Longreads + Open Thread

Longreads

• Victor Niederhoffer died this week. Here's John Cassidy's New Yorker profile of him from 2007. Niederhoffer was a squash champion, very risk-tolerant hedge fund manager, and collector of unusual art and unusual friends. A few people have gotten in trouble for how they phrased this, but: if you're memorializing Niederhoffer in any non-generic way, you simply can't avoid mentioning what a supremely weird dude he was. Junto was an eclectic meetup, and he never had any trouble being the strangest person in the room—both in the sense that it was usually him, and that he was supremely unbothered by this. The article mentions NYC Junto, a monthly meetup he hosted that I used to attend occasionally. Junto was a weird survivor of the Big Sort: there were young finance and tech professionals, septuagenarian libertarians who could tell you about meeting Ayn Rand, or people who were attending because they liked the speaker and had no clue what they were getting into. The only rule was that any time an audience member had a question, the speaker was required to pause and answer them. So once a month in midtown, you could basically participate in an improvised dramatic production called The Night the Comments Section Came to Life. Somehow, someone who'd had six kids by two wives and another on the side, who'd blown up multiple times at his own hedge fund and others, managed to keep this unstable-by-default arrangement going for decades. Nobody else could have pulled it off.
• Varun Godbole, who spent a decade working on deep learning, has a personal story about agonizing over the decision of whether or not to take an offer from an AI lab. This is lightly fictionalized, but keen observers of AI may be able to read between the lines and produce a good guess as to the identity of "Mario," who left OpenAI to start a more safety-focused lab called "Noetic." In one sense, the mental agony of deciding precisely which incredibly lucrative job to take is unsympathetic. On the other hand, the more of the world that gets mediated through AI, the more your life is affected by exactly this kind of thought process. It's nice to get a look at the thought process of the people who build the tools that do a growing share of the world's thinking.
• Substack convenes a panel where Kathleen Schmidt, John Paul Brammer, and Lincoln Michel discuss the story of a Call Me, I'll Hide the Body, a $2m book deal that fell apart over allegations that the author used AI to write it. What's striking about this one is that the author's agent was the one to make the call. There are plenty of places where anyone who feels like it can publish creative writing; you can serialize it on Substack, sell ebooks, self-publish, etc. If you go through gatekeepers, the assumption from a buyer at the other end is that the process is gatekept. But it can be hard for gatekeepers to resist social or financial pressure—it was probably hard for the agent to write the email turning down both $300,000 and also missing the chance to represent a hot new author. But it's the right call: it only gets easier to detect LLM use over time, so we'll all be judged based on the norms and evidence-gathering techniques of the future, not the present.
• Jasmine Sun visited a few towns where there's a lively debate about whether or not to build a datacenter. It's fun to read this as a story of completely out-of-touch rich people—they aren't rich financially, but they are in the sense that they can redirect resources towards their preferences. If a tech billionaire bought a stretch of land specifically because it was the last undeveloped stretch of real estate between two cities and he liked that part of the drive, you might view this as conspicuous consumption, or a wildly inefficient form of philanthropy. But for someone opposed to datacenter development, it's perfectly natural to impose that preference on everyone else. Another activist talks about a town that used to have a GM plant, and, since GM left, has been trying to redevelop it. Unfortunately, GM left behind some heavy metals and chemical contaminants in the spill, so nobody wanted it—until a datacenter came along and tried to buy it, at which point activists once again stepped in. "What if the project gets abandoned?" asks one of them. (The answer is that they'd be back where they started, except without the hazardous waste and with however many millions of dollars in taxes the datacenter paid.) The piece is rounded out with some pleasant talk about democracy, but given the arguments it presents, it's actually pretty critical: if you have a system that makes it easier to block things than to do them, you're achieving a kind of sad economic stasis that most people wouldn't vote for. Then again, there are a lot of places with cheap land and little economic activity, and some of them will probably be happy about the tax revenue.
• In Ars Technica, Alan Bradley remembers the Internet before search got good. In a way, we're returning to this: the way you'd find cool things online in the late 90s, as I faintly remember, went like this: somebody else found something cool, and they linked to it and told you why it was awesome. You'd follow a chain of such links and end up somewhere pretty weird—it was pretty common for personal sites to have a list such links, as a kind of collective effort to survey little bits of the web. And then search got good at cooking up recommended links for topics nobody had bothered to explore themselves, and that turned into the default discovery mechanism when you know roughly what you want but not exactly where it is. But LLMs are shifting things back to a more personalized model, because, if you have memory on, they'll generate answers that are tilted towards and contextualized by the user's existing interests. You'd get this organically if the way you found links was to visit the personal homepages of people you already knew and got along with, and now there's an even-more-similar-to-you source that's interested in whatever you want it to be.
• This week in Capital Gains, we talk about the question that comes up every time a hedge fund loses money on a directional bet: do we really have to call them 'hedge funds' if they don't necessarily hedge? At this point, it's just a linguistic curiosity—except that the biggest hedge funds, who keep taking more share in the industry, are hedged to a degree that would astound Alfred Winslow Jones. (This also features a fun bit of finance trivia: Ben Graham, father of value investing and operator of an early hedge fund, actually put a little money into American Research & Development, which is usually viewed as the first VC fund. It's a small world.)
• A Read.Haus user asks if the rise of multi-manager shops with tight risk mandates means that we'll see more market crashes that are invisible in overall equity indices—in other words, a crash where, say, half of the oil companies are up 10%, half are down 10%, but energy stocks overall are pretty flat. There are small moves like this all the time—the running joke on fintwit is that whenever a bunch of stocks in the same industry are behaving inexplicably, you say "hearing a pod blew up." And some of the time, it's probably true—once, in my investment research days, I saw a particular stock just keep dribbling down all day, ending the day -5% on no news. So I emailed a buy-side client who I knew was very close to the company, and very bullish, to ask what was going on. The email bounced, and there was my answer: he’d gotten fired, and his portfolio was unceremoniously liquidated. But big moves are less plausible: the pod shops all know that this dynamic exists, and they vary in their preference for cutting losses or taking on some directional risk in order to exploit the variance. Meanwhile, companies can react, by stepping up their buyback or quickly issuing more stock. And long-only investors, or even longer-horizon hedge funds, will also tend to take advantage of these dislocations. A crash happens when liquidity is suddenly scarce for everyone. If it's suddenly scarce for just one fund, that's too bad for them but probably not something the rest of us need to worry about.

