Longreads + Open Thread

This issue of The Diff is brought to you by our sponsor, MNX.

Longreads

• Cecilia D'Anastasio, Olivia Solon and Leon Yin write in Bloomberg about Stake, a crypto gambling company that has basically invented the optimal casino. They pay countless influencers to find viral content and put the Stake logo on it, they don't warn users about gambling addiction, and Bloomberg finds some strong circumstantial evidence that when streamers play the company's games, the games are rigged so they win. On that last point, the piece is a new frontier in data journalism: they recorded 1,500 hours of footage of people playing, and then used Claude to convert that into a list of bets and outcomes. This is one of the upsides of liquid content: sometimes, information is compelling in one form (a livestream of someone having incredible luck gambling) and then reads completely differently in some other medium (a spreadsheet showing that attributing their wins to luck is simply not credible). This piece is also a good example of companies going full-stack, for better or for worse; Stake's influencers and ads kept getting kicked off other platforms, so they created a streaming platform of their own, Kick, which hosts people who've been kicked off of other platforms for other reasons, and which seems to monetize mostly by funneling new gamblers to Stake. So they've basically recreated the concept of a red light district from first principles.
• Chris Walker argues that some maximalist cases of AI's impact miss the difficulty of good context engineering, which will probably keep AIs from replacing every job, everywhere, all at once. Instead, we'll see them get integrated into existing jobs, but the people doing those jobs will have a better local world model than AI. This is basically the white-collar equivalent of AI being unable to cook you dinner or do the dishes after—some last-mile problems are physical, but some involve knowledge that just isn't in the model, but is well at hand for the person doing the work. (If Pangram is to be trusted, this piece is an example of its own thesis: mostly written by humans, but with some chunks that got flagged as AI-generated. Writing out enough of your own thoughts to give the AI context to work out their implications is defensible; perhaps in the future we'll have text editors that give us live feedback on how surprising our writing is, compared to what an LLM would have done, so we can calibrate the exact level of humanity we want in every document.)
• Sam Buntz in default.blog on how Gen Z's media consumption is unmoored from time and context. This kind of flattening is inevitable. Even naming an era ("Victorian," "Middle Ages," etc.) means conflating what kinds of art people liked at the beginning with what they liked by the end. But it's more extreme today, because you can get algorithmically served a bit of the best of everything, without any connection between that and the next thing. Meanwhile, new media production is also going to find its way to you if it's legible based on what you like about older media. Interestingly enough, though AI takes away some serendipity and coherence from culture when it's applied to a news feed or a playlist suggestion, you can use LLMs to get yourself situated. They're a great way to ask what a given piece of art was influenced by, and which things that sound like clichés sound that way because you're experiencing an original work that everybody copied.
• Casey Handmer has fun questions about the Fermi Paradox. He points out that, if we're ever going to encounter aliens, it's because they can travel close to the speed of light, but that means that even if we're looking closely, we'd see them launch in our direction only a bit before they show up: "if they’re traveling at 99% c, we will see them only when they’re 99% of the way here. If they’ve traveled 1000 light years to visit us, we’ll see them (at best) 10 years before they arrive." This doesn't fully solve the paradox, because it doesn't explain why they didn't get here sooner, but it does leave the ultimate question appropriately ominous.
• Eric Hoel in The Intrinsic Perspective asks: if we can access intelligence-on-demand, why does the world feel dumber rather than smarter? Which might be a bit like asking: if Spotify has access to every kind of music in existence, why is anyone listening to the top 40 when they could be listening to Bach instead? The reality of intelligence on demand is that the kind we want is not necessarily that impressive, and that plenty of people had access to plenty of it already. LLMs raise the floor in many ways—I've used them to clarify math concepts that would otherwise remain mysterious, but that just means that the people who understand those problems are now, on average, a little worse at math because I've just joined that cohort. But this piece is a little too pessimistic, because we're still very early to learning how to work with models. I haven't been able to get one to produce prose that I'd want to publish under my own name, but they get closer every time. But LLMs have improved The Diff by speeding up research, clarifying technical questions, and warning about obscure references. Presumably they'll take over a larger share of tasks over time, and the nature of the output will evolve accordingly.
• On Read.Haus this week, someone asks why college endowments don't buy entire operating companies. (The question is almost certainly referencing The Counting House, which I haven't read—I've gotten a literal slap on the wrist for admitting this, so I'd like to reassure readers that it is on the to-read list.) The biggest universities are pretty big financial institutions, but the problem they have to solve for is that their donations are cyclical but their financial needs are at least slightly countercyclical. So they want a controlled proportion of their portfolio to be tied directly to growth. And they also want to avoid a situation where something they own might be an unpredictable cash drain: they already have capital calls from being limited partners in venture and private equity funds, but in that they know the total size of the capital calls even if they don't know the timing. (And they have some control over this timing, too; apparently in the financial crisis some of these kinds of investors gently suggested to funds that, if they wanted commitments to their next fund, they'd better not be too aggressive with capital calls when asset prices were depressed.) PE's dry powder gets a little more damp when every asset is cheap and it would be nice if there were another bidder.
• In Capital Gains this week, a look at the many ways to measure inequality, and the ways to mismeasure it. There's a very dishonest habit that some people have in discussing inequality, where when they talk about its negative effect, they mean consumption inequality—rich people buying Nth vacation homes while poor people struggle with evictions—but when they quantify it, they use wealth inequality rather than consumption inequality, because the numbers are bigger. But if that kind of inequality is supposed to be a problem, rather than an output of policies that have upsides and downsides, it's important to talk about it honestly.

