|
In this issue: - The Autistic Half-Century—A significant economic change over the last half-century or so is that the bundle of traits described as high-functioning autism has gotten much more economically rewarded than it was in the past, and this has compounding effects for both how many autism-friendly career paths there are and, over longer periods, how common autism is.
- Burn Rates—The higher your expected growth, the higher the optimal level of cash burn until then.
- Round-Tripping—ASML makes a very indirectly strategic bet.
- AI and IP—Anthropic sets a rough market value for IP—having your book included in an LLM's corpus is equivalent to selling about a thousand copies, but probably closer to selling a million people 0.1% of a copy.
- IPOs and Network Effects—It's hard for anglophone countries to maintain their own stock market.
- Spinoffs—Public markets replicate the PE portfolio structure.
The Autistic Half-Century
2025 has been a good year for valedictory memoirs from some of the 20th century's great executives, like Bill Gates' Source Code and John Malone's Born to be Wired. Gates has a few cameos in Malone's book, as a business partner and investor. One reason the two of them got along pretty well is that they had something in common: each one notes, early in his memoir, that he's autistic, and that his life (and his parents'!) would have been easier if he'd known that earlier.
When people talk about "autism," they're describing a wide spectrum of outcomes, including pretty serious disabilities that prevent severely autistic people from living on their own. When they use the term to describe themselves, they're referring to the more benign part—a combination of poor social skills, odd sensitivities (to food, texture, noise, etc.), and the ability-or-compulsion to focus for long periods on very narrow domains. For the purpose of this piece, by "autistic" I'm generally referring to the part of the spectrum where people can hold down jobs and live independently, and not the more severe varieties. This used to be called Aspergers, but that incredibly useful, motte-and-baily-crushing distinction has been deprecated for some reason.
This set of traits has existed for a long time; Heraclides of Pontus describes the maritime equivalent of being obsessed with trains in the 4th century BC. Isaac Newton was a vividly weird guy who spent more time on biblical and alchemical studies than on math and physics. (He was also a very literal, subtext-insensitive kind of person; he got a sinecure as Master of the Mint partly because the government wanted to reward him for his academic contributions, but insisted on actually doing the job, hunting down counterfeiters, etc.)
One reason to suspect that many public figures have an autistic streak is a two-sided selection effect: they're more often in circumstances where it's natural and normal for them to do what would otherwise be autistic. Describe Warren Buffett as someone who periodically has six-hour Q&A sessions with complete strangers, and he sounds pretty extroverted, but if you rephrase that as six hours of continuous riffing on a topic that's been his fixation since he was a teenager, and it sounds more like an autistic person who's been able to arrange his life so it's as autism-friendly as possible. The way I'd think of it is that if there's someone who spends forty hours a week messing around with model trains, they sound pretty autistic. But if you meet that person because you're at a model train convention, 1) it's 100% on-topic for them to start up a conversation with you by talking about model trains, and 2) you're already about 99th percentile for being interested in model trains, so they just don't look that weird to you. Similarly, lots of the people who strike up conversations with Gates or Malone are people who want to talk about their special interests. And when they were doing deals for set-to boxes and equity injections, a maximum-bandwidth bidirectional infodump was probably great fun for both of them.
What would someone born with those traits have done if they'd been born a century earlier? There were certainly plenty of careers that reward this bundle of traits. But they were less scalable. Presumably lots of mildly autistic people sorted themselves into more solitary, systematized career tracks, and many others were part of the vast majority of people who lived on farms. Urbanization creates opportunities for more specialization, which raises the potential returns of spending basically all of your brainpower thinking about what you specialize in. (The Pattern Seekers is not all that great, but it does mention some anecdotes from the early life of Thomas Edison, who did very well for himself in terms of historical impact and had financial outcomes that weren't too shabby, either. So he was an early beneficiary of this: more cities meant more demand for power plants, lightbulbs, and appliances.) Mass production and capital accumulation raised the returns for people who can think about complex systems, as did logistics networks like railroads—a business where day-to-day management could mean lots of interaction with telegraph machines (Carnegie’s first love, he could understand the incoming clicks by sound alone) rather than human beings, and a field that rewarded people who were good at mental arithmetic and could keep lots of datapoints in their heads.
But the period where traits on the spectrum were maximally rewarded probably starts in the 70s, thanks to a series of mostly unrelated developments:
- The PC gave many, many people with autistic traits something to fixate on, and this had a compounding impact—the time you spend getting an Apple ][ to do exactly what you want is time your peers are spending developing social skills, playing sports, etc. So, depending on the age at which it clicks for you that computers are unbelievably cool, you can end up burning the ships and having no real choice but to get good at them.
- The growth of global trade, whose share of GDP more than doubled from the 50s through the 80s, meant increasing country-level specialization, plus the meta-level specialization of managing supply chains. The more you think of things in terms of where it's most cost-effective to assemble a television set rather than what Bob down at the plant is going to do if he loses his job, the faster you'll outsource manufacturing. Meanwhile, the collapse of the postwar Bretton Woods system meant that currencies fluctuated more, and that capital moved more freely.
