Longreads + Open Thread

Longreads

• Felix Stocker has a nice essay on mining and society. Which sounds like the topic of the one humanities class a geological engineering major would grudgingly sit through, but which is actually a pretty pivotal question: many modern conveniences—especially including the batteries, windmills, electric motors,  and solar panels we use to reduce our reliance on emissions-heavy sources—require inputs that necessarily have to be dug up out of the ground, often in an environmentally-destructive way. When there's a debate over a mine, it's not so much big business versus the environment as it is environmentalism versus climate change and energy security.
• "Prop trading" can either mean elite systematic trading companies that don't take outside capital because their returns are so high, or it can mean businesses with a scammy or scam-adjacent reputation that ostensibly recruit independent traders and often rip off people who have dreams of having that job. Rob Carver analyzes some of the latter's offerings, specifically analyzing how these companies set up trial periods for traders. The usual setup is that they're given some target: make at least x%, lose no less than y%, over z days. As it turns out, at least one of these companies structures that task so it's basically indifferent to returns. Which is mysterious, until the last piece of the puzzle gets added: these firms are charging people a monthly fee over the course of their trial. The scammier they are, the less the details of the trial matter—in fact, they may want to actively select for traders who don't carefully walk through the math.
• In Fortune, Sharon Goldman profiles OpenAI's Greg Brockman. There's a running joke about how tech people will sometimes opine about topics well outside of their expertise, like inflation or culture wars or foreign policy or whatever, with excessive confidence. But one reason for that confidence is that some skills correlate across a variety of domains, and while excellence in one doesn't tell you everything, it certainly tells you something about someone's mental horsepower, focus, ability to identify the core issues behind some problem, etc. This piece is a datapoint in that direction, because Brockman is very hands-on technical ("Greg basically hacked together the first API one weekend, I think over Christmas,"), but is also negotiating the details of OpenAI's gigantic business development deals. Of course, it's an open question whether those deals will work out or not, but even getting them done is a challenge.
• Mark Humphries in Generative History has a fascinating piece on Gemini decoding centuries-old handwritten records in a very human-like way, by using context clues in the document to infer missing information. In a sense, the LLM's transcription was more than 100% correct, because it identified and fixed an ambiguity in the historical record (even expert human readers will occasionally miss something like this). One of the unique axes on which models perform well is that they don't get bored the way a person would, and are willing to check their work to make sure it's logically consistent even when the task is just to transcribe text.
• And on a similar note, this Dwarkesh Patel and Dylan Patel interview with Satya Nadella has an interesting side note on legibility: Nadella notes that AI makes it easier to move information from an Excel file into a real database, and that means it's easier to join across different datasets. Cheap determinism is a complement to more flexible but uncertain LLMs. Future historians will have a much easier time trawling through historical data, at least as long as someone pays to store it.
• In this week's issue of Capital Gains, we look at the economics of deals where both sides are basically indispensable to the other. They're a real-world instantiation of the "ultimatum game," and, as in that game, the theoretically optimal behavior is not what prevails in the real world, for good reason.
• This week in the Diff-bot chat, a user had a great question about interest rates and demographics. My model is that as societies age, it pushes real rates down, for two reasons: first, people approaching retirement accumulate lots of savings, and slowly tilt them towards lower-risk assets. And second, lower family formation means less demand for housing, and mortgages soak up a lot of the duration-seeking but credit-risk-avoidant capital out there.

Games

Switched up leisure time-sinks this week—let's talk about Europa Universalis V. This is the latest entrant in Paradox Interactive's series of grand strategy games revolving around world conquest, in this case covering the time period from 1337 to 1837.

The previous game in the series came out a decade ago, and while it's a wonderfully complex simulation, it's locked into some abstractions that separate it from what it's trying to simulate. That's true of any game, but one of the measures of a simulation is how well it trades off between fidelity to what it's simulating and fun. Over time, EU4 has tilted towards fun, specifically of the map-painting variety—the way the game is set up strongly encourages rapid conquest; if you're playing France, it's a good idea to expand into Iberia and Italy, get a toehold in England, start working your way through North Africa, etc. Over time, they've added downloadable content (more on this later) adding ever more entertaining ahistorical outcomes, from a resurgent Eastern Roman Empire to the very meme-flavored alternative history where the Teutonic Order becomes a nomadic horde of religious fanatics, slaughtering/pillaging/converting their way across central Asia. But historically, there's a very real sense in which France didn't even conquer France until late in the game's timeline. It's a bit immersion-breaking when you're supposed to be a 16th-century monarch and you're carefully micromanaging land you've conquered on the other side of the planet. That just wasn't practical with that era's technology or institutions! And this is common for historical grand strategy games. Giving players enough direct control over every aspect of a country is ahistorical, and leads to fun novelty strategies like "the chief of the Golden Horde converted to Christianity, became Holy Roman Emperor, and finished conquering the entire world by 1527."

