|
In this issue: - Tokenomics, Revisited—Crypto prices are in an upswing, but this boom is less extreme than the last one. It's a good time to revisit what parts of crypto actually serve a useful purpose—and why they're tricky to use.
- AI and Education—Schools try to detect LLM-generated essays, and the issue is messier than mere plagiarism or false positives.
- Money Laundering—Iran still sells oil, and now we know more about how.
- Agglomeration Economics—Density begets growth. Saudi Arabia tests the limit.
- Threat Models—Competitive for scarce resources at TikTok.
- Universal Hardware—Information-processing devices tend to get more use cases over time.
Today's issue of The Diff is brought to you by our sponsor, 2 Hour Learning. Tokenomics, Revisited
In the beginning, cryptocurrency was a concept without a use-case: it was an elegant way to decentralize a ledger that tracked how much money everyone had, so you'd know who had a fraction of a bitcoin and who had tens of thousands of them. But it was an academic matter as long as no amount of Bitcoin could buy anything.
Satoshi knew this, and speculated himself on what the first killer app would be. He guessed adult entertainment, but it was a different vice entirely that provided the initial Bitcoin killer app. But the killer app for crypto, more broadly, came later. And, in a rare case, there are two crypto killer apps, one a bit better than its bad reputation and another that's superficially appealing but has deep underlying problems:
- Crypto is a great tool for gambling, particularly gambling on crypto. You can trade it on completely decentralized exchanges, with decentralized providers of liquidity in both the lending and market-making senses. The upside to this is that it's a financial petri dish for constructing novel contracts, and gets to magic away some of the cruft that's accumulated in the traditional financial system.
- It's a financing tool for new projects that makes it easy to distribute equity broadly (through airdrops) or to systematically give it away to specific people who will use and/or hype the project.
There's a nice symmetry to this, where crypto gets adopted at two ends of the financial market—implementing hyper-liquid markets with complex derivatives and long chains of intermediaries on one end, and financing projects too weird even for early-stage investors on the other. That liquidity-provision doesn't have a lot of direct utility, unless you happen to be both a degenerate gambler and someone keenly sensitive to slippage, sub-human-perception-threshold latency, etc. But it still means that the overall crypto ecosystem can shuffle value around rapidly.
Using tokens as a marketing tool is also a much worse model than it appears to be. The problem is that it often works, and the better it works the noisier a signal it is about the viability of whatever it finances. If you use a thinly-traded token to recruit new developers, or get customers to use your product, there's a wide confidence interval for how much it really costs to acquire new customers. Which, for a while, doesn't matter at all: they're spending their own tokens, not US dollars or tokens that they'd have to buy with fiat currency. But unless those tokens are meant primarily to be a form of currency, they're getting valued on some dividend-equivalent basis, because the issuer has a buy-and-burn mechanic, or both. And both of those imply some way to create value independently of subsidized signups. A business can fund its growth by constantly issuing new equity, but only if investors remain convinced that the equity will produce more cash than it consumes.
This is not a crypto-specific problem. For example, it describes Snap, which tends to produce nice-looking free cash flow numbers because it's expanded its share count by 10% annually since founding, thanks to generous employee stock grants and generous support from public markets.(This creates a sort of invisible leverage: at a constant level of stock-based compensation relative to other costs, Snap's price-to-expenses ratio also sets the pace of their annual dilution, on a lag that's determined by Snap's equity comp vesting period and their employee turnover. And that lag probably changes if they need to replace people who left because the stock wasn't high enough to make equity compensation worth it.)
But Snap is a telling example in another way: they have an actual product, it's genuinely useful, and in the last decade they've contributed a lot to what the world thinks the right set of social network primitives ought to be, even if that mostly means contributing to other companies' market caps. The model of paying for equity does serve a function, and extending it to earlier companies that can pay directly in equity-like tokens rather than in cash they get from selling equity to VCs grows the space of ideas that it's financially feasible to explore—and means that it pays particularly well to think about new businesses that are possible with different arrangements of ownership and fundamental economics. That's a good thing to subsidize: the corporation is an amazing financial technology, but it's probably not the only end state of company structure, or the one that works best in every possible circumstances.
If there's a way for tokenomics to work well, it's this: tokens crystalize hype and convert it into dollars. That provides enough potential energy to convert the hype into something real. But they need to actually do that—token-based projects should shift, as quickly as possible, from issuing tokens directly to selling them and paying people mostly in a currency they don't issue. Market discipline is a powerful tool, so it's a good idea to raise money from the most discerning market you can in order to keep yourself honest about the value of what you're building.
