Longreads- Philo of M&DA has a great piece ranging from Sam Zell's real estate strategy to housing politics to Kuhn's model of scientific revolutions to the history of hand-washing in hospitals. Discursive and fun. The core point is that it's incredibly hard to embrace a new mental model, not just because of peer pressure, but because once you have a model that describes some part of reality quite well, it's very hard to consider holes in it. One of the important threads in this piece is that new models tend to sound more simplistic than legacy models, and since the real world is a complicated place, they look low-status—surely making money in real estate involves more than buying assets at below replacement value, and it can't possibly be the case that a complex set of symptoms that strikes intermittently in a seasonal pattern can be halted with one simple habit! But if everyone's model is wrong in the same way, then the biggest positive impact available is pretty monocausal, because everything else has already been optimized for. (The piece also has a good framing for why buying assets at below replacement value works as well as it does. It's not because you can liquidate them—obviously, the price you're paying is the money someone else is getting when they liquidate! Instead, it's that you have a lot of headroom for earnings improvement before any incremental competing supply shows up. So it's statistically protecting downside, but practically providing a path to upside.)
- Zhengdong Wang has a great annual letter, opening with some really thoughtful comments on AI scaling and capabilities. If you're asking a question like "which of these programs does matrix multiplication fastest on comparable hardware," you have fairly objective answers; if you ask "which of these models produces the most compelling ad copy," you have either a subjective question or one that punts to market judgments that may not capture some of the relevant externalities. (Select purely for that, and you'll discover that many of your products have sprouted fictional features and money-back guarantees.) The more powerful AI tools get, the harder it is to generalize about their abilities, as opposed to making domain-specific judgments.
- Cracking an old cold case, Ken Shirriff explains what was happening, at the hardware level, in Intel's notorious floating point division bug in the Pentium . This was a massive story at the time (even referenced in Saturday morning cartoons). The bug only makes sense in light of some clever optimizations for rapid division, which depended on a lookup table. And that lookup table was missing some values. This is slightly different from Intel's original explanation of the bug (or, at least, means that Intel's explanation was accurate but misleading). Which is proof that white lies for PR purposes eventually get discovered, even if it's decades after the fact.
- Will Tavlin in N+1 on how Netflix has moved towards forgettable, low-end-of-middlebrow content. This is a function of the invisible base rates problem: when Netflix launched its streaming service, it was great for people who loved movies; these people raved to their friends who merely liked movies, and there are simply more people with taste close to the median than who want sophisticated fare. So any service that's perfect for you is either on a path to oblivion (if they don't find a sustainable model) or on a path to really annoying its original fans (if it finds a way to scale). At least, that's the case if there are fairly fixed production costs and low marginal costs; if you want your tastes to be catered to indefinitely, choose media that anyone with a keyboard can create rather than the kind that requires camera crews, actors, and the like.
- Dwarkesh Patel has notes on China, a piece that doubles as an inspection of how much you can really learn from travel. ("[Y]ou're not going to learn about the risk of a war or the state of the AI race by gazing at skylines or chatting up taxi drivers. Of course you can learn about those things by talking to the princelings and researchers and CEOs. But if you have access to these higher ups, surely you can also get them on the Zoom call. And fwiw, this should you update you in favor of more Zoom calls, not less travel.") One of his most striking observations is that he had a hard time getting anyone to name public intellectuals. It's never clear how much value they really create (it's an instance of the problem above; if they're a public intellectual, they have wide appeal, but one way to get that appeal is to dumb down a message or become more of an entertainer than a thinker). But it's notable that, at least according to the people he talked to, China's much larger media ecosystem doesn't produce such figures the way the US's does.
