMs and machine learning now become less open? What is the role of academia in all of this? And how is GPTChat on medical questions?
Is ChatGPT moving toward the median voter?
1. ChatGPT no longer displays a clear left-leaning political bias. A mere few weeks after I probed ChatGPT with several political orientation quizzes, something appears to have been changed in ChatGPT that makes it more neutral and viewpoint diverse. 🧵https://t.co/bbrXWhWaWa pic.twitter.com/JHznag2Nve
— David Rozado (@DavidRozado) December 23, 2022
Ben Thompson interview with Daniel Gross and Nat Friedman
On the GPTs, this one starts with Ben:
…if text and images are all a commodities, that value increasingly comes not from the item itself but from the brand surrounding it. I mean, does that seem like a reasonable way this might play out?
NF: I think it does increase the returns to things that you trust, and I think it increases the returns to thoughtfulness, insight, surprising ideas that are true.
Or just surprising ideas that are not necessarily true.
NF: Yeah, I mean, if the models cause us to downgrade the appearance of authoritativeness, then that might be an excellent thing for society. If our societal adaptation is, just because it sounds formal and authoritative, maybe we shouldn’t trust it, that would be probably great. We become altogether more truth seeking. It’s like you can no longer judge people based on whether they wear a suit because everyone can afford a suit, and so wearing a suit may not be the perfect signal of reliability. I think that’s sort of where we are too. Just because you wrote four paragraphs full of complete sentences, doesn’t mean necessarily that you have an original or really thoughtful idea here.
The model so far cannot produce these big out of distribution kind of insights that kind of cause you to rewrite your whole model of the world in your head. I’m not finding that. I do occasionally find myself using ChatGPT for brainstorming and it’s like, “Gosh, how should I solve this problem?” And it’ll come up with sort of five obvious ideas. The problem sometimes is that I haven’t tried two of them.
Here is the link, yes you must pay to subscribe but worth it (I only pay for three Substacks or Substack-like products, this and Matt Y. and some NBA, Noah too.). Covers many different issues, interesting throughout! And some say the transcript is ungated.
The economic costs of depression amongst the young
A growing body of evidence indicates that poor health early in life can leave lasting scars on adult health and economic outcomes. While much of this literature focuses on childhood experiences, mechanisms generating these lasting effects – recurrence of illness and interruption of human capital accumulation – are not limited to childhood. In this study, we examine how an episode of depression experienced in early adulthood affects subsequent labor market outcomes. We find that, at age 50, people who had met diagnostic criteria for depression when surveyed at ages 27-35 earn 10% lower hourly wages (conditional on occupation) and work 120-180 fewer hours annually, together generating 24% lower annual wage incomes. A portion of this income penalty (21-39%) occurs because depression is often a chronic condition, recurring later in life. But a substantial share (25-55%) occurs because depression in early adulthood disrupts human capital accumulation, by reducing work experience and by influencing selection into occupations with skill distributions that offer lower potential for wage growth. These lingering effects of early depression reinforce the importance of early and multifaceted intervention to address depression and its follow-on effects in the workplace.
That is from a new NBER working paper by Buyi Wang, Richard G. Frank, and Sherry A. Glied.
Let’s get real about this stuff
Which media outlets and periodicals are at most danger of “being GPT-ed”? Your nominations? Or how about which particular features of those outlets?
Thursday assorted links
1. Words from the Chinese translator of Sally Rooney, a good piece.
2. Updating your syllabus for the GPTs. I’ll be doing this myself for the spring.
4. Are toy prices falling? (short video)
5. Chrome browser add-on logs your ChatGPT history.
6. From a reader: “Project Relate is a Google app that allows users with speech impediments to create a customized voice model so they can interact with Google’s voice assistant. After the user trains the model it can pretty effectively parse what they are saying even when other humans can’t.” Link here.