Can this really be true?Elsewhere
AI and Education
Bloomberg has a good roundup of the state of AI detection in grading. It focuses on the false positives, i.e. students who were incorrectly accused of using AI, in part because it's hard to get a good handle on how AI-generated papers get passing grades. (Especially in environments with a late-Soviet approach: students pretending to write papers that teachers pretend to read. That will be especially hard to detect because the sides that find this mutually satisfactory will find each other, and neither side will really complain.) There are a few dimensions to this:
- To the extent that AI-generated content detectors are finding traits of documents that uniquely identify them as AI-created, it's probably good for students to know if the creative output they're capable of can be generated by a machine, in the same way that it's good for students to know which kinds of arithmetic they should outsource to a calculator.
- But what if it's picking up something else? The article notes that English as a Second Language students are being flagged more often. And other sources indicate that lots of model fine-tuning happens in countries with low wages—where many of those workers may have grown up speaking a different language at home. In other words, these models might be worse at detecting people who write like AI, and better at detecting people who write like the real people who help AI figure out exactly how to write.
- The piece also mentions some countermeasures, like livestreaming work (a bit extreme) or just composing work in Google Docs so there's an edit history. There are already workarounds for this, like browser extensions that take a full-length essay and dribble it out into a document at a human-looking cadence. But, as with other kinds of dishonesty, the goal is not to make it impossible, just to make it inconvenient enough and risky enough that abiding by the rules is the simple and lazy solution.
As with other applications of AI, LLMs in academia are a labor-saving measure that introduces new monitoring costs. There are completely reasonable ways to use them in text composition—they're great for tip-of-the-tongue searches, and a good way to get unstuck if you're lost enough that you don't know what to Google to figure out what you're doing wrong. They are also a tempting substitute to sitting in front of a blank page and trying to whip out an essay. It's either up to schools to figure out better ways to grade or up to students to decide how they'll demonstrate proficiency if doing the work is optional.
Money Laundering
The Economist has a great investigation of Iran's sprawling oil-sales and money-laundering system ($). Money laundering follows a gradient, where it starts with people who know that the funds are illicit, moves through entities that willfully refuse to care, and eventually makes its way into the wider world where one dollar is equivalent to any other, even if one of those dollars was the proceeds of illicit activity.
It's a roundabout and expensive process: "The price usually tracks Brent, the global oil benchmark, minus a discount of $10-30 a barrel." Of course, that doesn't mean that the administrative overhead involved in laundering money for a tanker full of oil is in the tens of millions, just that there are economic rents being earned. There's enough demand for oil, and enough places with a lack of interest in enforcing US sanctions, to create an end market for the oil. But the population of able and willing co-conspirators is small enough that they can extract significant concessions.
Agglomeration Economics
Denser areas have higher GDP per capita, and the causation runs both ways: if you can make a lot more money moving to New York, the city will have more people, and if the city is teeming with people who want to make money, there will be more opportunities to do so. This raises the question: can you achieve something like New York City's level of economic activity merely by cramming people together? In particular, could you do this by packing people into a 400-meter cube whose volume is 20x the size of the Empire State Building, in Riyadh? Saudi Arabia is actually a sort of natural experiment in the limits of natural experiments. The country has a small population and effectively unlimited money and space, so they can afford wild economic experiments. Most of these will probably fail, but they'll fail in informative ways—seeing what happens when you create an extremely dense commercial district within a single building is informative about how much good density alone can do.
Threat Models
TikTok has fired an intern for sabotaging other interns' projects. There's more background here, in Chinese. The gist is that GPUs are a scarce resource, but companies don't want to allocate them for projects that aren't working. This creates an incentive: mess up something in someone else's config file, and your own project moves ahead in the queue. Internally, companies tend to be surprisingly high-trust environments, where things get locked down only after there's a problem. And the fact that there have been grumblings about hardware scarcity for years, but stories like this are rare, is evidence that that's the right approach.
Universal Hardware
It's common to observe that smartphones are systematically cleaning out pockets and purses: money, ID, keys—if you're carrying around some physical representation of information, it's going to get replaced by a device that can access and process arbitrary information instead. The phone can't replace everything, but that's where other devices come in. A few people have tested Apple's AirPods-based hearing aid. One way to miss changes in an industry is to focus on how the product is packaged rather than what value the customer gets. If there's already a device for pre-processing sound in order to make it easier to hear what you want, then it doesn't matter if it was first marketed for music and hands-free phone calls; it's a hearing aid, too.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - An AI startup building tools to help automate compliance for companies in highly regulated industries is looking for a director of information security and compliance with 5+ years of info sec related experience at a software company. Experience with HIPAA, FedRAMP a plus. (NYC)
- A well funded startup founded by two SpaceX engineers that’s building the software stack for hardware companies is looking for a staff product manager with 5+ years experience building and managing data-intensive products. (LA, Hybrid)
- 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. (NYC)
- A well-funded startup that’s building the universal electronic cash system by taking stablecoins from edge cases to the mainstream is looking for a senior full-stack engineer. Experience with information dense front-ends is a strong plus. (NYC, London, Singapore)
- A growing pod at a multi-manager platform is looking for new quantitative researchers, no prior finance experience necessary, 250k+ (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.
|
