Media Gluts Happen at the Level of Distribution, not Content

In this issue:

• Media Gluts Happen at the Level of Distribution, not Content—It always feels like there's too much content out there, but it's hard to find a case where there was a persistent overshoot. What happens instead is that filtering gets better, ad targeting improves, production costs decline, and we have more of exactly what we want instead of a better version of what just about everyone finds slightly appealing.
• Headline Risk—There's rarely just one weird thing happening at a company.
• The Crypto Treasury Trade—One feature of the crypto treasury strategy was that it gave every tiny company a known catalyst that would make their stock pop. The SEC worries that this led to obvious temptations.
• The Paradox of Decentralization—Governments that try to reduce the scope of their activities sometimes find that they need to be more interventionist than usual in the narrow set of things they still choose to do.
• TikTok—As with many other mergers, the hardest part of the TikTok deal was offloading a chunky bit of un-hedgeable risk.
• Signal and Noise in Credit—Lenders are happy right now, but nothing ruins their mood like a series of out-of-nowhere blowups.

Media Gluts Happen at the Level of Distribution, not Content

It's sometimes refreshing to read what people were saying about an industry decades ago, especially in cases where the narrative is "We have way more to read/watch/listen to than anyone could possibly want, and something's got to give." Depending on when you read that, it might have referenced the rise of hyper-specific Kindle Singles titles for every conceivably romantic configuration, exploding budgets at streaming video companies, a cable package with twenty whole channels, a broadcast channel that didn't sign off for the night, or, for that matter, how hard it was to walk through late 17th century London without being subjected to a blizzard of handbills featuring ads, political commentary, scurrilous rumors, and weird trends.

This provides a helpful baseline for thinking about the limits of demand for AI. There are only 24 hours in a day, and at some point we have to spend some of them grudgingly interacting with the physical world. But no society has reached a point where there's literally too much content—though we often reach a point where the framework through which we think about media leads to misleadingly negative assessments of when we'll reach Peak Content.

One of the problems we have is that content is only part of the bundle, and distribution matters, too. One reason for the disappearance of mid-century culture, at least in terms of the written word, is that a lot of that written word was distributed through magazines. A magazine doesn't work well if it's written by a single author—except for effect, like when The New Yorker [dedicated an entire issue to John Hershey's Hiroshima, or NYT Magazine doing the same for Nathaniel Rich, or Wired giving Neal Stephenson almost all the pages for Mother Earth Mother Board. And that means magazines more or less created a market for stories of 2,000-15,000 words, padding out other more enticing but more embarrassing fare. You can see this in other spaces, too. It's hard to tell if American TV manufacturers ever over-extrapolated demand, or collectively targeted more than 200% market share, or otherwise made the usual mistake durable goods manufacturers make during the deployment cycle, because what actually happened to them is that Japanese companies started selling better and cheaper TVs. It didn't matter if Zenith or Motorola overestimated the number of units they could sell if the real problem was selling them at too high a price. In fact, those two mistakes are really the same thing: overextrapolating demand usually means assuming too much demand exists at a price point the market won’t bear, and by definition, as prices fall, quantity demanded rises.

If you'd looked at the short story market in the 50s, and said that there were entirely too many neurotic young English majors moving to the Village in the hopes of getting something into Esquire, you would have been right for the wrong reason. What actually happened is that paperbacks got cheap, and became the default way to distribute fiction, so instead of paying an annual subscription for a shot at a dozen pages of Salinger, you could pay a smaller sum to get a few hundred pages of definitely-Salinger. The magazine was a bundle of upper-middlebrow fiction and nonfiction, and if you were in the target audience there was almost certainly something you'd read, but also very likely to be something you'd skip. Whereas, if you visited a bookstore, you'd know exactly what you were getting.

