What Do Proof-of-Work and Proof-of-Stake Really Prove?

Plus! Diff Jobs; Social Network Nuances; The Cost of Dropping Ads; Credit Outlook; Investment Niches; Wages

In this issue:

What Do Proof-of-Work and Proof-of-Stake Really Prove?

In crypto, the two primary models for governing a network are proof-of-stake and proof-of-work. The former is pretty straightforward: major changes get decided by voting, and votes are weighted by token ownership. So it's analogous to the system of corporate governance in a company with a single class of stock.[1] The latter model, proof-of-work, was popularized by Bitcoin, and uses a mechanism where miners compete to provide computing power, which a) makes it infeasibly expensive to attack the cryptocurrency, b) provides an otherwise zero-sum competition that people can use to credibly signal their affinity, and c) yes, consumes as much electricity as many countries.[2]

While the models are mechanically very different, they end up being similar, but subtly different, when we look at the incentives at work. Proof-of-stake makes sense in a model where the goal of decisions is to maximize the value of the network, and the people with the biggest stake in the network ought to have the most influence over how it develops. Proof-of-work makes a more subtle argument: blockchains are trusted because they're statistically very reliable, but the structure of the Bitcoin protocol means that you're always dealing with arbitrarily high levels of confidence, never perfect certainty. The proof-of-work model implicitly argues that raising this confidence level is always good, and that participants who don't mine crypto should pay the ones who do for keeping things secure.

Proof-of-stake and proof-of-work apply in general terms to many social phenomena, too. If you're trying to influence an outcome, whether it's making a political endorsement for a national election or suggesting a new spot for dinner, you're implying that you have some kind of stake in being correct. A proof-of-stake argument would be some kind of claim based on personal background, or just general high status.[3]

The canonical proof-of-stake interaction is citing previous experience. You see this in lots of places: investors turning down pitches in particular industries that have a long history of disappointing outcomes (a new location-based service! A dating site without the pathologies of nonstop swiping! Using machine learning to do litigation finance! Building a better [enterprise software company whose product is notoriously bad and whose sales team is legendary] by focusing on product rather than sales!). Sometimes an answer like "I've seen lots of these pitches, and they all sound good, but they always lose money" is enough (though it's more helpful to explain why—surely the winners in these categories won't be winners forever!).

On the other hand, proof-of-work in a social context is exactly what it sounds like. There's a phenomenon in stock pitches where the pitchee tries to ask detailed questions, not because they're curious about the answer, but because they're testing whether or not the person making the pitch did their homework. These can be very fun, especially if there's a good answer to the question that requires doing even more homework. There are some memos where the upshot is in the executive summary, and the rest of the content exists to be scrolled-through admiringly but never read. These are still useful—you'll put your thoughts down on paper, implicitly prep for a few of the more conniving did-you-do-your-homework questions, and have a handy piece of reference material for later. But to the extent that a memo's purpose is to catalyze some action, it doesn't matter whether the 5,000th word gets read or ignored; its existence is what really counts.

In a crypto product, the choice of proof-of-work or proof-of-stake is pretty binary; it's hard to build a stable system that uses both, though it's theoretically possible. In social interactions, there's a constant metagame about which of these will be the guiding standard. In the VC-early-rejection hypothetical above, it's going to be rare for the founder to have as much direct experience as the investor in evaluating ideas that sound good but don't work; for a typical VC—whose returns are power law-dominated and who spends much of their time on investments either ahead of the decision to invest or after the company starts falling apart and needs their help—they may have literally spent a majority of their professional career thinking about and working on that exact category. On the other hand, they're probably less up-to-date with what's changed recently.

In a way, some of the standard questions for an investment pitch, like "Why you?" and "Why now?" are actually requests to pit proof-of-work against proof-of-stake. There is some amount of local knowledge that can overcome proof-of-stake denials—the difference between accumulated wisdom and accidental bias is only clear in retrospect, and the person it belongs to is often the last to figure it out.

Proof-of-stake sounds lazy, but it has high returns in a few ways. First, it's a stock rather than a flow; the trouble with deciding every question the proof-of-work way is that it is, well, a lot of work. Whereas dipping into the well of wisdom is close to free.

The real synthesis is that proof-of-work leads to proof-of-stake: that experience comes from somewhere, and it generally took effort. So when there are holes in the reputation-driven model, there's not just room for learning—it's practically compulsory.

You can also view proof-of-stake as reputational capital, and proof-of-work as reputational labor: in both cases, the driving force is a prosocial fear of embarrassment. And that turns out to be a pretty important one—you can model a lot of human behavior, at many levels, by thinking of it as a continuous effort to be as visible as possible without ever looking bad.

