Longreads
- Todd Vaziri has a compendium of minor movie errors, including a deeply weird one in which a ghostly face appears in the background for a fraction of a second during Revenge of the Sith. It's really an essay about appreciating media by noticing the minor mistakes. For example, on of the case studies is that in Goodfellas: "The period-appropriate "movie" license plate dramatically dangles then completely falls off the car in the middle of the take, revealing the actual 1990-era license plate of the car used for the scene. This is an accidental and hilarious glimpse into the extremely specialized and detailed hard work that goes into making a Hollywood period piece (this portion of the film takes place in 1980), where every license plate of every car in the movie needed period-appropriate plates." (It almost creates an incentive to make tiny mistakes that only the superfans will notice, because getting something 99.99% right is the only way to draw attention to the fact that it's so close to 100%.)
- Mitch Moxley in the NYT on a group of young scammers who mostly met on Minecraft servers and executed a quarter-billion dollar crypto heist. One thing these illustrate is that civilization is actually quite high-trust, and given the opportunity most people will cooperate rather than defect in iterated games. And this is illustrated by the fact that the scammers in question weren't doing an incredibly sophisticated grift, but were still able to get away with a lot of money—and that they were incredibly bad at keeping their winnings quiet, to the point that they attracted enough attention that it as inevitable that they'd get caught.
- Tyler Cowen interviews Chris Dixon about crypto, decentralization, and permissionless innovation. (All very compelling ideas, but for some Diff skepticism see Can We Afford the Ownership Economy? and The Promise and Paradox of Decentralization.) Dixon's view is that a system without centralized control will give people more freedom, which is true. But it's also true that misbehavior often scales faster than good behavior, and that while a centralized entity might make decisions you don't like, it also really cares if other people are polluting the commons. It's possible to cleverly design a protocol that fights against this, but that's often most possible in a narrowly-scoped domain, and even then it's a challenge. Also fun is a bit towards the end where the two of them riff on philosophy for a bit, and the meta-philosophy of whether you read great books to actually absorb the ideas or to inhabit a worldview (once again, it is much more common for Tech Bros To Take Liberal Arts Classes than it is for poets and historians to shift a conversation towards their shared interest in compiler design or whatever).
- Carl Hendrick warns about the decline of deep reading. It's wonderful to abolish useless friction, and anything that can be done quickly is often better when done even more quickly, but that concept breaks down when the task at hand is fundamentally chunky and when breaking it down into smaller parts means losing something of the whole. There's nothing essential about 300-500 pages being the optimal size for telling a story or articulating an idea, but if that's been the default way to package things for long enough, then we're all inheritors of book-length ideas and ought to get good at digesting them if we're going to have any hope of producing something that future generations find worthy of preserving.
- Dwarkesh Patel interviews Ege Erdil and Tamay Besiroglu on the economics of, and future path of, AI. A very fun piece because it manages to be measured in some ways (superintelligent software is decades away) and completely wild in others (along the way to achieving that, GDP growth can be really, insanely high). A lot of what drives this argument is that 1) reality contains an annoyingly fractal level of detail, such that it's easy to describe some tasks at a high level and very hard to describe them in a sufficiently granular way that an AI agent can outperform a person at an entire job, and 2) even if there are some isolated areas of high growth, they'll run into real constraints that are difficult to deal with. Which actually loops back to the prior piece and the question of optimal chunking: some modern jobs encompass what used to be multiple jobs (many office workers do things on their own that were handled by secretaries in previous generations), and sometimes one job splits into many—"write the operating system and software that will run on this personal computer" used to be something that a team of a few people would do over a few months and is now multiple multi-year projects undertaken by different teams across different companies, roughly all of which will do the bulk of their work using tools written by other teams at other companies.
- In this week's issue of Capital Gains, the complementary relationship between online and offline communication, or why the fact that anyone, anywhere in the world can read arXiv papers about artificial intelligence has driven up housing costs in the Bay Area.
You're on the free list for The Diff. This week, paying subscribers read about why venture capital and media often go together ($), the case for search eating chatbots before chatbots eat search ($), and how it's hard to measure AI's market share in different categories becaues it transcends so many categories ($). Upgrade today for full access.