A Word From Our Sponsors

Trillions of dollars are flowing into AI infrastructure and DRW’s Don Wilson recently predicted that “the total dollars spent on compute will exceed total dollars per year spent on crude oil in the next 10 years.” Yet unlike crude oil, there’s no futures market for training runs, no benchmark markets to bet on the speed of AI progress, and no public market for OpenAI shares despite a valuation nipping at the heels of Saudi Aramco. MNX is changing that. As compute eats the world, it needs the same financial infrastructure as every other commodities market. MNX is the first exchange built to trade the full AI value stack—from H100 rental rates to frontier model benchmarks to AI equity.

Try their testnet platform here.

Books

Streetwise: Getting to and Through Goldman Sachs: Lloyd Blankfein worked his way up from gold salesman to CEO of Goldman Sachs just in time for "CEO of Goldman Sachs" to be a hated-by-default figure—his tenure started in 2006, and the company did fine during the financial crisis, so he achieved his peak public attention doing things like testifying before Congress or giving interviews to hostile journalists. And he gave them some great material! In one interview, when he was asked to justify the bank's existence, he rattled off the usual answer—allocating capital to growing businesses that need it makes us all better off—and then brushed the reporter off by saying that Goldman was "doing God's work." In the book, he calls this a "Lloydian slip."

A number of people took this seriously, or at least pretended very hard that they thought he was serious. But if you read the book, you repeatedly get the sense that he's just a generally funny guy who makes jokes when he should know better. At the time that he was born, in the mid-50s, that personality trait would probably preclude someone from having any shot at a successful career working for an investment bank. This book is an autobiography, but it also contains a nice social history of the collapse of one model of American class stratification, by someone whose birth gave him the dual advantage of a chip on his shoulder from some of his early experiences and many opportunities as norms shifted later on.

Blankfein grew up in a public housing project at almost the exact time that "the projects" were getting their current connotation. His parents, a postal clerk and a receptionist, couldn't afford to flee to the suburbs. But Blankfein managed to get ito Harvard, and then Harvard Law. (In a gratifying validation of a very minor Diff observation, Blankfein attended Harvard Law right after the release of The Paper Chase, and says that it seemed to influence not just students but professors, just like wiretaps revealed that mafiosi started talking like Godfather characters after that movie came out. The Diff has argued the same.)

