Longreads + Open Thread

Longreads

• Arpit Gupta was the only member of New York City's Rent Guidelines Board to vote against a rent freeze. He wrote about why. (My view has been that if you're going to control rents in NYC, the appropriate thing to do is to set a floor on annual increases for a given tenant. The city has thrived by bringing in talent from around the country and around the world, and should lean into that with an up-or-out model.) What rent control does over the course of multiple renewals is to create a subset of the housing stock that's cheap considering its location, but also low-quality considering the same—as long as NYC real estate is expensive relative to how much maintenance personnel make, you'd expect the housing stock to be unusually high-quality instead. But since renters are mostly paying for the capital they're borrowing, not for services bundled with it, there's just not much to cut. One of the problems with discussing rent control is that the upfront effect is very easy to see, and the long term effect is both more complicated to measure and harder to reason about. Wonks just reach a very different conclusion from the average voter.
• Joel Sobel: How to Count to One Thousand. The basic question is: if you have some information processing process which (for expository convenience) you can recursively apply to its own outputs—e.g. instead of counting to a thousand, you count out stacks of twenty and count your stacks—then you can start to reason about how fractal the process should get. And then this thinking can be extended to other processes that have some error probability per step. Which describes basically every multi-step process done by any person or organization. Simplifying it to the point of silliness is an illuminating exercise. (Via Elliot Lipnowski on Twitter.)
• Cory Doctorow has an interview in Jacobin on how to criticize AI. He makes a very revealing point here, noting that his science fiction career has taught him that who uses technology, and how, is more important than what that technology does. In one sense, this is so true it's tautological: we experience technology through its applications. On the other hand, this is very much a novelist's view of the world: the technology gives you a setting in which your characters can have interesting experiences. Whereas if you're living your life, not plotting a novel, one of the great conveniences of the price system is that you can be completely indifferent to whether your plumber is plumbing to pay for a boat, alimony, another addition to his rare book collection, whatever. You just care that the leak got fixed. The Doctorow view is very personal: at one point, he suggests that you should always be skeptical when your boss is excited about some technology. Which is actually a case where you want to think about the technology, not who's using it. If your boss is just a dummy, or is trying to weaken your bargaining position, by all means disagree. But if the bosses who are early adopters beat the ones who don't, you won't really have the choice of having an AI-obsessed boss or not. And if the whole thing is a boondoggle, the choice goes away that way, too. Which means that the winning move is to depersonalize the discussion: it's a technology, with certain characteristics, and will substitute for some things, be a complement for others, revalue still others, etc. Personalities and motivations will affect when and where that happens, but the outcome is more fundamental.
• And from the opposite end of the spectrum, here's former Trump admin AI policy guy turned OpenAI AI policy guy Dean Ball on the state of AI models. One important economic point he makes is that frontier models make their money right after they're released, because that's the time when they're the only way to get certain things done and they can charge accordingly. But that means that uncertainty about when they can be released can drastically cut the ROI. If you're going to be state of the art for a few months, max, then plausibly a delay of a few weeks eats a double-digit percentage of revenue, and can make the difference between a positive and negative ROI. So regulatory clarity, and a streamlined process, is essential. But then, he points out, there's a more complicated problem: unreleased models in internal use contribute a growing share to AI research, and wouldn't be covered under a release-based regulatory standard. So you can imagine a situation where the labs have models the government would never let them use, but that are invisible to regulators because there's no external customer. He ends up advocating a mixed public and private sector regime, which is probably the way things will shake out regardless: at some point in the near future, open-weight models with the right harness will presumably be able to replicate Mythos' capabilities. There's a mix of institutions that incentivize people to care about this in different ways, and they're all paying close attention.
• Stephen Witt visits a few robotics startups for The New Yorker. One fun point: a robotics founder says that "the ability to compensate for injury is one of the most important aspects of generalized physical intelligence." Having had a temporarily mobility-limiting injury recently, I spend a lot more time planning maneuvers like entering a car or picking stuff up off the ground. It's amazing how great muscle memory is, but you won't appreciate it until you've had to scrap your entire plan for getting leftovers from the fridge to the microwave. One of the debates in the article is on humanoid robots versus special-purpose ones. Some of this is a technical debate, but it's also an argument about timelines: the more humanoid something is, the more it's a drop-in labor replacement, as opposed to something you have to rearrange an assembly line around. (The book also makes a case for having people outside tech subcultures cover tech: "At one point, we were discussing a portable battery pack that Sleeper is designing for Neo. "It's actually a nontrivial problem," he told me. I asked him if he had ever used the word "nontrivial" before moving to Silicon Valley."
• This week in Capital Gains: why the dollar is different. Historically reserve currencies have derived their value from the expectation that they'd be redeemable in some precious metal that could, perhaps, be melted down and then minted into somebody else's currency. But the dollar is backed partly by the fact that the IRS expects payment in dollars, and partly by the related fact that dollars are a relatively easy currency to borrow in.

Books

Surely You're Joking, Mr. Feynman! (Adventures of a Curious Character ): One of my kids, who recently turned seven, has started getting seriously interested in electronics. If we're in the same room, he'll tell me about elaborate plans for drones, or try to bruit-force every breadboardable combination of LEDs, resistors, capacitors, photoresistors, and transistors. If we're in different rooms, and I can't hear him, there's a good chance that I'll find him disassembling an electronic toy with a screwdriver. It's like living with a Jawa. Anyway, it occurred to me that long ago, I had read some very entertaining stories about someone of a similar age, with similar interests, who had been a little more constructive and went on to accomplish some clever things. So I started reading him Surely You’re Joking, which does indeed open with charming anecdotes about Richard Feynman’s elementary school career as a radio repairman.