EU5 basically goes in the opposite direction; it's a simulation of the shift from early, weak states to centralized modern ones, and from economies completely dominated by local agriculture to manufacturing and global trade. At every opportunity, where EU4 had a discrete binary—a province is 100% Orthodox until the day a missionary finishes his work, at which point it's 100% Sunni—EU5 has continua. They track populations in each province, with a mix of different cultures, religions, and social classes. Populations shift between these, sometimes inertially but often through player intervention. And they model states' ability to project power by tracking how much control the central government has over any given location. That control level starts at 100% in the capital, but as it radiates outward, it declines—unless you build roads, or have a navy patrolling the coast, or invest in placing government representatives in these various far-flung locations. (This leads to its own funny emergent properties; Hungary's capital is far enough from their borders, and the country's economy happens to be configured in such a way that they don't trade much, so because of how the game models the diffusion of scientific knowledge, Hungary can end up being an isolated intellectual backwater. That's not especially historical, but it's really cool that "This place doesn't have access to modern technology because they don't have enough transportation infrastructure to expose people to it," a real historical phenomenon, is an emergent property of the game's design.)

It's impressive that they've managed to both make the game reflect the historical reality that pre-modern states exerted very little control over their populace, without making the game incredibly frustrating. It's still a little frustrating, but good games basically deliver a drip of structured frustration, where there are interesting obstacles put in the player's way, and interesting choices for how to solve them.

In EU4, one of the game design problems was that there wasn't much to do during peacetime (the main way to solve this is to get big enough that you're always fighting a war, or several, while preparing for the next few). They had a lightweight economic model, with different commodities produced in different places and trade routes that would shift over time—one emergent phenomenon the game didn't have to hard-code was the relative decline of Venice as a center of trade, as more Asian imports were routed over the ocean rather than land. But specific commodities didn't make that much of a difference—other than grabbing gold and cloves when the opportunity presented itself, random events like the Reformation reducing the price of incense had basically zero impact on the game. EU5 goes in the opposite direction. One of the core loops in the game is cranking out more lumber and stone so you can build tools and masonry, so you can build paper mills, alum mines, and a dye industry, so you can crank out scriptoria that produce books which can be used to build universities and libraries so your populace acquires the knowledge and wisdom necessary to slavishly obey their ruler. War is more of an inconvenient interruption to assembling a large industrial base.

What you're left with is a detailed interactive simulation of early modern economies, which will of course take all kinds of liberties with the truth but still have a system that produces realistic outcomes: a gradual decline in fragmentation, seafaring empires having more far-flung provinces than land-based ones, mountains marking natural borders, etc. All in all, an impressive accomplishment.

The usual story with Paradox is that they deliver an incredibly polished, compelling gaming experience, generally two to five years after release date. But this lends itself to a business model where they release a game, and then start shipping additional content that users have to pay for. In general, for pretty obvious commercial reasons, the paid content unlocks new mechanics and country-specific flavor that makes it easier to win. (Or, at least, easier to win eventually; when they offered a new content package for the wildly popular Eastern Roman Empire, the setup they used was that at first, the country is actually weaker than it looks on paper, but a skilled or lucky player who pulls off a few early victories gets a lot of momentum). For EU4, the game was priced at $40 at launch, and if you insist on buying all of the expansions, you'll spend $370(!). Or you'll subscribe for $8/month. This is actually a very user-aligned model, at least if you think alignment with users is "keeping them entertained" and not "ensuring that they put their non-work/non-sleep time to its highest and best use." For a product that appeals to a nerdy demographic, caveat emptor is probably the best approach.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• What are some other better-than-they-strictly-needed-to-be simulations out there?

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.
• 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)
• Series-A defense tech company that’s redefining logistics superiority with AI is looking for a MLE to build and deploy models that eliminate weeks of Excel work for the Special Forces. If you want to turn complex logistics systems into parametric models, fit them using Bayesian inference, and optimize logistics decision-making with gradient descent, this is for you. Python, PyTorch/TensorFlow, MLOps (Kubernetes, MLflow), and cloud infrastructure experience preferred. (Salt Lake City or NYC)
• A hyper-growth startup that’s turning the fastest growing unicorns’ sales and marketing data into revenue (driven $XXXM incremental customer revenue the last year alone) is looking for a senior/staff-level software engineer with a track record of building large, performant distributed systems and owning customer delivery at high velocity. Experience with AI agents, orchestration frameworks, and contributing to open source AI a plus. (NYC)
• Well funded, Ex-Stripe founders are building the agentic back-office automation platform that turns business processes into self-directed, self-improving workflows which know when to ask humans for input. They are initially focused on making ERP workflows (invoice management, accounting, financial close, etc.) in the enterprise more accurate/complete and are looking for FDEs and Platform Engineers. If you enjoy working with the C-suite at some of the largest enterprises to drive operational efficiency with AI and have 3+ YOE as a SWE, this is for you. (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.