Longreads

• Trevor Klee has a smart piece on how biology uses analogies rather than paradigms. Analogies can be very useful for taking a complex system and isolating one part of it well enough that it can be modeled, with the caveat that this model a) won't completely explain the system, and b) won't even fully explain the one component it's trying to model. This probably applies to any field where there are lots of moving parts and complex interactions.

• Elad Gil reviews the tremendous impact of recent changes in AI language models. One fun detail from early on: "Of the 8 people on the 2017 transformer paper, 6 have started companies..." AI seems to have a tighter feedback loop between academic advances and commercial implementations. This may be temporary—perhaps the private sector had accumulated a surplus of data and needed better tools to use it, and as the tools catch up they'll be relatively less important. But it's been a big factor in how quickly the field has advanced.

• Rohit of Strange Loop Canon has a great piece on Effective Altruism and "unknown unknowns," with the indefinite optimist takeaway that maximizing economic growth gives us resources to solve whatever problems turn out to be big ones in the future. Which is true, but which might miss an important question: big jumps in productivity often come from trying to solve specific problems. A world geared towards higher growth will be better at moving capital and talent into those areas, but it still needs a specific vision for what should be improved.

• Mark Bergen on YouTube's early "coolhunters," who manually curated content for the site's homepage. People who build services with user-generated content seem to over-invest in recommendation algorithms early on, when the signals are weak and small scale makes them gameable. YouTube dodged this bullet by having employees trawl through content themselves to find hits. This has the added advantage of providing a continuous stream of intelligence on how people are using the site. Even if the newly-discovered content isn't that great, its existence can be useful information about unusual use cases for the product. For YouTube, there are some genres that existed before the site (America's Funniest Home Videos started in 1989), but things like mukbang and surreal fever dream videos for small children are native to the medium.

• Cai Xia writes in Foreign Affairs about how Xi Jinping's hold on power is weaker than it looks. It's very hard to know how seriously to take pieces like this: factionalism and CCP propaganda are equally good explanations for a) why Xi's accomplishments get exaggerated by his faction, and b) why someone else might view him as a product of nepotism. (One fun detail: a group within the CCP that opposes lockdowns symbolizes their opposition by refusing to wear masks. Masks-as-partisan-signifier turns out to be pretty universal!)

Books

• Capital Returns: Investing Through the Capital Cycle: out-of-print classics are hit-or-miss. You might get something good, or you might find that the book's "classic" status means the core idea has been widely shared and is obsolete. Capital Returns is more on the useful end of the spectrum. The book's central claim is that 1) industry cycles are driven by supply and demand, and 2) supply is more predictable than demand. The book has an absolutely fantastic example of how hard demand is to predict: one of the hypothetical examples of hard-to-predict numbers was the demand for long-distance flights in 2020 (the book came out in 2016). Score one for the authors: that number was indeed a surprising one.
  
  One of the difficulties with actually implementing this strategy is that, while overinvestment is obvious in retrospect, it's harder to argue with at the time. During the commodities runup of the early 2000s, for example, there was a pretty straightforward demand-side case for higher consumption as the developing world got richer, and plenty of money was made on the way up by people who expanded capacity at the right time. In practice, while the book talks a lot about understanding the cycle, it also spends a lot of time on investments that don't experience much of a cycle—banks that try to stay boring and sit out new financial innovations, or companies like lock maker Assa Abloy that operate in steadier industries. Of course, a sophisticated theory is valuable even if its main application is to avoid a large category of mistakes; investors can do fine through a process of elimination.

Open Thread

• Drop in any links or comments of interest to Diff readers.

• I've been spending more time on the alternative data business recently. Are there any fun new datasets out there? (Feel free to reply if you'd prefer not to leave a public comment.)

Leave a comment

Reader Feedback

Nate Meyvis has thoughts on how the concept of stop losses applies to poker, in response to this subscribers-only piece. When you're thinking about the optimal way to act with uncertain information—including uncertainty about both the state of the world in general and your relative skill at evaluating it—poker is an even denser source of metaphors than finance because there's less narrative bias to confuse things.

A Word From Our Sponsors

Tegus is the first port of call for M&A professionals and institutional investors ramping up on an industry or company.

Get access to a database of 35,000+ expert call transcripts, spanning 5+ years, or schedule expert calls through the platform for a fraction of the usual cost.

When thousands of research analysts are pooling their expert calls into an on-demand database, using Tegus is table stakes. It's the leading platform for due diligence and primary research.

See the power of a Tegus subscription, and get up to data parity with your competitors, with a two week free trial through the Diff.

Try Tegus