As it turns out, this did not redistribute all of America's fiction spending to the mid-century greats. Instead, it shifted a lot of that spending to books about cowboys, criminals, spies, aliens, and possibly—if Frank Frazetta was available to do the cover—elves and orcs. When fiction buyers had more choices, they turned out to have “trashier” taste than the editors of big magazines in New York. But they also had more interesting taste! Gene Wolfe, editor of a manufacturing trade journal, who had contributed to designing the machine that makes Pringles, was probably not going to get a short story into The New Yorker. But he found his tribe in paperbacks, and definitely kept that audience entertained.

The magazine-to-paperback transition for fiction turned out to be a good preview for what online distribution would do to the written word. If you don't need a curated selection of articles, but can look for a specific work you've heard of, get a list of everything in a given genre, or even make the what-to-buy-next question as literal as possible by reviewing a list of which books people who bought the one you're considering also purchased. This ended up following rich veins of demand and opening up brand new mines in the form of genres that couldn't be imagined into existence but could be asked to exist. Paperbacks have limits—if you have a fixed cost for formatting a manuscript, commissioning a cover, etc., and you have to put actual ink on paper and ship it somewhere, you have to have some sense of what the demand for your product is, so to the extent that truly weird stuff got published, it was due to flukes and personal connections.^[1] One reason paperbacks on average don't have a great reputation is that the average paperback is likely to be a less-exciting sequel or derivative take on some already well-explored concept. This makes a lot of sense from a business perspective, and further encourages experimentation because any one hit is likely to be followed by one or perhaps numerous sequels. And once readers are hooked, they'll probably tolerate a dud or two and keep reading—When I was eleven years old, the single highest priority in my life was acquiring every possible scrap of information about Grand Admiral Thrawn, and I wasn't put off by some of the unfortunate misfires in that era of Star Wars merchandising. But it means that the average new one published is likely to be designed to keep a narrow, dwindling audience hooked, rather than as some grand artistic project.

There's a more general trend in media, where lower distribution costs encourage lower production values, which enables more niche products. When there are three channels, maybe one of them can compete to have the leanest cost structure, but realistically all of them benefit from being as broadly entertaining as they can be, which means doing a good job with the obvious concepts (e.g family sitcoms, news, talk-shows), and hiring popular stars (e.g. Lucille Ball, Walter Cronkite, Johnny Carson). Expanding the number of channels would, for pure supply-and-demand reasons, lead to a shortage of stars if their approach to talent persisted, but what happened instead was that the minimum production cost that made sense for a given show declined: it just doesn't have to be good if it's something audiences specifically want. But also, because there's a more continuous gradient of popularity, it's easier to match a show to its exact target audience: Walking Dead and Game of Thrones could have been sanitized into network-TV respectability, but the AMC and HBO audiences were already looking for something a little grittier.

And this, too, is a common dynamic in media cycles: differentiating a media outlet by staking out an audience on one side of some multi-modal, ideally bimodal, distribution. Two-paper towns often had a conservative morning paper and a progressive evening paper. Or, put another way: they had one paper that someone would read after taking a shower and before heading to the office, and a different paper you'd read after you took a shower when you got home from the plant. That split ultimately started to mean less as jobs shifted to the office and the relationship between social class and partisan affiliation shifted. But there were other options. Media platforms seek to subdivide attention as narrowly as possible. A newspaper or magazine has different sections in part as an affordance for readers, and in part as a way to target ads—you want your mutual fund ad in the business section and your beer ad in the sports section (unless there's a recession). The economics of cable TV—more total households, regional or national reach—meant that one of the best ways to do this breakdown was through political beliefs.