One thing I did not remember was how much of the book was devoted to seducing women. Almost every time he's talking about a woman who isn't a family member, it's a discussion of what to say to her at a bar so she'll go home with him, how to convince her fiancé to let her pose for nude drawings, or, when he's visiting a then-exotic foreign country, whether or not she's a prostitute. Sometimes, if you're reading aloud to kids, you can zone out a little bit. Not so with Surely You're Joking! As soon as you encounter a danger word (e.g. "she,") you're on high alert—our narrator is about to tell you how he solved yet another Two Body Problem. (From the perspective of 2026, you can see Feynman as an unanticipated beneficiary of the sexual revolution, and an illustration for why it entailed more tradeoffs than people thought.)

America would not be the country it is if it weren't capable of extracting strategically-indispensable work on the atomic bomb or Nobel-worthy physics research from shameless horndogs in addition to other types, and a lot of the book is about that process. A basic concept from programming is the if/then statement, and Feynman seems to have been constantly running an internal loop that asked: "If this is true, then what else must be or can't be true?" and constantly running experiments. Sometimes, he's trying to figure out beta decay; sometimes, he wants to know how ants find their way to his pantry; sometimes, he's curious about the efficient frontier between brute force and social engineering for safe-cracking?

One point Feynman makes, which many other very smart people do, is that some of his results were because he'd worked harder—but even that gets cast less as a statement about his work ethic (which feels like a moral judgment) and more about his curiosity (anyone can do it!). And he does portray himself as a deeply curious guy, who simply can't look at the world without formulating questions and can't come up with a question without inventing an experiment to answer it. One reason it's hard for smart people to talk about being smart is that, presumably, the subjective experience of working at the edge of your abilities is pretty similar for everyone. Whether you're trying to remember first-outer-inner-last or trying to prove Fermat's Last Theorem, if you're trying your hardest, you'll feel like that's how hard you're trying.

But in Feynman's case, it's a more plausible argument, because in addition to being good at physics, he cracked safes, played bongos, made art that people actually bought, and told some fun stories. These things probably use some of the same underlying traits. The safecracking, for example, was right between engineering and showmanship. He'd narrow down potential combinations by figuring out the tolerance of the safe (if you have to get one of 40 numbers in a row, it takes 64,000 tries; if it turns out that you just need the right number +/- 2, it's really 8^3 or 512, so you've lopped off two orders of magnitude). But he'd also find excuses to hang out in a coworker's office, idly spinning the dial on their combination lock, or come up with a sequence of funny notes to slip into their secret papers.

So, it's not the book to read if you want to be an amazing physicist, but it's a good one to read if you want to be habitually better at pretty much anything (or to be in physics what Feynman was in music and art: very impressive for an amateur and perfectly happy with that).

Open Thread

• Drop in any links or comments of interest to Diff readers.
• Any other books I should get for a seven-year-old who likes taking apart old electronics with his hands and assembling complicated drones in his head?

Diff Jobs

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

• Ex-Bridgewater, Worldcoin founders using LLMs to generate investment signals, systematize fundamental analysis, and power the superintelligence for investing are looking for machine learning and full-stack software engineers (Typescript/React + Python) who want to build highly-scalable infrastructure that enables previously impossible machine learning results. Experience with large scale data pipelines, applied machine learning, etc. preferred. If you’re a sharp generalist with strong technical skills, please reach out
• Ex-Citadel/D.E. Shaw team building AI-native infrastructure that turns lots of insurance data—structured and unstructured—into decision-grade plumbing that helps casualty risk and insurance liabilities move is looking for forward deployed data scientists to help clients optimize/underwrite/price their portfolios. Experience in consulting, banking, PE, etc. with a technical academic background (CS, Applied Math, Statistics) a plus. Traditional data scientists with a commercial bent also encouraged. (NYC)
• Lightspeed-backed team building the engineering services firm of the future is looking for founding members of technical staff excited about working alongside civil engineers to translate their domain expertise into the operating system that powers the next era of great American infrastructure. If you’re an engineer with strong product intuition, who's energized by access to users, and excited by the prospect of transforming how we design and construct our built world with frontier AI, this is for you. (NYC, SF or Remote) 
• A leading AI transformation & PE investment firm (think private equity meets Palantir) that’s been focused on investing in and transforming businesses with AI long before ChatGPT (100+ successful portfolio company AI transformations since 2019) is hiring experienced forward deployed AI engineers to design, implement, test, and maintain cutting edge AI products that solve complex problems in a variety of sector areas. If you have 3+ years of experience across the development lifecycle and enjoy working with clients to solve concrete problems please reach out. Experience managing engineering teams is a plus. (Remote)
• Series A startup that powers 2 of the 3 frontier labs’ coding agents with the highest quality SFT and RLVR data pipelines is looking for growth/ops folks to help customers improve the underlying intelligence and usefulness of their models by scaling data quality and quantity. If you read axRiv, but also love playing strategy games, this one is for you. (SF)

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.

And: we're now actively deploying capital into early-stage companies through Anomaly. Our focus is on defense, logistics, robotics, and energy. If you'd like to chat, please reach out.