Longreads + Open Thread

Longreads

• Brian Potter has an engineering history of the Manhattan Project. There are some commonalities with capital-intensive buildouts, but one point Potter makes is that there's a reason we don't have Manhattan Projects for everything: the initial pace was fast because there was, briefly, an atomic bomb race with Germany, and nobody wanted to be in second place. And then, later on, there was an institutional imperative to get a working bomb both to speed up for the war in the Pacific and (not that many people would openly express this one) a keen desire not to have retrospectively wasted unprecedented sums on something that turned out not to contribute to the war. One of the features of a war economy is that the government can commandeer immense resources (coincidentally, last week's Diff mentions something this piece also alludes to, that the government used its silver reserves as a substitute for copper). But another difference is that any steel, copper, labor, etc. allocated to R&D is material that isn't used for immediately practical applications, so the opportunity cost of this research includes loss of life—either directly, when soldiers aren't equipped as well as they could be, or indirectly when the war is drawn out. One of the great achievements of the entire project is that everyone managed to keep their heads on straight despite these incredibly high stakes.
• Jerry Neumann recently retired from investing, but has fortunately not retired from writing: this piece in Colossus is a look at where value is actually captured when new technologies are deployed. He points to the shipping container as an example of a new technology that created vast amounts of wealth, but not for the people who actually owned and operated containerized ships. Instead, the big winners in the US were retailers who suddenly had access to much cheaper goods, and East Asian manufacturers who could sell them for less. But, one of the reasons for this is that the early container ship companies, like SeaLand, didn't scale fast enough to lock down the market: "The business ended up being dominated primarily by the previous incumbents, and the margins went to the companies shipping goods, not the ones they shipped through." That might be a function of market structure, but it could also be a function of capital market structure—if there had been aggressive pools of capital like Softbank, crossover funds, the largest VC firms, and sovereign wealth funds, it's possible to retell that story with a happier ending where the earliest container shipping companies achieve enough scale to have an unbeatable lead, and really do capture the upside.
• Santi Ruiz interviews Dean Ball on how the Trump White House really works. There's always an unbridgeable gap between the org-chart/employee-handbook model of how an organization works and the question of who calls whom when something needs to get done, and this piece is a nice illustration of it. Even very centralized, top-down organizations have cross-currents, especially when everyone's theoretical responsibilities were established under different circumstances and every new situation interacts with multiple power groups. It's sometimes depressing to read about how much of DC behavior is about maneuvering for power rather than enacting policies. But, how could it be anything else? On any topic, there will be some entrenched interests that understand how things work and some insurgents with a view on how they could work, and DC's function is to figure out the right way to make these two groups meet in the middle.
• Tim Hua has a fantastic piece on LLM psychosis. There are some striking anecdotes out there about people driving themselves various kinds of crazy by doing what the LLM probably categorizes as fictional role-play and what they believe is unraveling the secrets of existence. But, instead of handwringing, why not figure out how big a deal this problem really is, and how chatbots vary in their handling of it? That's exactly what Hua did, by getting chatbots to role-play a user who's going gradually deeper into psychosis, to see how different bots respond. It's very hard to run controlled experiments on LLMs, but even if it's not a perfect proxy for how everyone thinks, it is, definitionally, a good proxy for the thoughts they bother to articulate.
• From Out of Pocket, a piece describing a phenomenon that's interesting-by-omission: if AI radiology is so great, why hasn't AI replaced radiology? This article should not be especially cheery to radiologists, because what it describes is more a market inefficiency than a technology problem: most of obstacles presented have to do with perception, market structure, and legal liability. The piece does make the interesting note that we don't have enough data on weird medical events, and that actual humans have a good enough world-model to handle the weirder cases. So, even in a model where AI does eat this professional specialty, there's a brief window where providing training data for the model is an efficient way to sell a talent for its liquidation value.
• In this week's Capital Gains, we consider the question of whether or not to trust the experts, and what it means to ask it.

Books

Prime Movers of Globalization: The History and Impact of Diesel Engines and Gas Turbines: It seems weird to describe a sober work of economic history as psychedelic, and yet reading this book allowed me to look at everyday objects and see fractal complications emanating from them in more dimensions than you can imagine. Globalization was partly the result of institutional factors, but many of those were downstream of technological changes that made them possible and necessary.

Smil is writing economic history from an engineering perspective, and that's a useful one to have. The ultimate answer to "what's the cheapest way to produce an additional ton-mile of shipping" is a rate-limiter on the entire system, in much the same way that human constraints like our low productivity at ages 2 and 92 is a defining feature of our economic and social systems.

One of the technologies he highlights is the diesel engine, which is the most efficient practical reciprocating internal combustion engine and thus, for its applications, determines how much oil, refining capacity, pipeline infrastructure, etc. it takes to move things. Gas turbines are his other prime mover, with a high power-to-weight ratio (making them good for aviation) but whose efficiency is highest when it can be used at steady load (making it good for power generation).

In both cases, there's a long path to production, starting with abstract theory and amateur prototypes eventually converging on a working product. In both cases, the path was gradual; Rudolf Diesel had the right general idea, but his original design had some serious flaws that he had to work through. Gas turbines were theorized much earlier than they were actually designed, because they depended on alloys and precision manufacturing that didn't become widely available until the 20th century.

So, in one sense, this is a book about path-dependence. Global trade was still viable when it was powered by sails, and grew plenty when coal first became available. But we'd live in a much poorer world if that's where things had stopped. And it wasn't obvious in advance that better engines would be developed, or what the underlying dependencies were, which is true today—drone delivery might have been motivated by the faintly ridiculous inefficiency of running errands in a car, where you might use a 2,000-pound vehicle to get to the pharmacy in order to pick up a few dozen milligrams of some active ingredient. But that was just a funny observation until smartphones created so much demand for lightweight, low-power components. So this book ends up diving into the specifics in order to defend a general term: the richer the world gets, and the higher real output is, the more we can afford to experiment and the more broadly we can amortize whatever those experiments lead to.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• We're going to do a piece on LLMs in healthcare, and would be delighted if any readers who work in the healthcare field would be open to chatting, off or on record, about how they and their colleagues are using AI. (We'd also be interested in cases where non-specialists were able to diagnose or address specific health issues their doctors missed, but we're mostly focused on how medical professionals are thinking about it.)

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.
• A transformative company that’s bringing AI-powered, personalized education to a billion+ students is looking for elite, AI-native generalists to build and scale the operational systems that will enable 100 schools next year and a 1000 schools the year after that. If you want to design and deploy AI-first operational systems that eliminate manual effort, compress complexity, and drive scalable execution, please reach out. Experience in product, operational, or commercially-oriented roles in the software industry preferred. (Remote)
• Fast-growing, General Catalyst backed startup building the platform and primitives that power business transformation, starting with an AI-native ERP, is looking for expert generalists to identify critical directives, parachute into the part of the business that needs help and drive results with scalable processes. If you have exceptional judgement across contexts, a taste for high leverage problems and people, and the agency to drive solutions to completion, this is for you. (SF)
• Thiel fellow founder (series A) building full-stack software, hardware, and chemistry to end water scarcity, is looking for an experienced software engineer to help build the core abstractions that enable global cloud seeding operations - from mission planning to post-flight analysis. If you have 5+ years building production systems with complex integration requirements, please reach out. (Los Angeles)
• 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)

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.