Longreads + Open Thread

Longreads

• Dylan Patel of SemiAnalysis on Why America Will Lose Semiconductors and what to do about it. China is currently the largest consumer of chip fabrication equipment, with growth faster than every other major region in 2021. This is driven by subsidies including ten-year tax holidays and billions of dollars in below-market-rate loans. The piece includes several policy proposals for reshoring chips. One important thing to note is that some of these have a long lag time: education, for example, won't affect the US's capacity to produce chips this year, but will be a limiting factor later on.
• This 2016 Wired piece looks at how Dropbox moved off AWS. This is a case where the margin for error is low and there really isn't room to ship an update if mistakes get caught late in the process—at one point, Dropbox ran a limited version of their service on their own hardware, with the stipulation that they needed 180 days without major bugs before they'd finish switching over. (Via the always enjoyable Interconnected.)
• Another look at infrastructure: the IMF has a map of countries by mean road speed. A lot of it is what you'd expect; there's a straightforward GDP correlation. But there are some countries that are interesting outliers, especially Saudi Arabia, Iran, South Africa, and Chile. So countries that experience long-term growth can have good infrastructure, but when a country benefits from natural resource export booms, it often has the budget and needs to build excellent infrastructure all at once.
• Null results are important to notice: this VoxEU story profiles a job-seekers' site that failed to have any meaningful impact on unemployment. Solving matching problems is always hard, and when the matching is automated, it sometimes means automatically matching one side to human filters on the other side, not trying to automate away the entire filtering process.
• In Tablet, a very lively interview between David Samuels and Edward Luttwak, each of whom is trying to be more unfiltered than the other. "The conversation has been edited and condensed for ease of reading, with all the controversial parts left in." It touches on topics ranging from intelligence failures to the importance of nicotine to why Goethe has been fully translated into Chinese but not English. Highly recommended.

Books

• Codes of Finance: Engineering Derivatives in a Global Bank: A fun book that looks at a trading floor from an anthropological perspective. Some of this is very useful for understanding why things go wrong at banks—traders, salespeople, quants, and back office personnel are all speaking slightly different languages, and that linguistic confusion can lead to losses. It has some segments that can be read as either taking the core idea very seriously or engaging in deadpan satire (there's a riff on how stealing someone's calculator is a Lockean approach to property rights, and it's the first time I've seen Deleuze used to explain investment banks).

Open Thread

• Drop in any links or comments of interest to Diff readers.
• Two labor trends from the last two years were a positive inflection in compensation for some already well-paid jobs in tech and finance, and an improvement in wages for the lowest earners. Is either one sustainable?

New - Reader Feedback!

A reader recently pointed out that open threads have a cold-start problem: if you read right away, there won't be any comments, and if you wait too long, you might miss responses. With that in mind I thought it would be good to highlight some feedback each week. In last week's post, I asked what makes some institutions hard to disrupt. From Philo:

I think it's mostly the characteristics of the industry rather than the individual institutions. A lot of our leading universities have been there since the 17th or 18th century, it's not just Harvard. Most of our leading banks were leaders since the 18th century (or late 19th century in the case of west coast banks), and so on. Organizations and industries are Lindy for sure.

It's interesting that people intuitively think there is turnover and natural replacement within industries but you don't see it that much after the first generation, we're all using most of the same consumer products brands that were popular 75-100 years ago. I suspect it's more consumer stickiness than recruiting employees - "a ham sandwich could run Coke" etc.

Which raises the question: what leads to an unusually high- or low-turnover industry?

Two other pieces of feedback from this post on applying the concept of type-safety to emails and other corporate communications ($): Umang Jaipuria has an earlier post expanding on the broader concept of a cultural vocabulary, and Nate Meyvis has thoughtful pushback. Specifically:

Programs that have good correctness checks can be bigger: stronger typing is one way to accomplish this, but only one. Good testing is another approach (I don't mean to imply that it's an orthogonal one). And I've seen strong typing prevent a code base from growing: complicated, ill-defined types can be so brittle that engineers either (i) leave the code in place, unable to change the types, or (ii) switch all those complicated types to Any (or some analogue of Any) when they need to change something.

There's often convergent evolution towards the same endpoint that uses mutually incompatible and path-dependent efforts to get there. Maybe if you're obsessed with type safety and you treat software as an instantiation of a theorem, you won't be able to inhabit the mindset of someone who uses good testing and might analogize their program to a factory—a factory where the ideal defect rate is zero but where the achievable one is nonzero and it's critical to weigh costs and benefits.

(Another conclusion here is that if you borrow a mental model from a field you're less familiar with, there is a good chance that someone who actually works in that field will respond with a much more thoughtful set of analogies. Please use this trick responsibly.)

A Word From Our Sponsors

Import millions of rows of messy customer data in minutes

“Hey, can you send me a spreadsheet with the following columns at minimum?”

Sounds simple enough, but ask 100 customers to do it, and chances are you’ll get 100 different spreadsheets, documents or formats sent back to you. Now try collating all that data into one place, without losing your mind. Or worse, asking your customer to do it.

If this sounds anything like your team’s reality then you need to get some help from Flatfile — who have built the ultimate data importing cheat code.

Their innovative platform lets your customers drag and drop their data into your system without formatting beforehand. Then the magic starts. Smart matching, file hooks and algorithmic data detection means that a messy file that used to take hours can now be imported, cleaned and validated in literally minutes. The best part? You can embed Flatfile straight into your product, or in a prebuilt hosted page.