Longreads + Open Thread

Longreads

• Dwarkesh Patel has a great interview with Marc Andreessen (video, transcript). This one is full of great riffs: the idea that VC exists to restore pockets of bourgeois capitalism in a mostly managerial capitalist system, what makes the difference between good startup founders and good mature company executives, how valuation works at the earliest stages, and more. Dwarkesh tends to ask the questions other interviewers don't.
• This John Carmack interview is also worth reading. Carmack has worked on gaming, rockets, VR, and now AI. One highlight worth mentioning is the fractally ML-related reason why he's choosing to work independently: "[B]ecause we don’t know where we’re going yet, there is actually a strategy inside machine learning where you need a degree of randomness—where you start with random weights and random locations and sometimes multiple ensemble models. So, I am positioning myself as one of these random test points, where the rest of the industry is going in a direction that’s leading to fabulous places, and they’re doing a great job on that. But, because we do not have that line of sight—we’re not sure that we’re in the local attractor basin where we can just gradient descent down to the solution for this—it’s important to have some people testing other parts of the solution space as well."
• Charlie at Sunk Thoughts has a meditation on the sunk cost fallacy and the difficulty of measuring counterfactuals. This piece goes in many interesting directions, but the core theme might be: accounting is hard enough in business, where you can operate with a singular, quantifiable goal. Measuring your results is necessarily harder everywhere else, but it's also a necessity to try.
• Richard Bookstaber on risk management and culture clashes at Salomon Brothers before and after it was acquired. Risk management is partly the science of dealing with agency problems: if a trader takes a risk that pays off, it results in a bonus; if the same trader takes a risk that doesn't pan out, they can blow up their firm and go find another job. Measuring risk is the first step to fixing this. But solving incentive problems within a big company, especially as it integrates with a bigger one, is a struggle. For example: "Traders found that they could profit by arbitraging the difference in the lower amount Citibank charged for working capital and the higher amount Smith Barney paid for working capital. They extracted working capital from Citi and supplied it to Salomon Smith Barney, earning the difference." If a large, sophisticated firm can find itself paying traders for doing a carry trade with the firm itself, you know it's a hard problem.
• Nintil has a good piece on ML and biology. The essay makes some very interesting points on how drug discovery differs from other cases where we've seen impressive results: we don't have comprehensive understandings of some of the systems we're looking at, and we don't have good ways to collect all the data we'd want to continuously refine our models. There's an extended comparison with self-driving cars, and it's an important one: even a very bad self-driving car can collect lots of sensor data, and improve itself, but the tests we run on humans create many orders of magnitude more data than we can capture. (At least for now.)
• And in The Diff's educational newsletter, Capital Gains, this weeks' explainer is a primer on multi-manager/pod/hedge funds, which dives into how the oldest hedge fund model is bigger and more profitable than ever. You can sign up for the newsletter for free here (now with a referral program, where rewards include access to a discord). Upcoming pieces include: how to think about book value; the first ten minutes, hours, and years of getting up to speed on a stock; and the reason companies love just-in-time inventory.

Books

• The Little Book of Valuation: If you want a quick and effective summary of roughly what analysts are doing when they put a price target on something, this book is a great primer. Valuation is a series of easy-to-understand rules that are also very hard to implement, but that's true for many other domains.
• Money Men: A Hot Startup, A Billion Dollar Fraud, A Fight for the Truth: It's a cliché to say that a business book "reads like a spy novel," but this book, a finance-gonzo narrative about both Wirecard's collapse and the FT's efforts to expose it, does actually have a lot more tradecraft than the typical white collar crime narrative. Wirecard's fraud was a fun reversal of the usual norm in criminal behavior. If you do something, you have to hide the evidence. But in their case, they were faking revenues and profits and their big task was to hide the absence of evidence—so a key part of the fraud was sham acquisitions that explained why their fictional profits didn't create actual cash.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• What does the world look like a year from now if there's a "soft landing," and the US doesn't get a serious recession but does see a gentle return to the post-crisis growth trajectory?

Reader Feedback

Responding to last week's longreads, Andrew Antes asks:

The a16z piece has me wondering if AI will be sort one of those areas where most of the upside is widely distributed and not captured by any central entity - I think you catalogued a bunch of these awhile back. Like the airline industry creating all sorts of unrealized value.

Maybe that upside capture is represented by the cloud infrastructure companies’ massive revenue sink.

Value creation is always a good first priority for new industries, but value capture ultimately matters a lot, too—especially because industries that can't manage it are run by capital markets and professional managers rather than founders.

A Word From Our Sponsors

Built by investors, for investors, Tegus is a one-stop investment intelligence and research platform that modernizes research processes.

Our ecosystem combines at-cost, on-demand expert calls with a 55K+ transcript library; quantitative financial workflows that streamline research across company disclosures, management presentations, earnings calls and filings; and 4K+ fully-drivable financial models and company benchmarking data, including every KPI and comparison that matters. Better research starts with Tegus.

For a limited time, The Diff readers can trial the Tegus platform for free at http://www.tegus.com/thediff/.

Diff Jobs

Companies in the Diff network are actively seeking talent! If you're interested in exploring growth opportunities at unique companies, please reach out. Some top current roles:

• A hedge fund is looking for an experienced alternative data analyst who can help incorporate novel datasets into systematic strategies (NYC).
• A fintech startup is looking for investment researchers who can build new strategies on top of an AI platform (NYC).
• A company using ML to build better tools for writing higher-quality code is looking for software engineers who care deeply about performance and like tackling seemingly impossible problems. (Washington DC area)
• A high-growth provider of market data to both retail investors and institutions is looking for a staff-level frontend developer. Experience scaling a product/platform is ideal. (US, 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.

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.