I Still Hate Flexible Spending Accounts
According to a new report in Money workers lost billions in so-called flexible savings accounts:
…44% of workers with FSAs in 2019 forfeited money. On average, the amount lost totals $339 per person.
…With reliable data on how often workers forfeit, how much they forfeit and how many FSAs workers hold, we can now reasonably estimate that workers forfeited approximately $3 billion in 2019 and $4.2 billion in 2020.
As I said in my post from 2017 (no indent), I hate “flexible” spending accounts, i.e. those accounts where you put say $1000 in tax-free but you then must submit a bunch of health or education receipts to claim the money–and the “benefits manager” tells you half of the receipts you submitted are no good so you have to trawl through your files to find more–or lose the money. The whole process is demeaning. My hatred of this process, however, pales in comparison to that of Scott Sumner who gives a correct analogy:
Imagine a government that took 10% of each person’s income, and put in in a wooden box. The box was placed at the end of a 10-mile gravel road. Each citizen was given a knife, and told they could crawl on their hands and knees down the road, and then use the knife to cut a hole in the box, and retrieve their money.
Scott’s point is twofold. First, there is a lot of waste in crawling down the road. Second, taken in isolation, it looks like the plan at least offers people an option and so, in isolation, flex accounts and their ilk appear to benefit taxpayers. In the big picture, however, the total amount taken in taxes is somewhat fixed by politics and economics so if we got rid of the spending accounts, taxes would probably fall in other ways that are difficult to predict but nonetheless real.
Some want to crawl down the gravel road, fearing that if they abolish the program the government will not reduce their tax rates, instead the money in the box will be diverted to welfare for the poor, or higher salaries for teachers. I can’t deny that this might occur, but if we don’t even TRY to build a good country, how can we possibly succeed? Isn’t it better to try and fail, rather than not even try?
I agree with Scott. If I am going to be forced to pay taxes I’d like to hand over my cash standing like a man and not be given the option of crawling to recoup some bills the tax collector magnanimously throws on the floor.
Canine Coaseanism
We are for a while caretakers for a dog, and so I have started thinking what kind of trades I might make with the beast. Of course for Darwinian reasons dogs have co-evolved with humans to be fairly cooperative, at least for some breeds (and this is a very smart, easily trained breed, namely an Australian shepherd). So the dog’s behavior (my behavior?) already mirrors some built-in trades, such as affection for food. But what kinds of additional trades might one seek at the margin?
One thought comes to mind. I would like to signal to the canine that, when I get up from the sofa, he does not need to follow me because there is no chance I will offer him a food treat. It would be better if he would just stay sleeping. And yet this equilibrium is impossible to achieve. Nor does rising from the sofa quietly succeed in fooling him, he follows me nonetheless.
Overall, though, I conclude that the current (spayed) version of the dog is already fairly Coasean in his basic programming.
One look at our future
The following comes from a deleted tweet, so I have paraphrased in some places and also removed references to a few specific individuals. If the author wishes to reclaim ownership of these ideas, I will write a separate blog post crediting him:
1. AGI and Fusion are what matter. Fusion accelerates AGI, since AGI is just exaflops spent on training.
2. GPT 3.5 (ChatGPT) is civilization-altering. GPT-4, which is 10x better, will be launched in the second quarter of next year.
3. Google is worried, but Microsoft is all-in, and is building many more data centers to lead the charge. Bing search is getting GPT integration next year.
4. Model configuration and training parameters don’t matter. Intelligence is just GPU exaflops spent on training.
5. If #4 is true, civilization becomes a decentralized crypto network, where computers are contributed for training and earning tokens. Querying the model costs tokens.
6. One last centralizing force — gradient descent is synchronous. Needs high GPU coordination and fast network bandwidth. Current trend is civilization centralizing with Microsoft laying 10x bigger Open AI data centers.
7. Gradient descent is the process of error correction, where an N-layer model predicts an output. When that’s far away from the target, we correct all the layer weights slightly to re-aim.
