Longreads + Open Thread

Longreads

• Ryan Moulton has a beautiful piece on the limits of screens, and the human eye, in seeing color. It's a quirk of evolutionary history that mammals went through an evolutionary bottleneck in which they were mostly active at night, so we ended up with relatively worse vision, particularly color vision compared to birds. And we've faced separate hardware constraints when we determine what range of colors to display on screens. This piece is wonderfully cross-disciplinary: it talks about the physics of color, evolution, the history of computer standards, and what to take a closer look at when you're walking in the woods.
• A few months ago, Zilian Qian wrote this piece in ChinaTalk about the black market in Claude tokens. It's always interesting to see complex illicit supply chains, because the main form of contract enforcement is reputational damage, and where there are many opportunities to take advantage of people. Model resellers are basically providing a pseudonymized (to the buyer) source of tokens, and apparently they sometimes substitute worse-but-cheaper models. But at the same time, this business appears to be subsidized by Chinese labs' demand for distillation. Which makes pricing the advanced models even more complicated: if there's a big price premium for a small quality premium, it actually subsidizes the least trustworthy of the token-resellers, and makes the market more of a market for lemons.
• Chris Gillett asks why the big problem in powering datacenters is connections to the grid, not power generation. There's a sense in which every economy, whatever its formal description, is a mix of a market economy and a planned economy, just at different levels. In sectors with relatively few externalities, all of that central planning takes place within firms, but for something like electricity—a complex supply chain prone to monopoly, and something that went from a luxury to a human right over half a century—there's more regulatory intervention. And some of it has been less flexible than it could be. In more purely market-based systems, a bottleneck shows up in prices and those prices are both information and a bounty for acting on it. It's completely understandable that since US electricity consumption flatlined a decade ago, our grids are administered on the assumption that very little happens, and it happens with substantial advance notice. But that's suddenly not the case.
• Saloni Dattani has a retrospective on why the Covid vaccine was approved so quickly. This is actually a positive version of the story above: there was a system that, originally for very good reasons, had a long, sequential process for approving drugs. But parallelizing is a great way to get things done faster, and caching things you'll need ahead of when you need them is another nice trick. It turns out that the healthcare establishment contains multitudes, and some of them had been ploughing ahead with research on mRNA vaccines generally before we found the ideal use case. Covid probably shouldn't set the benchmark for how the average drug approval process works—the piece lays out many of the Covid-specific things that made it unusually fast. But it's a good reminder that some systems can be surprisingly effective when given the right goals.
• Noah Smith wonders if his writing has an impact any more, because populism weakens the influence of wonks, paywalls mean that you can be an independent journalist at the cost of preaching to the choir, and AI-generated content soaks up attention. I'll take the other side of all of them: populist movements do plenty of dumb things, but they still need some kind of internal elite that actually implements things, and that elite does read; sufficiently popular articles can break containment (I.F. Stone, who was writing independent political newsletters before it was cool, broke the story that the Gulf of Tonkin incident had not been described accurately); and readers—the good ones—still want to know that they're reading something written by a human. (If you must publish AI-generated content, please just share the prompt. That way, I can have ChatGPT write your essay in the context of my interests.)
• In this week's Capital Gains, we consider the long path to IPO. It's not just Sarbanes-Oxley, though that contributed; the financial system has evolved so different kinds of investors specialize in different stages, and the ones who can do multiple stages are also big enough that it makes sense for them to do both. In some ways, we're still overcorrecting from the dot-com bubble. Then again, how many people really bought and held Microsoft? How many bought what they thought was the next Microsoft in 1999? Overcoming the wealth destruction of the dot-com/telco bust is a very high threshold.
• A Read.Haus reader asks if we are the paperclip maximizers, consuming all of our natural resources to make AI. This is a nice way to push back on the idea that AI is aligned by default with the profit motive, because it's too expensive. If AI can run on cheap consumer hardware, unfriendly AI is more of a risk, but if you somehow need to secretly raise a hundred billion dollars, the list of potential supervillains, both people and nation-states, is pretty short. But part of being aligned to the profit incentive is being aligned to the regulations that already govern it; we already set limits on natural resource use, and when there's a big swing in the value of some resource, there also tends to be regulation on how quickly it's used. AI is increasingly politically unpopular, so the risk is probably in the other direction: poorly-considered laws that are meant to do something about AI, rather than well-crafted ones that treat its existence as inevitable but the outcomes of that existence as subject to human input.

Books

1873: The Rothschilds, the First Great Depression, and the Making of the Modern World: In the summer of 2007, a book came out detailing the crash of 1907, when an opaque banking system dependent on short-term financing to back longer-term loans suddenly unwound. A few months later, the S&P 500 hit an all-time high and then pulled back; a year after that, the financial system was on the brink as financial intermediaries dependent on short-term funding discovered that they couldn't roll it over.

So, it's with great trepidation that I cracked open a book about how a massive capex spree that led one industry to completely dominate equity market performance suddenly evaporated into a crash.

Unfortunately, the parallels continue: what sets the scene for the bubble of the early 1870s is a massive episode of credit expansion driven by the government's response to a crisis: after the Franco-Prussian war, France was forced to pay enormous indemnities to Germany, equal to about 4% of Germany's GDP. Germany had followed a lopsided economic development model, where it still had a fairly agrarian economy, with relatively little manufacturing and a small financial sector, but had a very modern army. Acting on the principle of comparative advantage, they went after a neighbor who had those complementary advantages. But the German economy just didn't have useful ways to absorb that liquidity, and it eventually went where credit expansion usually goes: into real estate. France, meanwhile, had a surprisingly easy time issuing the bonds it needed to pay off the indemnity, and was actually able to pay it early (even more liquidity!).

Today, you can pretty straightforwardly split the world into rich countries, middle-income, and poor, and periodically countries outside of the rich category will go through an economic cycle where investors are betting that they'll converge. The late 19th century equivalent of this had the UK as a special category—much in the way that the US is at this point economically distinct from other rich countries—and sometimes, other places would make a serious effort to catch up. The book points out that France was surprisingly successful at this, with a slight twist: the British financial system was unusually good at financing companies globally, while France specialized in the more strategically valuable business of financing foreign countries. Austria went through a land boom similar to Germany's, and, like Germany, ended up over-levered.

One of the most familiar bits of the whole narrative is global contagion: the Austrian and German booms collapsed, but America's railroad boom kept on going—for a little while, until an underwriting by Jay Cooke failed, Cooke's company went under, and suddenly the US was in the same deflationary crisis as Europe.

But the other thread of this narrative is that in a world with smaller financial systems, liquidity is more of a finite resource. If money is supposed to be backed by precious metals, then credit expansion is relative to a fixed supply, and any time investors get skittish, they'll rapidly demonetize anything that isn't gold or silver. Now, we have a bigger financial system that can sustain even larger global financial flows, but we also have central banks that can inject whatever liquidity the system needs. That's true as a consequence of the deflationary late 19th century, and it means that while some elements of history can repeat, some bugs in the system have been patched (and, of course, these patches have introduced brand new bugs that tend to reveal themselves whenever equities are 25%+ off their peak).

Open Thread

• Drop in any links or comments of interest to Diff readers.
• Noah's essay has some good thoughts on the role of modern punditry. Who else has written interestingly on this?

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.