The whole book at three sizes: a sentence, a paragraph, a page. Everything here is argued properly in the chapters — this is the package, not the proof. (Editor-drafted distillation, provisional.)
One sentence
Nearly everything you know arrived secondhand through a machine that compresses ideas at every step and decides which ones travel at all — and because that machine is tuned to reward attention rather than truth, Lossy is about how to build people and institutions that can carry hard truths through it intact.
One paragraph
Every idea that reaches you has made a journey — from the world, through measurement, expert consensus, news, and finally something shareable — and at every step of that journey two things happened to it. It was compressed: the claim survived, but the instructions for reading it correctly (the methods, the conditions, the “we’re not sure yet”) were stripped away. And it was selected: a gate chose it over a thousand alternatives, using criteria — publishable, newsworthy, clickable — that have nothing to do with whether it’s true. Compression is nobody’s fault; you have a finite budget for understanding hard things, and no fix changes that. But the gates are owned, and today the dominant ones are tuned to maximize engagement, which means what reaches you is systematically the most arousing version of anything, not the most accurate one. Run that machine long enough and corrections lose to myths, raw emotion becomes the floor of what spreads, and the question “is this true?” quietly becomes “are these my people?” The book’s answer is not to retreat into smaller, like-minded groups — that makes it worse. It is to build bridges: deep experts trained to translate across worldviews, and institutions funded outside the attention market, designed to survive distrust rather than win the feed. Not victory — survival, long enough to hand the full-resolution world to the next generation.
One page
Start with the trip an idea takes. A real finding almost never reaches you as itself. It reaches you as a headline, a chart screenshot, a thing someone repeated at dinner — the last stop on a pipeline that runs from reality through measurement, analysis, expert consensus, journalism, and finally the meme. Every stage re-encodes the idea for the next audience, and the book’s first move is to notice that even the most respectable stage is already doing it: “Raw data is misnamed. The ‘raw’ is the first compression, pretending to be the source.”
Two different things happen at every stage, and you need both to understand the mess. The first is transport — the telephone-game repackaging everyone already suspects. What matters is what transport drops: usually not the claim itself, which travels fine, but the decoding instructions — the scope, the method, the uncertainty. The claim arrives; the manual for reading it doesn’t. The second is selection — at every stage a gate decides which ideas move at all, and each gate has its own test: fundable, publishable, newsworthy, shareable. Most explanations of our information crisis only see one of the two. The book’s structural claim is that transport loss is like friction — a cost, nobody’s decision — while selection is criteria, and criteria can be set: “You cannot aim a cost. You can aim a choice.” That asymmetry is the whole game. Whoever sets the gates’ criteria steers what a society can know.
Why does compression happen at all? Because you are finite. Add up the hours a person can actually spend absorbing difficult material and it’s a tablespoon of weeks out of a lifetime. Compression isn’t a moral failure; it’s the only possible response to that budget. Every artifact — a textbook, a sermon, a statute, a chart library — picks a point on the same curve: more faithful and harder to use, or more usable and less faithful. You can’t have both in one object. That’s the honest part of the machine.
The dishonest part is who tunes it. Today’s dominant gates are tuned to engagement, and “engagement is a proxy for revenue, dressed up as a proxy for user benefit.” Under engagement-tuned gates the damage acquires a direction: “the pipeline doesn’t break at random. It fails in a direction” — corrections are always less dramatic than the myths they chase, so the correction loses every race. Strip a claim’s decoding key and it can land three ways: roughly right; inverted — a true finding decoded into a false belief, with nobody lying anywhere; or as a pure tribal flag, where “the operative question has quietly changed from is this true to are these my people.” Run a mass network on these gates long enough and what reliably spreads converges on raw strong emotion — shared, mostly, without being believed in any way that changes behavior: “The belief is the alibi, not the engine.”
The institutions that were supposed to compensate are losing — but not how you’d think. Civilizations have always run a two-part repair: preservation keeps the full, uncompressed form alive somewhere (archives, journals, scripture, source code), and training keeps producing people who can read it. The modern collapse is asymmetric: the archives mostly held; the training hollowed out. And nothing censored those institutions — they were simply out-competed, because engagement-bait costs nothing per reader and depth costs real money: “You can fight a captor; you cannot fight an equilibrium.” The asymmetry is what makes it urgent: “you can replace a corrupted book, but you cannot replace a corrupted generation.”
Then a new machine arrived that owns every dial at once. A large language model is gate, option-menu, content-generator, training data, and training objective in one place, under one owner — the most concentrated selection surface any medium has ever had. It points both ways. Used faithfully, it’s the first real relief the receiver budget has ever been offered: a patient tutor that can decompress any source down to the preconditions you’re missing. Left to the same economics that tuned social media, it’s the old failure at industrial scale — the summary quietly replacing what it summarizes: “The paper is technically still there; nobody reads it. The summary is technically not the paper; everybody reads it.” Which way it goes is not a technical question. It’s the same ownership question the rest of the book asks.
So what do you do? Not the intuitive thing. Retreating into smaller, like-minded networks curdles into echo chambers that discredit the outside world. The book lands on integration — moving complex truth between groups that don’t share starting assumptions — and commits to what that takes: deep specialists deliberately trained to translate across worldviews (not generalists); institutions that do preserve-and-train work funded outside the attention market, so they can’t be out-competed by free outrage; everything designed for survivable polarization — auditable by skeptics, functional under partial distrust — rather than for a trust that isn’t coming back; and AI used inside those institutions, under their own custody, never rented from the machine it’s meant to counterbalance. The goal is explicitly not to beat the engagement machine. It’s the monastery’s goal: “They never beat the dark age. They just outlasted it.” That work runs on the timescale of universities and cathedrals, not product launches — and the book’s closing claim is that it is exactly as worth doing at that timescale as it sounds.
Ready for the full argument? Start at Chapter 1 — The Information Landscape, keep the glossary open in a tab, or wander the landing page’s other ways in.