  1. Like the company example, most of the network's functions, i.e. executing individual transactions, are not put up to a vote; it's the big choices that determine what the serially-repeated behaviors are that get voted on. ↩︎

  2. Though sometimes the country comparisons get annoying, since the comparison is either to small-but-recognizable countries, or framed as "more electricity than N whole countries," but that is more of a reminder that the world has a fair number of countries that are some combination of tiny and poor. It ends up being a non-sequitur: "Bitcoin mining consumes 160 terawatt-hours of electricity annually. Also, Bhutan and Kiribati exist." ↩︎

  3. A matter of longstanding fascination for me is that there are limits on cash donations to political campaigns, at least nominally, but there's no limit to in-kind donations in the form of endorsements. Oprah's endorsement of Obama apparently accounted for over a million votes in the primary, and in the general campaign Obama spent just under $11/vote, so that was at least an eight-figure in-kind donation. Measuring the exact value of this is hard, though, because it depends on the behavior of incremental voters. Presumably the cost of a marginal vote is far higher than that of an average vote. Ironically, the fact that very rich people can spend far more than the nominal donation limit on getting their preferred candidate elected makes the system more democratic, by allowing plutocrats to compete with celebrities instead of leaving celebrities—yes, including James Comey—as the only people with the right to make unlimited campaign donations. ↩︎

Diff Jobs

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

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.


Social Network Nuances

It's been interesting to watch Meta's Threads continue to develop. Meta has two big advantages here: Twitter has already demonstrated the core interactions of a status-update social product, and Meta has a roughly two billion daily active user head start. But that also means that, in comparison to other social networks, Meta's opportunity cost for new threads features is much, much higher. So one of their recent rollouts was the ability to turn off automatic sharing to Facebook and Instagram. This is a striking demonstration of Twitter's status as a product rather than a feature: it really is a different kind of social network, with different norms and standards. The fact that even Meta wants to offer some separation between its various services is good evidence that social media is not a winner-take-all business, even if individual clusters of users and interaction mechanics do end up dominated by just one platform.

Disclosure: Long META.

The Cost of Dropping Ads

Amazon and Apple cut a deal where Apple paid to keep ads for competing products off its product pages ($, Insider). These inline ads are incredibly profitable but tricky to implement ($, The Diff). They provide a revenue lift, and the brutal economics of auction-based ad systems also mean that advertisers are making targeted, relevant offers. But they also clutter the shopping experience. Individual users might be mildly annoyed, but if those ads are subsidizing a better selection, fast shipping, a generous rewards program for the store credit card, etc., they may be worth the hassle. Merchants, though, especially if they're large merchants with distinctive products, can cut a deal. This is another case where large, monopolistic companies care about what would otherwise be an inefficiently-mitigated externality (granted, the negative externality of ads on product pages is, in this case, the product of another monopolistic company). Sometimes good outcomes for users are the result of a balance of powers between ruthless, profit-maximizing competitors.

Disclosure: Long AMZN.

Credit Outlook

A ratings agency is a unique media company, but it remains a media company. As The Diff has noted in many contexts—and as you are, statistically, experiencing right now—a fun feature of low-marginal-cost media is that any given piece of content can function as either a product offered to paying subscribers, or as a giveaway meant to entice those subscribers.

This applies to ratings agencies, too. One weak theory of the agencies is that they actually provide alpha; if you sell when something is downgraded and buy when it's upgraded, you win. But they're really about discovering and justifying consensus, and showing their work so investors can articulate their own investment case rather than having to laboriously reconstruct the consensus view to figure out if the market disagrees with them. So when you see a headline like Moody's changing the US's credit outlook to negative, remember that the main calculation they're running is: what is the earliest that they can plausibly do this, and what's the risk that S&P will scoop them? In one sense, they're twelve years behind, but there's always room to take a newsworthy story and recycle it with a modern spin.

Investment Niches

Investors are raising funds to take control of startups that aren't growing fast enough to be venture-fundable, but that might make sense as a small standalone business ($, FT). This basic model has existed for a long time, and tends to be on a syncopated cycle with the rest of the venture funding system: an upturn in funding catalyzes the growth of new companies, some of which will reach escape velocity and make sense as publicly-traded, others of which will top out at a smaller scale. Historically, more of the companies in the latter category would be M&A targets for big tech, but as big companies run into antitrust obstacles (and have management teams distracted by the need to incorporate AI into everything) the niche gets more attractive.


This post is a good look at the complexities of answering a simple question: are American workers, in the aggregate, better-off than they were pre-pandemic? And how do those outcomes compare to the immediate pre-pandemic trend? The answer is that workers are better-off for the most part, though inflation has taken a bite. But one interesting wrinkle is that compensation growth has been more skewed to wages than benefits recently, reversing a longer-term trend where real wage growth was limited and compensation mostly rose in the form of higher healthcare benefits. One result of this is that there's a stronger wealth effect: if you get a 5% raise, but it consists of flat wages and a 20% increase in the cost of your employer-provided health insurance, you don't feel much richer. Whereas if that same raise means an increase in your weekly paycheck, it's more likely to get spent.