Where does that leave AI? If lowering the marginal cost for something makes it okay in equality and exceptionally well-targeted, LLMs are definitely another datapoint for the chart (though that chart now uses a logarithmic scale). If fragmented media segment their audiences, we should expect there to be more Claudes and Groks, i.e. chatbots fine-tuned to be on opposite ends of the political spectrum from one another.^[2] It also implies that we'll find a way to keep consuming more, at least as measured by some variable that we in some sense pay for. We're getting far more pixels per second than we got when YouTube or Netflix first launched, and while the relationship between that and revenue isn't linear, it is directional. There will come a point, soon, when labs stop talking about the number of tokens they've produced over a given period because there's too much variance in return-on-token for that to be the right metric. If older media economics prevailed, it would also imply that LLMs would be exceptional at playing to the lowest common denominator. But that's less of a risk when every media consumer is consuming something slightly different. And it ignores the fact that AI companies can price-discriminate, by selling better performance at a premium. There might be more money in raising the economic output of the top 1% than that of the rest of the economy, and that's arguably mission-aligned for the big AI labs because acquiring more wealth implies producing unique tokens when prompted about future business events. It will be interesting to watch which labs move in the direction of a populist product that overfits on previous media cycles, or who view their task as building a product to 10x the 10x engineers (and doctors, and lawyers, and so on). Either model might win in the near term economically, but one of them has a distinct advantage in tilting AI products towards having a detailed, sophisticated model of reality.

Diff Jobs

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

• A startup is automating the highest tier of scientific evidence and building the HuggingFace for humans + machines to read/write scientific research to. They’re hiring engineers and academics to help index the world’s scientific corpus, design interfaces at the right level of abstraction for users to verify results, and launch new initiatives to grow into academia and the pharma industry. A background in systematic reviews or medicine/biology is a plus, along with a strong interest in LLMs, EU4, Factorio, and the humanities.
• A transformative company that’s bringing AI-powered, personalized education to a billion+ students is looking for elite, AI-native generalists to build and scale the operational systems that will enable 100 schools next year and a 1000 schools the year after that. If you want to design and deploy AI-first operational systems that eliminate manual effort, compress complexity, and drive scalable execution, please reach out. Experience in product, operational, or commercially-oriented roles in the software industry preferred. (Remote)
• A leading AI transformation & PE investment firm (think private equity meets Palantir) that’s been focused on investing in and transforming businesses with AI long before ChatGPT (100+ successful portfolio company AI transformations since 2019) is hiring Associates, VPs, and Principals to lead AI transformations at portfolio companies starting from investment underwriting through AI deployment. If you’re a generalist with deal/client-facing experience in top-tier consulting, product management, PE, IB, etc. and a technical degree (e.g., CS/EE/Engineering/Math) or comparable experience this is for you. (Remote)
• YC-backed, Ex-Jane Street Founder building the travel-agent for frequent-flyers that actually works is looking for a senior engineer to join as CTO. If you have shipped real, working applications and are passionate about using LLMs to solve for the nuanced, idiosyncratic travel preferences that current search tools can't handle, please reach out. (SF)
• Ex-Bridgewater, Worldcoin founders using LLMs to generate investment signals, systematize fundamental analysis, and power the superintelligence for investing are looking for machine learning and full-stack software engineers (Typescript/React + Python) who want to build highly-scalable infrastructure that enables previously impossible machine learning results. Experience with large scale data pipelines, applied machine learning, etc. preferred. If you’re a sharp generalist with strong technical skills, please reach out.
• Fast-growing, General Catalyst backed startup building the platform and primitives that power business transformation, starting with an AI-native ERP, is looking for expert generalists to identify critical directives, parachute into the part of the business that needs help and drive results with scalable processes. If you have exceptional judgement across contexts, a taste for high leverage problems and people, and the agency to drive solutions to completion, this is for you. (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.

Elsewhere

Headline Risk

A while ago, The Diff wrote about the bizarre situation at Northern Data, a company that was being incredibly promotional about the future but oddly sloppy at recording their profits in the present ($, Diff). They've since been raided by German tax authorities over something that was not at all a feature of that original Diff article. But this is one of the reasons that sketchy companies can be so volatile: they'll sometimes be fractally crooked, such that they're overstating the growth potential of a stream of profits that is itself overstated. For a short seller, this means that the ideal position size is small, so it's hard for them to offset low-information bulls. Companies like this take a lot of patience to bet against, but usually a company that gets caught in one problem will turn out to have related ones, too.