- All that volatility also started shaking up financial markets; more money sloshing around means more liquidity, which means there are more markets that are viable to trade in and that smart traders can take bigger positions. The rise of hostile raiders and high-yield bonds in the 1970s and especially 1980s fits an autism-adjacent approach: big banks operated with a sort of clubby understanding about what behavior was and wasn't becoming of an investment banker, but lots of smart, ambitious people at Drexel and the like noticed that these rules were not written down anywhere, and that it was actually pretty lucrative to just kick underperforming managers out and find someone competent to run the company, even if this might earn you some glares on the golf course later. The entire pitch for high-yield bonds was a "Well, akshually": their interest rates more than compensated investors for the added default risk, and moreover all else being equal a higher-rate bond has less duration risk, which was something very salient to investors who'd survived the rates environment of the prior decade. 4% underwriting fees are plenty of compensation for dealing with snide remarks from white-shoe firms about "junk bonds."
- CRSP started publishing historical data in 1962, Nasdaq launched in 1971, the Black-Scholes formula was first published in 1973—each of these are plausible start dates for trading and investing starting to get more systematized. If you can trade electronically, you don't have to restrict yourself to one specific trading floor, and if you're the fastest, you can be fastest at many different things. So instead of a trader's skillset being a mix of personal relationships, industry knowledge, handicapping instincts, and a healthy appreciation for risk, these skills could get narrowly-specified. Someone who's good at objective analysis but too timid to bet big, or too stubborn to cut losses when they're wrong, can be a lot more useful if they're doing the research and somebody else is constructing the portfolio, setting risk limits, etc.
- Of course, if a PC is a bicycle for the autistic mind, the Internet is a hypersonic jetpack. Many people who launched and grew social networks were themselves quite socially-awkward, which meant that socializing online for them was relatively easy but also that they were better at converting social interactions into PHP scripts because they had to think about them explicitly. An autistic person's social script can't match the performance of a neurotypical's, but the autistic person's code has well-named variables and thoughtful docstrings. And many of the services that these people created also make life easier for autistic people. You can DoorDash food without talking to another human being, you can get product recommendations without asking anyone to recommend a particular product to you, you can use LLMs for solo therapy sessions, and Uber literally has a setting you can select that tells the driver to not make small talk. All of these are in one sense pretty minor affordances, and none of these products would be best-described as targeting the autistic market explicitly, but they do allow the loneliness that those with autistic traits tend to deal with to be more comfortable.
The analog world is built around neurotypical preferences. How could it not be? If you hate how bad and loud the music is in various public places, remember that these places are made for the public, and that's basically what the public wants. The digital world is built by and for autistic people, and if you want The Internet to be a place where you spend all of your time reading about Rust, or the Eastern Roman Empire, or certain practical applications of matrix multiplication, you can more or less arrange your online life so these are all mainstream interests.
This framing even describes some of the conflicts that modern tech companies have engendered. People complain about The Algorithm showing them rage-bait news, snuff films from Ukraine, scantily-clad women, scantily-clad men, etc. But The Algorithm is, basically, whatever you seem likely to engage with based on your past behaviors and those similar to you. The neurotypical user says "This isn't what I wanted," and the autistically accurate response is "But it's exactly what you asked for."
You should expect this to get more common. Autism diagnosis rates have risen over time, so in terms of pure numbers they'll have a bigger impact. Part of the rise of autism is just that we have better diagnostic tools. Malone and Gates probably received informal diagnosis of autism-spectrum disorders on the playground, in the form of being called a "geek" or "spaz" or something, and at least some of the growth of autism as an identification is that it's not as pejorative as it used to be. But it's also quite heritable, so if people with autistic traits start being more likely to have kids together—perhaps because of assortative mating by education, which rose from the mid-20th century through the 90s before declining; perhaps because couples are meeting online; Autistic couples are also more likely to read Bryan Caplan, take him as gospel, maybe run a few monte carlo simulations, and commit themselves to maximizing the net present value of future visits-from-the-grandkids; maybe they nerd out about religion and read-and-reason their way into believing that any form of birth control is intrinsically evil. If the material rewards from these traits keep rising, they won't be as economically constrained as other couples: fertility rates tend to be u-shaped with respect to income, and higher within-couple variance lowers the opportunity cost of one parent downshifting their career for a while.
