Links to Consider, 12/15

relative earnings and labor force exit; Allison Schrager on interest rates and capital allocation; Gary Marcus on AI; Scott Alexander on crypto;

Pinghui Wu writes,

Labor force exit rates decline with a worker group’s expected earnings but increase with their reference earnings, defined as the average earnings in a state across all prime-age workers. Compared with a 10 percent increase in expected earnings, a 10 percent increase in reference earnings has the same sized but opposite effect on the labor force exit rate, suggesting that workers discount their own expected earnings by their peers’ earnings.

If lower wages for me make me want to quit working, that could be ordinary labor supply elasticity. But if higher wages for you make me want to quit working? That sounds like status loss, which is the author’s point. So if Richard Reeves thinks that men can be encouraged to work in low-status jobs in the health sector, he needs to think again.

For

Josh Barro

,

Allison Schrager

writes,

The tech industry got so frothy and crazy (and the entire crypto market became so valuable) because interest rates were low for a long time. When interest rates are low, debt is cheap and bad ideas receive funding, resulting in things like endless bad Netflix content. Low rates also prop up equity in both public and private markets. The value of a stock today is its expected future profit divided by today’s interest rates. Or, lower interest rates means higher stock valuations, especially tech stocks which are presumed to have big payouts far in the future.

Low rates also encourage investors to take on more risk — they have to if they want to earn more than the paltry returns bonds offered the last few decades.

I don’t think you are obliged to call this Austrian macroeconomics, but other people do.

Gary Marcus

writes,

Pastiche, in case you don’t know the word, is, as wiki defines it, “a work of visual art, literature, theatre, music, or architecture that imitates the style or character of the work of one or more other artists”. GPT-3 is a mimic.

…GPT”s heavy use of a technique called embeddings makes it really good at substituting synonyms and more broadly related phrases, but the same tendency towards substitution often lead it astray.

Pointer from Tyler Cowen. Marcus links to a previous essay, where he wrote,

As AI researchers Emily Bender, Timnit Gebru, and colleagues have put it, deep-learning-powered large language models are like “stochastic parrots,” repeating a lot, understanding little.

The latter essay argues that “deep learning” is in fact too superficial, and that it needs to be supplemented by what he calls neurosymbolic methods. When I was looking into automated mortgage underwriting, almost thirty years ago, there was a division between statistical systems and expert systems. Statistical systems won. If a credit scoring algorithm treated as important a variable that human underwriters had long overlooked (unused credit was such a variable), it worked better than a system that tried to follow the logic of human underwriters. I think that integrating deep learning (which reminds me of statistical credit scoring) with a more expert-system approach is not going to be easy.

Scott Alexander

writes,

Crypto is an interesting technology that had one terrible piece of bad luck: its standard-bearer, Bitcoin, went up in value 10,000x over a few years.

When something goes up in value 10,000x, it’s hard to think of it in any other context. Whatever it was before, now it’s “that thing which went up in value 10,000x”. And so both crypto believers and detractors have treated crypto primarily as a thing for going up in value6. Believers are excited that it did go up that much, hope it might go up more, and fall for a thousand scams that promise continuing going-up. Detractors correctly point out that buying things only insofar as they go up in value makes them Ponzis, and mock crypto for not having gone up in value enough recently.

I’ve had a similar thought recently, which I express as the worst thing that happened to crypto was the run-up in Bitcoin price. That is because it focused people’s attention on speculation, which in turn elevated “pyramid scheme” into a leading use case for crypto.

My longstanding aphorism is that standard banking is adjacent to government and crypto is adjacent to crime. You may think that government is so bad that you would rather take your chances with being adjacent to crime. But in my judgment, the amount of financial capital and human talent going into crypto start-ups is more than what is warranted.