Longreads + Open Thread

Longreads

• In Bloomberg, the real story of the great SEC EDGAR hack, in which a group of traders got access to filings, including earnings announcements, before they were made public. It's partly a reminder of how much rickety infrastructure is out there—EDGAR dates back to the early 90s, and has been iteratively patched since then, but apparently still has a fair number of vulnerabilities. It's also a story about the limits of law enforcement, and the internal politics of regulators: the reason there's a story here, so long after the fact, is that the biggest winners from the hack were out of reach of US law enforcement.
• From Pradyumna Prasad and Minh Nhat Nguyen: if you pit LLMs against each other in debates, they're overconfident, even if they're reminded to calibrate their confidence. If two LLMs go head-to-head, each one tends to estimate greater than 50% odds of being right, which is mathematically impossible. On the other hand, consider the training data: if you argue, it's probably because you think you're mostly right—but to have an argument, your interlocutor needs to think the same thing! It's a bit like the longstanding finance puzzle that, given that markets are fairly efficient, very few people should be trading. Which is true! But you get a lot more interactions if both sides are just a little bit overconfident. If this problem is going to be solved, one way to do it would be via synthetic data, specifically producing cogent arguments in favor of something that was later proven wrong. (Via Marginal Revolution.)
• Eric Levitz on the decline of reading, specifically the decline of reading longer works like books. Culture is drifting back towards oral rather than literate culture—but actually drifting further in that direction than it used to be! In oral cultures, ideas only stay alive if they're continuously reproduced; you can't listen to the same epic poetry or songs that your grandparents did if there's no one left to remember them. But the selection is much, much faster if every bit of content is being continuously ranked against everything else based on real-time signals, and everyone has access to a global selection of content. Oral culture used to mean memetic selection for being the catchiest song or story in the village, but now competition for catchiness is global rather than local. One irony of all of this is that the platforms that enable this tend to get built by people who can absorb complex ideas, often through books, and maintain towering abstractions in their heads. So there may be a cycle: hyper-orality requires a hyper-literate priesthood, and if that priesthood can't be replaced because TikTok is simply so compelling relative to The Structure and Interpretation of Computer Programs, the whole thing falls apart. (If you're trying to address this problem yourself, one technique that has worked for me as a forcing function is to produce a weekly email that discusses five or so longish pieces of online writing and perhaps a book review or two.)
• Clara Collier on The Origin of the Research University. This piece is a reminder of how new the university-centric research model is, and how schooling can function perfectly well without research. (We don't expect middle school biology teachers to come up with novel frog-dissection techniques in their spare time, though this kind of thing occasionally happens.) As it turns out, research universities are yet another example of an institution that's downstream from measurement: to have a good school is subjective, but to credibly claim to run the best requires some kind of ranking system. If you're early to some field that's going to be competitive later on, it will look to future readers like you were playing on easy mode: in the 1790s, Duke Karl August of Saxe-Weimar decided to upgrade his university, and he was able to hire Hegel, Schiller, Fichte, and both Schlegels. It's the academic hiring equivalent of mining a few hundred Bitcoin with your laptop in 2009.
• Oscar Sykes on the invention of inflation targeting. It's a good story about uneven state capacity, and about how hard it is to get an electorate to support somewhat wonkish, technocratic policies. As a general rule, the political dynamic of inflation is that every voter hates 1) inflation, and 2) any specific policy that targets inflation. So you want your central bank to be apolitical if it's going to be effective. But that means that technocrats are better at implementing them: one of the effects of a trusted inflation target is that people engaging in behaviors that drive up inflation, like hoarding goods or borrowing money to buy real assets. So someone who can't easily get voted out can do a little bit of a temporarily unpopular policy and let the broader economy handle the rest.
• In Capital Gains this week, we ask: what is "synergy" supposed to be, and when does it show up? If mergers were always value-destructive, they'd never happen; if they were always value-creating, The Diff would be a product of OmniCorp, the company every reader worked for. Reality is somewhere in the middle.

Books

How Countries Go Broke: The Big Cycle: Ray Dalio is famous for many things—well-timed recession calls, many more poorly-timed recession calls, a surreal company culture, running the world's largest hedge fund, writing a pretty solid macro newsletter, etc. But he also wants to be famous as a philosopher who advocates radical transparency and relentless cultivation of excellence. Which means it's a challenge for him to write a book on macro without also making it a bit of a self-help and philosophy book, too.

The solution is partly typographical: How Countries Go Broke makes abundant use of bold text to showcase the high-level arguments (and Dalio expects readers to skip the details they find less interesting), while the philosophical aphorisms get a little red dot next to them. The result of this formatting choice is that the whole book sometimes reads like one of those sales letters asking you to sign up to learn about a little-known penny stock that's guaranteed to produce dynastic wealth.

The actual content is better, though there are some quibbles. Dalio's core model is that there are small-scale credit cycles that are well-understood, but also large-scale ones that take place over multiple generations and don't get enough study. But is that really true? An equally valid model is that historically, cycles were much longer because of limited communications bandwidth—if the cause of some downturn is a bad nutmeg harvest, you'll be waiting months until the news propagates from the Banda Islands to Amsterdam, and even from there Paris and London will have a week or two before the crisis hits them. If you look at the rise of financial leverage from the 1950s to today, you can see a cycle of similar length—but you're actually looking at a secular change, where longer lifespans increased global demand for financial assets, rising productivity made it economically possible to keep all of those retirees alive, and a global financial system meant that countries with different wealth levels and average ages could trade with one another.

And even this exaggerates the extent of the change.Some of the leverage cycle is invisible because of accounting. Debt levels were low in the 1950s if you look at one-balance-sheet debt, but the US and Western Europe were building out a much more extensive safety net that put lots of economic liabilities on the collective balance sheet. Some of that safety net took the form of broadly labor-friendly policies—if GM, Chrysler, and Ford are obligated to pay above-market salaries that automatically rise over time, that's a liability, even if it flows directly through their P&L without showing up on the balance sheet until the pension liabilities start to accumulate.

It's incredibly tempting to come up with grand historical theories of cycles, especially because these cycles do exist. But their implementation is different enough that excessively accurate pattern-matching is a sign of overfitting, not a big discovery: what you actually want to look for is the same meta-pattern showing up in completely different domains—New Deal hiring compared to tech hiring, for example, or pre-Reformation monasteries' parallels with modern universities. Institutions evolve so much that you're necessarily drawing parallels between two very different organizations if you compare a modern government to a 19th-century one, and if you happen to find something identical about them, it's a suspicious coincidence.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• Let's say you want to be, not on reshoring and not on a return to the trade status quo, but on ongoing volatility. What do you buy, what do you sell, and what do you build?

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• Deerfield-backed Series A company building agents for healthcare administration (prior authorization, eligibility checks, patient scheduling) is looking for a senior software/AI engineer to build backend services and LLM agents. Experience building and monitoring production-quality ML and AI systems preferred. (NYC, Hybrid)
• Ex-Citadel/D.E. Shaw team is building a new financial market within a multi-trillion dollar asset class and looking for a data scientist with classical and generative ML experience..  (NYC, Boston)
• 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 product manager with 3+ years experience building product at high-growth enterprise SaaS businesses. Technical background preferred. (LA, Hybrid)
• An OpenAI backed startup that’s applying advanced reasoning techniques to reinvent investment analysis from first principles and build the IDE for financial research is looking for software engineers and a fundamental analyst. Experience at a Tiger Cub a plus. (NYC)

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.