- Elad Gil has a good look at AI economics. One interesting possibility he explores is that there will simultaneously be a hardware-driven secular decline in the cost to train a specific model, and a parameter count-driven increase in the absolute cost needed to train the most recent model. In that world, the best LLMs will be proprietary (unless you're up for launching a Kickstarter campaign with a billion-dollar funding target), but older models will be open-source. What's really interesting to think about here is that smaller, older models are more mentally tractable: when they make mistakes, it's a little easier to see why they were misled. So that's a world where AI profits will be concentrated but practical AI knowledge will be diffuse.
- Rowan Zellers on taking a job with OpenAI rather than staying in academia: there has been a very rapid shift from AI results being published by universities to being published by private companies ($, Economist), which is an important inflection to keep an eye on—turning a money-versus-status tradeoff into a money-and-status versus less-of-either trade can divert a lot of talent.
- And on the topic of career decisions, Praveen Seshadri writes about selling a company to Google and then quitting a few years later. One of the striking details is the claim that managers care more about keeping employees happy than about keeping customers happy. Businesses naturally produce some level of dissatisfaction over time, through unavoidable inconveniences, nonzero transaction costs, etc. And companies implicitly have to decide whether these problems are absorbed by the company or by its customers. Neither of these choices is especially fun, but over time a bias towards not making something the customer's problem is probably a bias that leads to better automation and more thoughtful processes.
- Bloomberg has a good piece on ION, an acquisitive fintech company that handles back-office work for traders, which briefly ground to a halt after a cyberattack. There's a lot of invisible infrastructure in the world, and one thing that makes it visible is that sometimes it breaks.
- A good piece on the history of the Rust programming language. One handwavy approach to describing languages is that there's a continuum from treating a computer as a physical machine, with the program as a list of instructions for that machine, and treating the entire enterprise as a way to make abstractions that get translated into computation through a process the programmer doesn't have to think about. The tradeoff is between higher maximum execution speed from telling the computer exactly what to do versus lower implementation speed from, well, painstakingly telling the computer exactly what to do. And low-level code has wetware memory limitations; it requires a more robust model of what the computer is actually doing. Rust pushes the efficient frontier out a bit by making certain classes of bugs impossible to write, while remaining low-level. Languages are partly a technical phenomenon and partly a social one, and anecdotally the Rust community has pretty high morale; one of the risks of adopting a new language is needing to find engineers who know it and like it, and an evangelical community helps to solve for that.
- This week's Capital Gains looks at why just-in-time inventory management is so appealing—there's a big financial impact, but it's also (shades of Rust!) a way to run a business "close to the metal" and to really understand how suppliers and customers are thinking. You can subscribe to Capital Gains here to get a weekly breakdown of a concept in finance, economics, or strategy.
- The Great Leveler: Violence and the History of Inequality from the Stone Age to the Twenty-First Century. You probably heard about Thomas Piketty's Capital in the Twenty-First Century, the book arguing that as long as the return on capital exceeded the rate of economic growth, the rich would end up owning everything without external shocks. But you might have decided that this book did not sound nearly metal enough, in which case you should check out The Great Leveler, which takes a similar approach but focuses almost entirely on the wars, revolutions, state collapses, and pandemics that reset economic inequality. This book raises a fun question: what if the average rate of return on wealth is indeed higher than the rate of economic growth, but this is offset by higher variance? Great instances of leveling are typically bad (total war, plague, communism, etc.), but in those scenarios the rich have more material wealth to lose. The book is heavily statistical; as it turns out, you can get pretty good indications of historical income distributions by measuring house sizes, for example. Bits of it are horrifying, like the various descriptions of communist takeovers with their landlord-execution quotas and the like. And some are poignant, like the detail that after the collapse of Ancient Egyptian civilization, people still decorated coffins with hieroglyphs—but they'd forgotten how to write, so the hieroglyphs they used just spelled out nonsense. And, of course, it's one those books that was suddenly retrofuturist, since it was written in 2017 and has a chapter speculating on what the long-term impact of a pandemic would be.
- Gates: How Microsoft's Mogul Reinvented an Industry--and Made Himself the Richest Man in America: an excellent biography covering Bill Gates' career up to the early 90s. One of the retrospective surprises in this one is just how much Microsoft's business model had to evolve even before the cloud and AI era: they spent their early days in a very tenuous position where competition and piracy were always threats, and diversified out of necessity. Throughout the book, there's always some other company that's an existential threat, and until the last few pages Microsoft is generally not the biggest software company in its category, but there's a lot of turnover at the top. One way to read the Microsoft story is that it's about a company that sought to reduce variance rather than maximize gains at all times—it was one long series of Kelly Bets in a field where competitors irrationally pushed all their chips to the middle of the table.
- Drop in any links or comments of interest to Diff readers.
- An old joke in macro, from Simon Kuznets, is that there are four kinds of economies: developed, developing, Japan, and Argentina. Are there any industries where this dynamic happens, i.e. where there's one exceptionally great company and one company that mysteriously can't figure things out?
Taylor Pearson responds to last week's Longreads, particularly to the link on volatility:
Would quibble with the Verdad piece that you need an asterisk that volatility is a better measure of risk for assets with negative skew than ones with positive skew. You tend to see a decent difference between Sharpe and Sortino in trend strategies for instance where there is positive skew.
This is true! Volatility represents risk within some bounds, but there are definitely cases when an investment outcome is sensitive to some specific real-world binary outcome, and if that outcome is fundamentally unpredictable, volatility will be an imperfect-at-best measure of risk.
And from Nick Mazing, Director of Research at AlphaSense, responding to Thursday's Three Stories of Ads and Scale ($):
Incidentally, I went through the redlined (aka blacklined) META 10-K yesterday, and spotted that the word "growth" is often now replaced with "scale". Exact match search, growth dropped from 81 hits to 37 hits. Scale is up from 15 to 23.
A Word From Our Sponsors
If time is money, why are you wasting it?
Maximize investor research and diligence returns with an end-to-end platform that solves inefficiencies. Tegus streamlines the information investors need to move quickly, build conviction and make better decisions to outperform the market.
Right now, The Diff readers can trial the Tegus platform for free at http://www.tegus.com/thediff.
Companies in the Diff network are actively looking for talent. A sampling of current open roles:
- A VC backed company reimagining retirement wealth and building a 401k alternative is looking for a founding CTO. (NYC)
- A successful crypto prop-trading firm is looking for new quantitative developers with experience building high-performance, scalable systems in C++. (Remote)
- A company building ML-powered tools to accelerate developer productivity is looking for a mathematician. (Washington DC area)
- A company building tools to enable zero-knowledge proofs is looking for multiple roles, including a senior cloud infrastructure engineer. (Remote)
- A profitable AI startup is looking for ML engineers to help build new services to help small companies accelerate their growth. (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. We're always onboarding new companies, so the available roles change frequently.
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.