Longreads + Open Thread

Longreads

• Santi Ruiz interviews Jennifer Erickson and Greg Segal on kidney donations. It's a dire piece. The good news: "So if we can get patients a kidney transplant, not only can they live a much better quality of life, which is what Greg and I care about, we can also save the taxpayer up to $1.5 million per patient in foregone dialysis." The bad news: "An astounding one out of every four kidneys that's recovered from a generous American organ donor is thrown in the trash," and "Seventeen percent of kidneys are offered to at least one deceased person before they are transplanted, because the system doesn’t do appropriate data hygiene to pull deceased patients off of the list." Reading this article, one question that comes up a lot is: do any of these mistakes ever ruin someone's day? Obviously this happens to the patients, or their survivors. But what about the people working in kidney transplants? Do they not feel bad when they make a million-dollar mistake by letting a kidney thaw before it's transplanted? I have made a million-dollar (albeit nonfatal) mistake at work before, and was pretty bummed even before my boss checked in to make sure I felt appropriately bad.
• Felix Stocker on talent, and really on appropriately-channeled obsession. There is sometimes a debate of sorts over whether extreme success comes from raw talent or hard work (or luck!). This piece makes the argument that it's the confluence of talent and effort applied towards that talent—you can work extremely hard to get pretty decent at something you're not naturally good at, but it's a demoralizing process. Whereas working hard at the things that don't even feel like work is a recipe for extraordinary achievement.
• The only two things you need to do to get rich are compound your net worth at a reasonable rate, and do that compounding for as long as possible. In Bloomberg, Richard Dewey and Aaron Brown cover the second half of that in this interview with the extremely well-preserved card-counting, options pricing, and stat arb pioneer Edward Thorp. It’s a good look at the benefits and limitations of apply knowledge across domains—figuring out convertible arbitrage doesn’t directly tell you anything about how to avoid a heart attack, but Thorp does quote his risk of dying in a car accent in terms of basis points per year, so at least some parts of his intellectual framework apply to both money and fitness.
• Sara Fish, Yannai A. Gonczarowski, Ran I. Shorrer: Algorithmic Collusion by Large Language Models. This is a fun piece, particularly in light of the recent Diff post on algorithmic collusion with simpler models. The paper shows that GPT-4 agents do end up pricing like cartels. Less sophisticated models don't. That's telling; as the agents get smarter, they start to make more optimistic assumptions about other peoples' behavior, and they do things that are more positive-sum (for them, not for society—but they've been told that the "sum" they want to be as positive as possible is P&L). As with other LLM stories, one should always consider the source—specifically, what kinds of text will show up more or less often in text than they do in thoughts and behaviors. The training data probably has more sellers talking about how to make money and more consumers talking about how to pay less, and the bot is being asked to play the role of a seller. (The corpus also has many, many examples of Patrick McKenzie and Marc Andreessen telling people to charge more!)
• Dan Williams: People are persuaded by rational arguments. Is that a good thing? This is a really fun one, and something I've wondered about: saying that issues should be resolved through reasoned debate, or that consent is the guiding principle in interpersonal ethics, amounts to saying that sufficiently persuasive people should be allowed to do whatever they want. And who argues for this? Almost by definition, those same persuasive people who benefit from it! In a way, of course, this is a ridiculous argument, since it amounts to saying that we should not necessarily do things just because doing them makes more sense than not. But that's just because language can be tricky: a "rational argument" can exist in some Platonic sense, but to make it practical we have to say "an argument that the listener finds rational," which could be an irrational argument that pushes the right buttons. So the real conclusion is one about calibration, not outcomes. Yes, we should generally do whatever seems reasonable—but we should also remember that this is exactly what power-seeking people who are good at sounding reasonable want us to do!
• This week in Capital Gains: on when and how you can compare stocks to flows. Remember, financial markets are a nonstop attempt to measure the correct ratio of stocks (i.e. asset prices) to flows (the cash they generate over time). So you could compare market cap to net worth (because both are a function of these markets), but you can’t compare either to GDP (unless there was a market for slices of countries' future GDPs, or of their taxes).
• And in The Riff, we discussed pivots in hard tech, security and privacy, and how much law enforcement is too much. Listen with Twitter/Spotify/Apple/YouTube.