8. Some other batch of stuff I didn’t understand and cannot paraphrase.
9. A lot of the configurations work equally fine. If you throw the same GPU exaflops at the model — they perform more or less the same. Probably that is why it is evolutionarily easy to invent the brain. Open AI is at 10 exaflops right now vs. 1000 for the human brain. Probably going to equalize in five years.
10. Models are so good already that only expert training matters anymore. Co-pilot for X is in play. Anyone building a Co-pilot for my browser? Browsers are largely text-based, which GPT fully understands.
There you go! Speculative, of course.
Wednesday assorted links
Ohio teen fact of the day
The number of 11th and 12th grade males experiencing gambling problems, such as lying about how much they lost, or being unable to control their gambling, rose to 8.3% in 2022 from 4.2% in 2018, according to one survey of 7,500 7th through 12th graders in Wood County, Ohio.
People who research and treat problem gambling say the line between gambling and videogaming is blurring. Videogames, which are often played on smartphones as well as computers and game consoles, include features that mimic gambling activities like roulette and slot machines.
The FDA’s Lab-Test Power Grab
The FDA is trying to gain authority over laboratory developed tests (LDTs). It’s a bad idea. Writing in the WSJ, Brian Harrison, who served as chief of staff at the U.S. Department of Health and Human Services, 2019-2021 and Bob Charrow, who served as HHS general counsel, 2018-2021, write:
We both were involved in preparing the federal Covid-19 public-health emergency declaration. When it was signed on Jan. 31, 2020, the intent was to cut red tape and maximize regulatory flexibility to allow a nimble response to an emerging pandemic.
Unknown to us, the next day the FDA went in the opposite direction: It issued a new requirement that labs stop testing for Covid-19 and first apply for FDA authorization. At that time, LDTs were the only Covid tests the U.S. had, and many were available and ready to be used in labs around the country. But since the process for emergency-use authorization was extremely burdensome and slow—and because, as we and others in department leadership learned, it couldn’t process applications quickly—many labs stopped trying to win authorization, and some pleaded for regulatory relief so they could test.
Through this new requirement the FDA effectively outlawed all Covid-19 testing for the first month of the pandemic when detection was most critical. One test got through—the one developed by the Centers for Disease Control and Prevention—but it proved to be one of the highest-profile testing failures in history because the entire nation was relying on the test to work as designed, and it didn’t.
When we became aware of the FDA’s action, one of us (Mr. Harrison) demanded an immediate review of the agency’s legal authority to regulate these tests, and the other (Mr. Charrow) conducted the review. Based on the assessment, a determination was made by department leadership that the FDA shouldn’t be regulating LDTs.
Congress has never expressly given the FDA authority to regulate the tests. Further, in 1992 the secretary of health and human services issued a regulation stating that these tests fell under the jurisdiction of the Centers for Medicare and Medicaid Services, not the FDA. Bureaucrats at the FDA have tried to ignore this rule even though the Supreme Court in Berkovitz v. U.S. (1988) specifically admonished the agency for ignoring federal regulations.
Loyal readers will recall that I covered this issue earlier in Clement and Tribe Predicted the FDA Catastrophe. Clement, the former US Solicitor General under George W. Bush and Tribe, a leading liberal constitutional lawyer, rejected the FDA claims of regulatory authority over laboratory developed tests on historical, statutory, and legal grounds but they also argued that letting the FDA regulate laboratory tests was a dangerous idea. In a remarkably prescient passage, Clement and Tribe (2015, p. 18) warned:
The FDA approval process is protracted and not designed for the rapid clearance of tests. Many clinical laboratories track world trends regarding infectious diseases ranging from SARS to H1N1 and Avian Influenza. In these fast-moving, life-or-death situations, awaiting the development of manufactured test kits and the completion of FDA’s clearance procedures could entail potentially catastrophic delays, with disastrous consequences for patient care.