The Internet, the more autism-first way of viewing the world, is increasingly interposed between you and the external world. You'll still talk IRL—about things you read about online. You go out to restaurants—that you saw on Google Maps. As smart glasses get more common and interactions with LLMs get more ubiquitous, an increasing share of reality will be filtered through consumer apps, meaning that everything you see will be filtered by an autistic person's idea of what a neurotypical person would like. So, in the end, there's a paradoxical separation that leads to synthesis. Someone who was born on the spectrum a century ago would probably have a slightly worse life because of it, and feel a bit separate from their peers. They might, through effort and luck, construct a wonderful life for themselves, but the world didn't provide a lot of the raw material to make that easy. Today, they're much more likely to get sorted into a peer group in which they're comparatively normal, and to be able to filter out some of the rough textures and annoyances of a 50th-percentile life. But in another sense, all this separation makes them more integrated with society than they've ever been. A disproportionate share of the software and hardware I used to write this article, and that you're using to read it, was produced by people on the spectrum. You are, in some tenuous but meaningful sense, interacting with all of them. And, mercifully, you aren't even attempting to make eye contact.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - A startup is automating the highest tier of scientific evidence and building the HuggingFace for humans + machines to read/write scientific research to. They’re hiring engineers and academics to help index the world’s scientific corpus, design interfaces at the right level of abstraction for users to verify results, and launch new initiatives to grow into academia and the pharma industry. A background in systematic reviews or medicine/biology is a plus, along with a strong interest in LLMs, EU4, Factorio, and the humanities. (Toronto, Remote)
- Ex-Optiver/DRW quants with over a decade of experience in HFT and AI are reimagining time series forecasting from first principles. They are building a research lab, initially monetized via derivatives trading. The team is hiring a founding engineer (Python/C++/Rust; distributed compute, ML infra) and a founding AI researcher to rethink how machines model the future. No finance experience needed. (SF)
- 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)
- 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. Elsewhere
Burn Rates
OpenAI has modestly adjusted its projected cash needs through 2029, to $115bn from $35bn ($, The Information). AI bearishness takes two forms: the demand-side argument that the products just aren't that great and will never be that big a business, and the supply-side argument that AI is amazing but the space is too crowded and overcapitalized for anyone to make money. This update is actually net worse for the first argument, but helpful for the second: the reason OpenAI's projected losses are rising so fast is that their revenue numbers are rising fast. Every time revenue goes up, there will be another set of inputs where scale effects mean they're worth doing in-house. OpenAI had no business owning fabs or power plants a few years ago, or designing their own chips, but if doing that today means getting the compute and the power they need to meet the future demand they expect, then it's worth it. This shifts costs earlier in time, because it's creating a fixed cost to replace a variable cost. So far, OpenAI's growth is answering the demand-side question quite nicely, but they'll need to set a lot more scaling records to justify another $80bn in spending to get there.
Round-Tripping
ASML has become a big strategic investor in Mistral. This is very fun from the perspective of commoditizing the complement: if there are more AI labs, there will be more demand for GPUs, and probably more skunkworks projects to develop custom chips. If some of those competing companies are open-weight, the factor that determines marginal cost for the cheapest models will be purely on the hardware (and power) side. So ASML is doing what it can to ensure that demand for the most semiconductor capital equipment is as high as possible, but also that the end market is as fragmented as possible.
AI and IP
Anthropic has settled a class-action lawsuit for $1.5bn over their decision to pirate some of the half-million or so books they trained on. This works out to about $3k/book, or what a typical author would get if they moved another thousand or so copies of their book after having earned back their advance. This actually seems reasonably fair. Just as a ballpark, the collective very partial attention of all of the users of an LLM seems like it would be equivalent to selling around that many extra copies of a book. As AI companies get bigger, there will have to be more refined ways to think about how to split the upside from people providing training data, but there's a possible model where LLMs end up being an active subsidy for very niche writing, because it will have a hard time finding its audience in bookstores but can't help but find that same audience in the context of an LLM chat session.
IPOs and Network Effects
So far this year, 231 companies have gone public in the US, and 7 in the UK ($, FT). Which raises the question: how global should the stock market really be? Within countries, there's typically consolidation in listing venues over time; trades happen where there are already other trades happening. On the other hand, even within the US that leads to some frictional costs, and the financial industry professionals on the West Coast who get to the office before 5am might fondly wish that a few of the trillion-dollar West Coast-based companies they track were listed on the Pacific Stock Exchange, with West Coast hours, instead. That problem is a lot harder if Japanese and market professionals have to work the night shift to trade Japanese stocks. So there's still some regional gravity for exchanges. But many of the biggest companies in a given country are multinationals, and if the investor base is English-speaking and assets tend to get managed from the US, it's generally easier for these companies to have a US listing. So the same cultural and economic ties that make UK-based companies bigger than they'd otherwise be have, over time, turned the London Stock Exchange into something closer to a regional exchange.
Spinoffs
One of the drivers of private equity's growth is that they can lever up at the company level, and have a bunch of higher-volatility bets with capped downside, which makes their portfolio perform better. Levering a company is economically similar to owning an option on in, with the strike price equal to the value of the debt, and if that's combined with streamlining a company to focus on a narrower business, it can lead to some exciting returns (and a few blowups). but public markets can do that, too, and there's been a burst of spinoffs lately ($, WSJ). Because of how low fees have gotten and how liquid the assets in question are, diversification is cheaper for most investors than making single-name bets, which means the right structure for the market is have diversification at the portfolio level and to make each individual stock as focused and idiosyncratic as possible.
|