Books
Clashing over Commerce: A History of US Trade Policy: You might be tempted to read this book to understand what's going on with tariffs right now, but by the time you get through the introduction you'll recognize that that's a terrible idea. The book's framework, which seems accurate, is that US tariff policy was historically driven by conflicts between different interest groups—concentrated ones that wanted their industry protected at the expense of consumers who were too diffuse to fight for their own interests, or cotton and tobacco exporters advocating free trade for their own benefit while manufacturers pushed for higher trade barriers to protect their interests. But there aren't that many big interest groups today that fight for tariffs, and a book where most of the action takes place in Congress doesn't tell you much about the world of improvisational executive orders, tweet-driven rumors, and re-Truthed policy pronouncements.
On the other hand, as a history of the tariff debate in the US as it existed from roughly 1776 through last month, it's very good. The book frequently refers to the economic impact of tariffs (and in a balanced way, noting that they probably didn't hurt consumers all that much when they were high, but that they probably didn't so much create as accelerate US manufacturing growth). But it's mostly about the Clashing, not the Commerce, i.e. how did politicians come up with their plans and how did they rope together electoral majorities to put those into action.
One of the fun contrasts in the book is that while there's a lot of vociferous rhetoric on both sides, there's also some surprisingly sophisticated economic analysis. At one point, the book quotes a 19th century politician who says that for every $1 of tax revenue produced by tariffs, there's $5 of private gain. And then, half a dozen pages later, estimates that the impact of tariffs on government revenue at the time as about 0.5% of GDP, but that the impact on producers was 2.5%. These politicians, operating with the benefit of much formal economic theory, also understood that tariffs on raw materials and intermediate goods get passed through to consumers of final goods, and that a tariff reducing inputs of manufactured products would reduce exports of agricultural ones.
Some of the political debates are quite familiar—Taft wanted a tariff board to identify countries with discriminatory trade policies so they could be subject to higher tariffs (the board couldn't find any, and was later disbanded). Some of them are surreal: high tariffs and high growth in the late 19th century meant that the US budget reached a crisis-level surplus that would have drained too much liquidity from the economy—Congress sprang into action to create an expanded and very easily-defrauded set of Civil War pensions to avert catastrophe.
The nature of the debates, and the general direction of US tariffs, shifts roughly a century ago. For most of the period covered, tariffs were the government's primary revenue source, in part because the government was small and in part because they were cheaper to administer than a sales tax. This led to a fun statistical artifact: throughout the 19th century, periods of high tariffs tended to have high economic growth, while lower tariffs often presaged recessions. But this was an artifact of tariffs as a tax source and inflexible monetary policies: governments imposed tariffs when the volume of trade declined to the point that the prior rate of tariffs didn't generate enough revenue, and when there was a budget surplus, they cut taxes. The book notes that the recessions in question usually started with financial crises, and weren't directly linked to trade—in fact, some of these crises involved declines in the value of farmland, an asset class that would be expected to outperform with low tariffs.
The more recent chapters in the book look at a time when trade barriers were coming down, both those imposed by governments and those imposed by the cost and inconvenience of actually moving goods from one place to another. That's a world where tariffs are less of a political question and more of a technocratic one. At least, most of the time.
Open Thread
- Drop in any links or comments of interest to Diff readers.
- What are some niche tail risks that don’t get enough attention?
Diff Jobs
Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:
- YC-backed startup using AI to transform how companies quantify and optimize engineering productivity is hiring formidable full-stack and AI engineers. Experience with React + Typescript, Go, or Python on the ML side a plus. (SF)
- A multi-stage, fintech focused investment firm is looking for an investment associate to support thematic opportunity identification, diligence, and execution. Investing experience OR high-growth operating/investment banking/consulting experience and demonstrated interest in fintech required. (NYC, London)
- A hyper-growth startup (10x growth in 9 months) that’s turning customers’ sales and marketing data into revenue is looking for a head of deployments who is excited to work closely with customers to make the product work for them. Experience as a forward deployed engineer and leading enterprise deployments preferred. (SF, NYC)
- Well-funded, fast-moving team is looking for a full-stack engineer to help build the best AI powered video editor for marketers. Tackle advanced media pipelines, LLM applications, and more. TypeScript/React expertise required. (Austin, Remote)
- A Google Ventures-backed startup founded by SpaceX engineers that’s building data infrastructure and tooling for hardware companies is looking for a software engineering manager with 7+ years experience building large scale distributed systems. (LA, Hybrid)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.
If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.