When he graduated, it was still a point in corporate history when Cravaths, Swaines, and Moores did not like to mix much with Proskauers and Roses; he wasn't able to get into one of the big white-shoe firms, but spent a few years doing tax law at Donovan, Leisure. Deciding he didn't particularly enjoy tax law, he jumped to gold sales at J. Aron (even though this is a book about working as a professional trader, the only time he specifically portrays anyone yelling, in all-caps italics, it's his mom when she finds out that he's going to use his Harvard education to be a gold salesman). J. Aron was able to print money without taking meaningful risk when gold trading volume was high, but around the same time Blankfein joined, Goldman acquired them at the peak of that cycle. At that point, Blankfein ended up in another system, stratified by class instead of religious background: the J. Aron traders were scrappy, the Goldman bankers were polished, and each side treated the other like The Other; he mentions a few times that "we" meant "J. Aron," and Goldman was "they." At one point there's a literal Upstairs, Downstairs moment: during the crash of 1987, Blankfein and his fellow currencies and metals traders take the elevator up to the equities trading floor to watch the chaos. They get escorted back down by security.

He joined the trading business right between the era of trader-as-international-spy and the era of trader-as-spreadsheet-jockey, and got to do a bit of both: trips to South Africa to negotiate gold purchases, but also the beginnings of Goldman's SecDB risk-management system. Blankfein built up Goldman's currency-trading business, and gradually expanded both his personal management mandate and the extent to which Goldman was a trading rather than banking business. At one point, when he's expanding the trading business internationally, he engineers the same chip-on-the-shoulder setup he'd experienced: they staffed their Tokyo office with ethnic Koreans.

By the 2000s, Blankfein has been promoted to COO, which he argues is an underrated job ("The CEO is the public face of the company—the one who gets subpoenaed") and then, when his boss gets tapped to be treasury secretary, he makes it to CEO. Just in time to get Goldman's books flat ahead of the financial crisis!

One of the peculiarities of the social responsibility of financiers is: if all of them avoided taking undue risk, we'd have fewer financial crises (and fewer booms, of course). But when a crisis happens, the sloppy CEOs get fired, and so the cautious ones are the only ones who were running banks in the run-up to the collapse and who are still running them after. If you flatten your model down to "banks caused the crisis," without asking which banks contributed what, you'll tend to be maddest at whoever's least at fault. So the arc of the book is the classic one, where someone who's good at a technical or business-management skill crushes every problem in that domain, at which point the only problems left are in unfamiliar fields like politics and PR.

Capital allocation has a brutal form of the Peter Principle. Not only do you get promoted up to your level of incompetence, but you manage the most capital right at the point where whatever strategy you're best at has peaked and gotten overextended. It's not uncommon for people to have good career results in terms of average annualized returns, but to have negative lifetime returns in dollars because their fixed income arbitrage strategy's AUM peaked in 2007 or because they were able to raise so much money to bet on growth equities in 2021. Blankfein managed to avoid that: he was the right manager for taking Goldman's currency trading business from nonexistent to dominant, but also the kind of person paranoid enough to buy credit default swaps from AIG on AAA-rated mortgage-backed securities, and then to buy credit default swaps on AIG, too. That's a rare combination of skills, and the US financial system is more stable for it.

Open Thread

• Drop in any links or comments of interest to Diff readers.

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 to help highly technical organizations wrangle and autonomously test/debug complex codebases is looking for a senior design engineer to own their design system and build the visual abstractions customers rely on to simulate their software systems, find bugs, and quickly remediate them. A compelling portfolio, a rare blend of design and engineering chops, and a deep understanding of how the internet and browsers work required. (D.C.)
• A pre-IPO, next-generation chemicals company that’s manufacturing the mission-critical inputs for a sustainable American reindustrialization is looking for a CFO to own the capital raising roadmap and allocation strategy end to end. Experience turning corporate strategy into a data-driven narrative and advising on late stage capital raises and/or IPOs preferred. (Remote, Houston)
• 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)
• 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)
• 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. (SF, NYC)

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.