Books

By the Numbers: Numeracy, Religion, and the Quantitative Transformation of Early Modern England: Go far back enough in history, and just about everyone is illiterate and innumerate. How did we bootstrap from that to this—a textual world where it's perfectly normal to use abstractions like fractions, negative numbers, probability, and the like in everyday conversation? The answer: not easily.

This book is partly a cultural history (more on that in a bit), but a lot of it is the story of how difficult it is to get people to adopt new and unfamiliar mental models. Arabic numerals are far more convenient than roman numerals for both calculation and storage. But they're not perfect, and early critics made the important point that they're easier to forge: you can turn a zero into an eight, add a digit, or otherwise mangle numbers; the unwieldy nature of roman numerals is actually a security feature! There's a similar debate over when and how to adopt a new calendar (England took a century and a half to adopt the Gregorian calendar, mostly because doing so meant letting the Pope tell them what to do).

The real fun of the book is taking a peek at what everyday life was like in a world where people need to deal with quantities but don't have especially good tools for doing so. Some of the early anecdotes sound like world-building from a D&D group that consists entirely of Federal Reserve employees. For example: one way for mostly-illiterate and mostly-innumerate people to keep track of a contractual agreement to pay a debt is to take a stick, make notches corresponding to the amount owed, and then split it lengthwise so each party has a complete record. The two halves can be joined later as a sort of audit to see if either side modified their half. (The half of the stick that the creditor kept was called the "stock," which is where we get the modern term.) These tally sticks could also be issued by the exchequer, and used as a substitute for hard currency—so in fifteenth-century England, M0 might be a quantity of gold coins while M2 was a bunch of sticks. The Bank of England was buying back discounted government tally sticks in 1694, and the last of them were collected and burned in 1834.

No review of distant history is complete without noting the fun analogs to the present. This book abounds with them, especially in the probability chapter. At one point, we learn that mostly-for-fun prediction markets like Manifold are hardly anew phenomenon: "In the eighteenth century, it was possible to purchase insurance on the likelihood of births, marriages, cuckoldry, highway robbery, the fall of a besieged city, and even death by drinking gin." Meanwhile, during the great plague of the 1660s, Samuel Pepys writes about how he was too distracted to work because he and his colleagues were all busy talking about the latest weekly death stats. (And, yes, there were debates over whether the plague numbers were wrong, featuring early discussions of excess mortality.)

Open Thread

• Drop in any links or comments of interest to Diff readers.
• What are some "dogs that didn't bark" in AI? There have been layoffs here and there, but no mass unemployment among call center operators; AI-generated spam exists, but the big platforms remain usable; the economy is growing, but at the high end of normal rather than an order of magnitude faster. Any transformative technology takes a long time to roll out—electrifying factories and households took a few generations, and cars took a long time to reshape cities. But it's good to sometimes pause and see which predictions were flat-out wrong and which were merely early.

Diff Jobs

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

• A CRM-ingesting startup is on-boarding customers to its LLM-powered sales software, and is in need of a backend engineer to optimize internal processes and interactions with customers.
• A company building ML-powered tools to accelerate developer productivity is looking for ML researchers with a knack for converting research into Github repos. (Washington DC area)
• A company building the new pension of the 21st century and building universal basic capital is looking for fullstack engineers with prior experience in fintech. (NYC)
• A diversified prop trading firm with a uniquely collaborative team structure is looking for experienced PMs with a strategy ready to deploy. (Singapore or Austin, TX preferred)
• A private credit fund denominated in Bitcoin needs a credit analyst that can negotiate derivatives pricing. Experience with low-risk crypto lending preferred (i.e. to large miners, prop-trading firms in safe jurisdictions). (Remote)

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.