Longreads + Open Thread

Scale, Solo Inventors, State Capacity, China, Juul


  • Packy McCormick has a good writeup of Scale. AI feels like magic at times, but is actually fairly close to this classic Far Side cartoon: there's a lot of human effort required to train those models. Scale has an interesting loop: as it gets better at training models, it needs fewer human inputs, so it's a business that continuously improves its unit economics as it grows.

  • The myth of the myth of the lone genius: solo inventors go from mythical to common if you round up contributions slightly. There have been team efforts behind discoveries, but some of the big team efforts involve multiple discrete discoveries made by different people. It's surprisingly hard to figure out what kind of time management is most likely to produce big discoveries (other than working hard), and the diversity of success stories implies that there's a strong inventor-process fit dynamic at play.

  • China's state capacity through the lens of party drug availability.

  • Larry Summers and Jordan Schneider on China: One of Summers' good open questions is how China has so many successful technology companies given the country's low average GDP per capita. Exclusive access to a big market is a factor, sure, but it is striking just how many big Chinese tech companies there are.

  • long retrospective on Juul. It's a somewhat poignant piece, because there was a time when Juul had regulatory benefits ahead of it, rather than costs: "The FDA is trying to carry out this long-term plan that began under Scott Gottlieb, but that had actually been talked about for decades. They call it the Tobacco Endgame. What that means is the FDA has decided that there’s a huge potential public health benefit in getting people off smoking cigarettes onto potentially less harmful products, like Juul." This didn't work out, partly due to mistakes on Juul's part, partly due to politics, and partly because previous anti-smoking campaigns were so effective: cigarette regulation in the US had substantially reduced youth smoking, and Juul's product was an unusually good one for people who liked nicotine but couldn't openly consume it. So Juul was simultaneously helping the FDA achieve a long-term goal and reversing some of their shorter-term efforts.


  • The Republic: it's a rewarding but difficult exercise to figure out personal virtue by thinking about the ideal government and working backwards to what this says about the ideal person. One of the interesting points on tyranny in Book IX is that the tyrant's life is miserable and low-agency, because unjust rules spend all of their time clinging to power, which doesn't allow them to do much of anything else. This has interesting parallels to Zero to One's point that running a monopoly is simply a lot more fun than running a company in a competitive business. Even if monopolists work hard, it's towards growing their lead and creating new complements to the core business, not just keeping what they have.

  • Dealers of Lightning: this book looks at Xerox's Palo Alto Research Center. The thumbnail sketch I had in mind was that PARC invented much of the PC revolution and didn't commercialize any of it, and while that's largely true, the story is more nuanced. It's also a story about one of the downsides to monopoly: if the core business is lucrative, and there's a budget for R&D, the results of that research are both economically uninteresting and potentially a threat. Also worth reading because of the theme that many of the most enthusiastic researchers at PARC were enthusiastic because of Moore's Law. When a cost curve starts bending downward, it sparks a lot of creativity, because anything that's cool-but-too-expensive today will be simply cool by the time it comes on the market.

  • The $1,000 Genome: The decline in the cost of genome sequencing has been faster than Moore's Law, although the regulatory environment means that the impact of this happens on a delay. This book is about a decade old, but captures a very interesting moment when genetic information became widely available to consumers and increasingly useful for medical care. (One especially prescient part of the book speculates that the next time there's a SARS-like epidemic, the germ will be sequenced early and a cure will be produced before the disease spreads. Which is close to what ended up happening.)

Open Thread

  • Feel free to share any long-form writing and books—including your own!—that Diff readers would enjoy.

  • Calling the top in an industry's influence doesn't mean calling the top in the stock price. I sent a long email to a friend in early 2018 with some thoughts on regulation as a headwind for Facebook and Google. This was about a month before Cambridge Analytica broke—but since that email, FB is up 93% and GOOGL up 117%, both well ahead of the S&P. If you'd shorted Microsoft when the DOJ filed its antitrust complaint in 1994, you would have lost about 18 times your money by late 1999. And Marc Rubinstein has written about how hard it was to short Western Union even though the business was in terminal decline. Part of what's interesting to me about content recommendation companies is that they're companies that get upside from the regulatory downside of Big Tech. Are there other good examples today where there's a short thesis that might not play out, but the same facts support a long thesis in something else?

Leave a comment