BooksSBF: How The FTX Bankruptcy Unwound Crypto's Very Bad Good Guy: There was a very splashy book about SBF by someone who wasn't deeply familiar with crypto and really liked their subject, and this book is a nice inversion as a work by an experienced crypto journalist who is pretty annoyed at how badly SBF ruined the party (at least for a while). The book has plenty of first-person accounts from SBF himself during his ignoring-the-lawyers-and-talking-to-the-media arc, and plenty of skeptical pushback as well. And it's a good look at the general tension in the crypto world: successful companies tend to centralize the ecosystem, even though it's built on decentralization. That's hard to avoid given the incentive structure at work, but it also leads to lots of infighting and the occasional blowup. One thing the book walks through is the question of whether the best way to understand Alameda and FTX, from a running-a-scam perspective, was a) FTX was a captive exchange that Alameda, as a prop trading firm, could readily exploit, or b) Alameda was a built-to-be-dumb liquidity provider that subsidized activity on FTX, making the exchange look more active and allowing Alameda to cover losses by selling FTX equity or borrowing against FTX's own token. And the answer turns out to be a bit of both: some traders praised FTX for offering tighter spreads than other exchanges, but one higher-frequency trader noticed that his algorithms were faster than the competition everywhere but FTX, and inferred that FTX was running its risk-management algorithm ahead of every trade except those by one participant, Alameda. (Since Alameda had an infinite credit line, it didn't make sense to run those checks, but it did mean that the Alameda/FTX relationship would be obvious to whoever was fastest. As a general rule, if you're running a scam that involves markets anyone can access, it's very, very dangerous to run the scam in such a way that the most competent participants know exactly what you're up to. At that point, you're just one more source of alpha.) It also explains something about FTX's weird marketing. Because the exchange was so liquid, they attracted plenty of institutional interest. But that meant that these institutions were also trading against other sophisticated institutions. Cutting checks to Tom Brady and Gisele Bundchen and Larry David was actually a way to bring retail traders onto the platform, so the institutions would have someone they felt comfortable trading against. Alameda could already be that dumb counterparty on its own, but it couldn't overcome the reputation that FTX was full of ruthless professionals. So it's a weird 3D-chess version of a marketing campaign: FTX ran ads focused on retail investors in order to convince institutions that these retail investors existed, explaining why such institutions were able to trade with uninformed counterparties! One note on this book is that it's worth flipping through a copy or downloading a sample before reading it, because the writing style is somewhere between "gonzo journalism without leaving the house" and "entire book dictated via Voice Memo in one long Adderall-fueled writing session." It transcends "breezy" and veers off into "inexplicable"; there are sentences that seem out of order, and an explanation of decentralized crypto exchanges that involves an extended metaphor about wizards trading magical stones. Here's the conclusion to a discussion on crypto and anonymity: "Whether or not bitcoin (and its ilk) is anonymous is kind of like the question of who killed JFK. No one has a decisive answer about it, but anyone with a strong opinion sounds a little crazy once they start talking. So. Forget about all that." Okay! (There are actually several sections that start out explaining some mildly technical issue and then end with this throwing-up-hands textual maneuver. But there are also pretty good explainers on things like the Uniswap/Sushiswap kerfuffle.) Overall, this book was a good addition to the FTX canon, with some unique perspectives from industry participants. There are still lingering questions, but they mostly linger because the most plausible answer leads to incredulity. Did SBF really think he'd be able to pull this off, and somehow make enough money to fill a hole in the FTX/Alameda balance sheet? Did he just not realize that crypto prices could go down? Or, even worse—what if he ran the numbers basically accurately, made a positive-EV bet, and there's a near possible world where even today he's a popular philanthropist who can dismiss any ethical questions as rumors spread by his less successful competitors? Open Thread- Drop in any links or comments of interest to Diff readers.
- It’s a new year, and that’s a classic time to decide between doubling-down on or completely inverting last year’s big ideas. Place your bets!
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - A premier proprietary trading firm is looking for smart generalists to join their investor relations team, working with external investors, rating agencies, and the internal finance team. Investment banking and/or investor relations experience preferred. Quantitative background and technical aptitude a plus. (NYC)
- A company building the new pension of the 21st century and enabling universal basic capital is looking for a general counsel to help lead capital markets and regulatory efforts. Experience in an early-stage context is a plus. (NYC)
- A hyper-growth startup that’s turning customers’ sales and marketing data into revenue is looking for a staff-level product engineer with a track record of building, shipping, and owning customer delivery at high velocity. If you like to build, this role is for you. (NYC)
- Ex-Ramp founder and team are hiring a high energy full-stack engineer to help build the automation layer for the US healthcare payor-provider eco-system. (NYC)
- YC-backed AI company that’s turning body cam footage into complete police reports is looking for a tech lead/CTO who can build scalable backend systems and maintain best practices for the engineering org. (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.
|
