Longreads + Open Thread

Longreads

• John Psmith of The Psmiths has an alternatively laudatory and brutal review of Vaclav Smil's Energy and Civilization. It's an open-face praise sandwich! I've read a few Smil books, and he is as far as I know unique in writing history books that take the form of Fermi calculations that turn out to be accurate. (The Diff briefly wrote about another Smil book here.) Smil, like many other thinkers, is better at applying good abstract ideas to situations that aren't immediately politically salient, but tends to struggle when those abstractions lead to uncomfortable conclusions in the present. Specifically: the story of human progress is a story of cheaper and more abundant energy, not of conservation. In fact, the fields where we see the most progress are usually the most flagrantly wasteful ones—the text you're reading on your screen right now could be displayed in a perfectly readable way with 1970s technology; the profligacy of bandwidth consumption, graphics (why zoom the text when you can squint!?), etc. is a demonstration of how much progress computer hardware has made.
• Michael Nielsen asks how AI is impacting science. This is one of the more exciting pieces I've read recently, because the question it focuses on is: can increasingly complex and inscrutable models point us towards undiscovered general principles? Can a 93 million-parameter model be boiled down into a handful of general laws and some edge cases that might be explained by measurement error? One promising sign he points to is that this appears to have happened in at least one domain: chess. Perhaps there are others.
• Ian Bogost writes in The Atlantic about how professors are dealing with ChatGPT. What does grading look like in the age of AI? It probably looks a bit like the interview prep process for the kinds of jobs college students want to get: your interviewer might be impressed that you took a month off to learn algorithms or to refine a stock pitch, but they're mostly focused on what you can deliver and what you can explain on the fly. So a class that assigns homework as a study aid, but treats it as an immaterial proportion of the grade while using timed in-person exams to evaluate learning will probably be the equilibrium. Another option: some academics have written tests that ChatGPT will fail, at least for now. If it's trained on a corpus of public discussions, there are some topics where the median quality of those discussions is bad enough that it will reliably get the wrong answer. This is, of course, unfair to students who are trying to be merely acceptable rather than brilliant. But to employers, in some of the domains where college grads get hired, "acceptable" is now an API call away. Many school assignments end up being proof-of-work rather than a demonstration of mastery, and as this particular kind of work gets easier to accomplish, that will have to change.
• Nadia Asparouhova has a piece in Tablet on the intra-tech split on politics. In one sense, this is inevitable regardless of which side is right, and regardless of whether "right" means "correct about the best policies for the country" or "correct about which set of policies narrowly benefit the industry." When an industry starts to get partisan, there's internal pressure to comply, but there's also an immense external benefit—in the US system, at least, party coalitions tend to approach 50/50, and which side wins an election is a toss-up. So people who support the less popular view within their industry have more influence on how policy works in that industry. (The career advice here may be something like: if you're a huge Sanders fan in high school, flip a coin between majoring in aerospace engineering or petroleum engineering, and if your career goes well then for roughly half of it you'll have a good shot at being your industry's spokesperson in DC.)
• Tyler Cowen interviews economist Simon Johnson. This piece has some great digressions on institutions and the impact of technology. One of the important points it makes is that new technologies make the world richer, but not necessarily right away: factory workers during early industrialization had a very low standard of living. It's worthwhile to think of the appropriate discount rate to apply to things like that: more or less everyone is happy to be descended from people who didn't get the immediate benefits of technology if it means we get to enjoy them, but there's a point at which that trade simply isn't worthwhile.
• In this week's Capital Gains, we look at when bond prices behave like stocks, with the help of some options theory. The legal nature of a security doesn't necessarily correspond to how it behaves over time; you might think you own a stock when what you really own is mostly a short position in oil (this has hit airline investors from time to time) or an esoteric bet on interest rates (as with the regional banks).

Books

• Trust: The Social Virtues and The Creation of Prosperity: perhaps the best "non-economist does economics" book of all time, this work is Francis Fukuyama's investigation of which cultural and legal norms lead to high-trust societies, and how this predicts the magnitude and nature of economic development. One of the most important things the book does is to dig into cases where it's easy to get things superficially wrong: the US has a highly individualistic culture on paper, but also has a long history of civic organizations, clubs, movements, etc.; we're more communal than we look! He also notes that German and Japanese labor markets look very different on paper, but argues that they're similar in practice: asking for a contractual lifetime employment guarantee would be a social blunder in a context where it's a given, so the structure of some German employment arrangements on paper matches what Japanese companies did in practice. The book came out in 1996 and is very much a product of its time (at one point, when discussing the norms-violation of spam, he pauses to explain to readers what the "Internet" is). And this applies to some of the specific arrangements he talks about, like the aforementioned lifetime employment. Fans of dunking on Francis Fukuyama's predictions, i.e. people who have read half of the title of his most famous book, will be delighted that, two separate times in the book, he uses semiconductor fabrication in Taiwan as an example of the kind of industry that is, in his model, implausible. But a book like this is worth reading both because it gets many big ideas right and because some of the mistakes are instructive: reading a book with cogent but mistaken predictions is a reminder that the present is even more improbable than it looks.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• On the topic of energy abundance: what's going on with fusion? How much of this is fluff, how much is real, and how can a non-physicist tell the difference?

Reader Feedback

Great responses to monday's post on why equities build wealth, mostly on Twitter. The most common critique is that even if equities and bonds have similar risk-adjusted returns, we can look at other moments of the distribution and see differences: equities are more leptokurtic, and levering up bonds won't change that. Which is true, but still leaves mysteries: real estate is also a big source of wealth, and while it's hard to gather good data on this, it doesn't seem like a high-kurtosis asset class. It could be that extreme success in real estate owes more to leverage than to skill (or perhaps to skill at structuring leverage well and holding off creditors until the cycle turns). But kurtosis can't be the entire answer, because you can increase it by selling your equities and replacing them with call options on the same assets. Perhaps the transaction costs eat the profits there (the big options-related fortunes come from market-making, i.e. having a bias towards selling options, not from buying short-dated YOLO options). It's been a fun discussion so far.

Diff Jobs

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

• A fintech startup that lets investors trade any theme as if there were an ETF for it is looking for a senior backend engineer. (NYC)
• A successful crypto prop-trading firm is looking for new quantitative developers with experience building high-performance, scalable systems in C++. (Remote)
• A proprietary trading firm is seeking systematic-oriented traders with ML experience—ideally someone who has displayed excellence in DS and ML, like a Kaggle Master. (Montreal)
• A startup building a new financial market within a multi-trillion dollar asset class is looking for generalists with banking and legal experience. (US, Remote)
• A VC backed company reimagining retirement wealth and building a 401k alternative is looking for full-stack engineers with prior experience in fintech. (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.