Longreads + Open Thread
Coase; Xi; Mr. Beast; AI; Monzo; Trader Joe
One of life's great pleasures is taking a throwaway argument entirely too seriously and demonstrating that it's completely wrong. Ronald Coase's 1974 essay, The Lighthouse in Economics, is a classic case. Lighthouses are often used to illustrate the idea of a public good, since it's hard to charge for their use. But Coase digs through history and finds that they were privately-owned and operated for a long time, albeit with regulators creating rules that allowed them to charge. This is not purely a story about how the private sector can provide a surprising array of goods without government assistance: it's really a story about how the creation and allocation of property rights is so important to determining what gets built. (This story was prompted by the news that there was a bidding war for a lighthouse recently, though that happened for unrelated reasons.)
Political systems have formal and informal structures. Nowhere in the Constitution does it say that a government program can be vetoed through tactical leaks to major media outlets, but that happens all the time. In post-Mao China, part of the informal system is that there have been party elders who keep an eye on the current leader and keep things stable. Zhuoran Li writes about how this is not the case. One thing that stabilizes political systems is to use the same legitimacy criterion for different positions, but to do it at different lags: Supreme Court justices and House Representatives are both selected through a process that requires some popular approval, albeit indirectly, but in the latter case it's based on what was popular in the last election, while justices get in based on what was popular when the people who nominated or confirmed them were in power. That makes the system less responsive to change, but also means it's less prone to overcorrection. Companies can do this, too; a board of directors will often have some members who were CEOs of similar companies decades earlier, and who can help the current CEO apply what they’ve learned.
Xi Jinping is not the only powerful person with obscure motives and a cult of personality with millions of members. Mr. Beast is a YouTuber whose schtick usually involves weird contests with constantly changing rules and large prizes. This piece looks at one of those contests—a Mr. Beast fan gets $500,000 if he can spend 100 days in a circle feet in circumference. It's a video that would have a very different moral if it were fiction. But, as the author of the piece notes, it is unfortunately what audiences want. (The video is roughly 18 minutes long and has had 34 million views, so one person got paid $500,000 to spend 100 days in isolation but that led to 421,000 days of viewing time.)
Yan LeCun has a long interview in ZDNet arguing that recent AI advances are approaching a local maximum. Creating artificial intelligence is partly a process of defining intelligence more rigorously. The usual way we think about it is when we describe how to do things humans find difficult, which means we're building systems that don't start with the things we do easily. ("How do you figure out whether 173 is a prime number?" is most easily answered in pseudocode, but "What process do you use when you recognize a friend you haven't seen in years?" is a) easier for people, and b) more of a black box for machines.) The general winning bet in AI over the last few years is that systems that are hard to explain in detail can still produce impressive results if enough data and hardware get thrown at them, and we don't necessarily have to build things in the same order that evolution did to get the same result. But hitting a local maximum for a promising new technology is always something to pay attention to.
Tom Blomfield has a nice firsthand account of the growth of neobank Monzo. Lots of good lessons here: the path dependence of building a very minimal minimum viable product and then getting enough media attention from it to realize that it was on the right track; finding the metrics that predicted which users would stick around, and then building marketing around attracting those users; and the benefit of things that don’t scale (Monzo’s team met lots of its initial customers in person because they hadn’t built the functionality to mail them cards).
Via Snippet Finance.
The Diff is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
Becoming Trader Joe: This book is a riot. It's one long pivot from convenience stores to what Trader Joe's is today. The author has a fun libertarian streak; he talks gleefully about evading export quotas on tuna and describes California's alcohol pricing regulations as "quasi-fascist." Many companies get built around a single extreme decision, often in the form "sell the best X" or "sell the cheapest X." In Trader Joe's case, their economies were constrained by a policy of paying people well—which ended up encouraging efficiencies in every other part of the business. When employees have a good reason to stick around, they're less of a cost and more of an investment, and Trader Joe's got very good at steadily increasing that investment's ROI. Even though Trader Joe's wound up with a distinctive store, many of the decisions that got them there were incremental. For example, there were protectionist quotas against foreign cheeses, but at the time those quotas were written, US consumers didn't eat many varieties of cheese. As a result, Trader Joe's could sometimes sell brie for less than the price of Velveeta.
Several readers have recommended this one—thanks in particular to Philo, who writes the excellent, MD&A for being the most persistent.
Drop in any links or comments of interest to Diff readers.
Part of the consumer Internet growth story a few years ago was that regulations written before the Internet didn't necessarily cover the equivalent activity if it was done on a smartphone (or, at least, companies hoped the rules didn't apply). And that has been part of the investment case for some crypto projects as well. Are there new leaks in regulations that will appear as AI gets rolled out?
Companies in the Diff network are actively hiring right now. Contact us if you're interested in hearing about. Some profiles for which we have interesting opportunities:
Early-career finance/consulting with a combination of research skills and an interest in enterprise software. (US, multiple locations)
Developer with an appreciation for functional programming and an interest in fixing education. (US, remote)
DevOps engineer interested in infosec and crypto. (US, remote)
Entrepreneurial person who can launch existing products in new markets and handle both client outreach and operations. (US, multiple locations)
Hybrid consultant/coder who can help clients change their development process to eliminate some categories of technical problems. (Multiple roles here, US and UK.)
If you're interested in hiring through the Diff network, please reach out.
In last week's Longreads, Calvin McCarter asks how much of US's tech industry dominance is driven by demand rather than supply:
Even in the early days of the PC, the US had a large community of techie consumer hobbyists eager to spend money on a Mac 128K or IBM PS/2. This community also was happy to give feedback, and magazines like BYTE and PCMag also represented the "voice of the customer," enabling and forcing PC makers to respond to consumer desires and complaints. In contrast, while Japan was a leading market the latest electronic hardware, neither Japan nor Europe had such a large number of early PC adopters.
This seems true for a lot of tech growth areas. In the early Internet-era, websites were generally able to extract more ad revenue (per impression and per click) from America than elsewhere, which meant the customer base of a web startup was concentrated in the US. Much of the demand for crypto coins and web3 products has come from libertarians, also a disproportionately-American population. The demand for enterprise software, SaaS, and cloud computing is disproportionately from tech startups which are setting up their tech stack and operations from a clean slate, and tech startups are also disproportionately American. And even big, established American companies (many of them grown-up-startups themselves) are disproportionately more open to trying out new enterprise systems.
This is a great question. And as some of the examples show, it has feedback loops: we have lots of demand for enterprise software because the kind of company that probably spends the most of its incremental revenue on additional enterprise software is other enterprise software companies: every new sales representative means a new Salesforce, Docusign, and Zoom license, not to mention more Uber/Lyft and Brex/Ramp spending. This is a point raised more than once in The Power Law: once an investment category has been established as hot, there's more money in figuring out what new categories will emerge as a consequence.
A Word From Our Sponsors
Leveraging AI to drive deeper insights
Stream by AlphaSense provides investment research teams and analysts with unique, high-quality expert insights on demand. Sign up for a free 14-day trial here to make faster and more cost-effective investment decisions with AI technology that understands your sentiments.
With Stream, you’ll have instant access to:
20,000+ on-demand transcripts of expert interviews conducted by experienced buy-side analysts. Get the operator’s perspective, first-hand, on the companies and sectors that matter to you.
The fastest-growing expert call transcript library in the industry.
Hand-picked buyside interviewers (500+ hours experience).