The Crypto Treasury Trade

One question about crypto treasury strategies is why so few companies did it at first, but an equally good question is to ask why so many did it later on. Did everyone think that every crypto treasury company would trade at a perpetual premium? Did they have some specific plan to quickly turn a stock price pop into a durable increase in the company's value? One missing clue is that the SEC is investigating stock price moves that happened ahead of crypto treasury announcements ($, WSJ). To the extent that cryptocurrency is currency, it's definitionally immaterial if a company buys some, in the same way that a company with a Japanese supplier shouldn't need to make a real-time disclosure that it's taken a position in Yen. So there's a narrative where it's innocuous to tip people off. On the other hand, insider trading ahead of meaningless information that leads only uninformed traders to bid up an asset is still insider trading ahead of information that's material to a prospective buyer or seller. If nothing else, the presence of insider trading as a theoretical legal gray area explains why so many companies used this strategy. They were moving some of their assets over to a very old currency, of being owed a favor.

The Paradox of Decentralization

Argentina has been trying to reduce the government's role in the economy, but, to soften the blow, has been supporting their own currency. This is not an egregious decision, and it arguably serves both political and economic interests for Argentina's top 10% earners to be able to afford expensive imported goods. But that makes Argentina's libertarian turn partly dependent on foreign exchange rates, which forced them to seek a US bailout of unclear proportions ($, Economist). In general, the countries that make their currency artificially cheap will tend to grow fast, while the ones that make their currency artificially expensive will be able to consume more foreign-produced luxury goods. So there's some obvious predictive value in which direction a country sets that rate. Fortunately for Milei, implementing many of his ideas would eliminate exactly the kinds of jobs that enable people to passively consume imported luxuries. So the whole situation is one more example of him betting on himself.

TikTok

Last week's TikTok deal was confusing, particularly the headline $14bn valuation, less than half of the ranges people had been kicking around beforehand. As it turns out, one reason for this is that ByteDance will still retain more than half of the profits, through a combination of a licensing fee for its algorithm and a remaining equity stake. Considering the different dimensions of the negotiation, this makes some sense: the US was aiming less for economics and more for control, and the smaller the economic sied of the deal, the easier it is to convince bidders. These buyers are getting a piece of a big, successful consumer app, one that's done well enough that larger incumbents have had no choice but to copy it. But they're also buying into unknown political risk from two different countries whose leaders like to intervene unpredictably in tech companies. Even if they like the business they're buying, there's no reason to add more Xi Risk or Trump Risk to a portfolio that will inevitably have some already.

Signal and Noise

There's been a little excitement in credit markets recently. On one hand, the best-rated corporate borrowers are occasionally paying less interest than the US government ($, WSJ), presumably because index funds can't turn down assets that fit their mandate just because those assets are overpriced. But this morning a levered auto parts roll-up collapsed into bankruptcy, partly due to off-balance sheet financing ($, WSJ), a few weeks after Tricolor Holdings, a used car chain and subprime auto lender, also went bankrupt. Autos will tend to have more exposure, and more unpredictable exposure, to tariffs, but that kind of issue should affect the entire credit market. There's nothing worse for credit than a situation where almost every company is fine and an unpredictable subset of them have had their earning power permanently impaired. If creditors can't figure out which is which, they have to reduce their exposure to everything. Right now, the anecdotal headlines are telling a worse story than credit spreads, and in the short term excessively trusting lenders are an easy solution to credit stress—they'll let existing borrowers roll over loans, and credit expansion creates demand whose impact is disproportionately positive for the most distressed borrowers. But that's only a good thing if there's attendant growth in productive capacity, and in the late stages of a growth cycle, it can be hard to find those.