Clement and Tribe nailed it. Catastrophic delays, with disastrous consequences for patient care is exactly what happened. Thus, Harrison and Charrow are correct, giving the FDA power over laboratory derived tests has had and will have significant costs.
Best movies of 2022
The Lost Daughter, TV but still good, a movie of sorts, based on Ferrante.
Belle, spectacular Japanese anime.
Licorice Pizza, a good normal movie, captured California well.
Compartment Number Six, with a new meaning after the war of course.
Petit Maman, French, short, plays mind games with you, profound.
The Quiet Girl (Irish)
Vesper, Lithuanian, dreamy, Tarkovsky influence but faster-paced, an underrated movie this year.
Tár, really quite good and interesting.
Decision to Leave, Korean crime drama with Hitchcockian twists and inspirations.
The Fabelmans, Ignore the cloying preview.
Saint Omer, French-Senegalese courtroom drama.
Clytaemnestra, Korean movie, one hour long.
EO, Polish movie about a donkey, better than you think.
Overall an abysmal year for Hollywood, a pretty good year for the movies. I haven’t yet seen Oppenheimer or Bardo, so their absence on the list should not be taken as a negative signal.
Why are Americans spending less on holiday gifts?
That is the question behind my latest Bloomberg column:
The research is clear: Americans are becoming less generous over the holidays. Not to sound too much like a Scrooge, but this is not necessarily a bad thing.
In 1999, Americans said they planned to spend $1,300 (converting to 2020 dollars) on holiday gift giving. In 2020, that amount was about $800. These numbers are based on Gallup data, but retail sales figures show a broadly similar pattern. From 1935 through 2000, gift-giving tended to rise with disposable income. Since 2000, gift-giving has fallen as national disposable income has risen.
…One hypothesis is that Americans are simply getting less generous. Yet charitable giving is robust, so that’s probably not right.
An alternative possibility is that Americans are too rich for gifts to make sense. It’s not only that billionaires are hard to buy for; the rest of us are, too. You might think your friends already have most of the important things they need, so how can you buy them something meaningful at the margin? This logic doesn’t hold for all Americans, but perhaps the higher earners account for a big enough share of the gift-giving total that it exerts a downward pull on the numbers.
The cheeriest scenario — again, speaking strictly as an economist — is that Americans are realizing that gift-giving often doesn’t make much sense. If you give me a gift and I give you a gift, neither of us is quite sure what the other wants. We might both be better off if we each spent the money on ourselves. Under this hypothesis, Americans are not becoming less generous, they are becoming more rational.
Another rationale for gift-giving is that it tightens familial and social bonds. Perhaps it does, but it is not the only means for doing so. More and better communication — which has also become cheaper and easier over the last two decades, with email, texts and cheaper phone calls — may make gift-giving seem less essential.
There is also the possibility that we, as a society, have lost that “Christmas spirit,” whatever that might mean. After all, secularization is rising and churchgoing is declining. Christianity is less central to American life. Whether this social development is all good or bad will of course depend on your point of view.
What do you all think are the likely causes here?
*Paved Paradise: How Parking Explains the World*
Henry Grabar has written an excellent book on how badly America has screwed up its parking policies, and in turn ruined many of its cities, or at least parts of them.
Here is just one little tidbit:
Parking on the curb was not such a great option in those days: in 1990, 147,000 cars were stolen in New York City — one theft for every fifty residents. My parents joined the club shortly before Christmas, 1993. Their next car went into a garage, a prewar building with a cable-hauled elevator serving three floors of parking…
The revival of Manhattan around the millennium, which often took the form of apartments replacing parking, and the concurrent decline of car thefts, which fell by an astounding 96 percent between 1990 and 2013, put a lot of pressure on the curb.
I wish such books would spend more time discussing whether dense urban areas are simply a fertility trap. Nonetheless excellent work, and everyone interested in urban economics (or parking) should pick this one up. Due out in May, you can pre-order here.