Books

Seveneves: My seven-year-old has gotten in the habit of asking my elaborate hypothetical questions involving complicated feats of engineering, space, and unusual ways for people to die. Usually, all I can do here is offer a little technical guidance (e.g. "coating the cruise missile in a powerful neurotoxin probably wouldn't kill more of the people you're launching it at, but it might kill the people who were trying to launch it.") But one time he asked me what would happen if the moon exploded, and, fortunately, I had some detailed reference material on exactly that. (As with Surely You're Joking, Mr. Feynman, I recommend reading ahead and censoring some details and some lengthy passages. At a few points you'll zip through several pages with "And then all of the people died in various horrible ways.")

The basic setup is exactly that: some unknown force splits the moon into seven rocks, these rocks start colliding with one another, that process accelerates, and it quickly becomes apparent that in about two years, the earth will be incinerated by a cloud of meteors, which will stick around for thousands of years.

It's very much a book written for nerds, both because it has lots of detailed descriptions of clever ways one might try to stay alive in space and because the good guys tend to be fighter pilots-turned-astronauts, roboticists, pop scientists with sterling academic credentials, etc., while the bad guys tend to be politicians with Dickensian names. If a character ever gets themselves out of a tight spot by writing a Python script, you know what kind of reader this book is meant to appeal to.

In some ways, this book is a time capsule of fifteen years ago: there's a character who is a loving parody of Elon Musk: he calls himself as an "Asp-hole," is impossible to manage, and—so it doesn't pick up every beat—he doesn't tell everybody what he's planning but does get things done. And it was written when a global mass-casualty event was a hypothetical, rather than a reality, so it imagines that most of us would react with some combination of calmly going about our lives or spending all of our waking hours helping to avert disaster at great personal cost. There's some artistic license there, for sure.

The story starts deliberately anchored to modern technology, basically asking how humanity survives the end of the world with mid-2010s tech, and with various strains of mid-2010s nerd culture. Over time, the book necessarily has to make more and more imaginative assumptions about technology, culture, and the interplay between them (there's a fun little riff about a society where consumer electronics use cheap, lagging-edge chips, because all the good stuff obviously needs to be saved for robots). So it eventually evolves into a sword & sorcery & spaceships, albeit with more discussions of how one might implement magical-seeming tools, vehicles, etc. Which is still fun, but a different genre: unless you're deliberately setting a science fiction book entirely in a short span around the present, it naturally drifts into fantasy. But the readership of these genres has substantial overlap, so putting both genres into the same book works reasonably well. 

The archetype the book celebrates is the non-performative survivalist. The book features miners, nuclear submarine captains, and of course astronauts, all of whom are expected to exhaustively plan for every possible contingency and also to be ready to improvise when there's inevitably some problem they didn't think of. In the real world, these people get periodically praised when they do something visibly heroic—landing a plane on the Hudson River; refusing to launch a retaliatory nuclear strike on America without getting confirmation that the equipment isn't acting up; ignoring instructions to remain in place on 9/11, evacuating thousands of people, and then dying while trying to get more of them out; or, thematically, coming up with a way to jury-rig a modified CO2 scrubber on the ground and then implementing it in space. But most of these people spend most of their lives making sure things don't go wrong rather than dealing with it when they do. But the fact that light switches, faucets, seatbelts, and supply chains all do most of what they're supposed to do most of the time relies on this sort of person, and it's nice to wrap a hymn to them around a pretty fun story.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• Robots are getting smarter as we control them with more sophisticated models instead of deterministic rules. But not only do models have spiky capabilities, but they have spiky implementations. Choose some random task, and a billion years of evolution beats a few decades of intelligent design. So, are there any interesting new tasks where robots are getting close to human-level skill? 

Diff Jobs

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

• High-growth startup building dev tools that help highly technical organizations autonomously test and debug complex codebases is looking for someone who can help scale and manage their rapidly expanding facilities footprint: everything from lease negotiations, to proactively anticipating / catalyzing new office acquisition, to making sure those offices are stocked with the best snacks. If you want a seat on a rocketship and enjoy fixing things that break–literally and figurately–this one is for you. (SF, DC, London)
• Well-funded, frontier AI neolab working on video pretraining and computer action models as the path to general intelligence is looking for researchers who are excited about creating machines that learn from experience, not text. Ideally you have zero-to-one pre-training experience and/or are a high-slope generalist who’s frustrated that the big labs aren't doing this. (SF)
• 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) 
• Ex-Anduril, Ex-Abnormal Security, Ex-Bridgewater, fast growing startup bringing agentic cybersecurity to 99% of businesses via MSPs is looking for platform and machine learning engineers. Startup experience preferred; what matters most is that you've grown in scope and handled ambiguity over the last few years. (SF)
• AI Transformation firm with an ambition to build an economic world model to run swathes of the private, unstructured economy is looking for Systems Engineers, Platform Engineers, and business generalists who understand how to solve problems.  

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.