Archive
131 essays, grouped by category.
Policy & Power
By the time a law addresses a technology's harms, the technology has usually already moved on to its next form.
A chip, a cable, or a cloud region is no longer just engineering — it's a lever in a relationship between governments.
Psychology
When what you did clashes with what you think, one has to give. The action already happened and the belief is private — so people end up sincerely believing whatever justifies what they did.
The same behaviour gets explained by character in others and by circumstance in yourself. The asymmetry is not mainly self-flattery — it is what you can and cannot see.
A fair coin flip for equal stakes feels like a bad bet to almost everyone. The asymmetry behind that instinct quietly governs pricing, negotiation, investing, and why reform is so hard.
Wartime analysts mapped damage on returning bombers and proposed armouring the places full of holes. A statistician pointed out those were the survivable hits — and inverted the entire dataset.
The quiet, bookish stranger sounds like a librarian. There are a hundred times more salespeople — so the shy ones alone outnumber every librarian in the country.
The people most certain they're right are disproportionately likely to be at the exact skill level where certainty is least justified.
The mind doesn't search for the truth by default. It searches for confirmation, and calls whatever it finds the truth.
Science & Math
A rare disease, a 99% accurate test, a positive result — and roughly a 9% chance you're ill. The gap comes from one number most people leave out entirely.
Around thirty per cent of numbers in real datasets begin with 1, and only five per cent with 9. People inventing figures don't reproduce that — which is how the pattern became an auditing tool.
The statistics taught in most classrooms describe a world of averages. A surprising amount of what actually matters lives in the extremes instead.
The hardest problems left in quantum computing are as much about error correction and business cases as they are about qubits.
A treatment beats its rival for small stones, beats it for large stones, and loses when you combine the patients. Nothing was miscalculated — which is what makes it worth understanding.
One die is flat. Ten dice added together make a bell curve. The shape is a property of adding independent things — which tells you exactly when to expect it and when not to.
An unusually good or bad result is disproportionately likely to be followed by a more ordinary one — not because of any correction, but because extremes are rare by definition.
Seven heads in ten flips is unremarkable; seventy thousand in a hundred thousand would be extraordinary. The same probability, different room for chance — with consequences most people miss.
Elliptic curve cryptography secures a huge share of the internet's traffic, and almost none of its users could describe what it actually does.
Finding a biosignature in another planet's atmosphere wouldn't prove life exists. It would just make the silence a lot more interesting.
Society & Politics
The thing that gets you hired and the thing that actually teaches you are separating faster than most institutions have adapted to.
You can recognise thousands of faces and actively maintain far fewer relationships. The famous figure of 150 is contested — the underlying constraint is not.
The moment a job can be done from anywhere, the candidate pool for it stops being local and starts being global.
Most of human urban history was built around a 15-minute radius. The last century of car-centric planning was the actual anomaly.
Ideas that are ordinary today were once outside the range of what a respectable person said aloud. Mostly the facts didn't change — the boundary of the sayable did, and it moves by repetition.
A Nash equilibrium isn't necessarily a good outcome. It's just a stable one — which is a very different thing.
Every herder who adds a cow to the shared pasture captures the whole benefit and pays a twentieth of the cost. Nobody is greedy, everyone is correct, and the pasture dies anyway.
The prisoner's dilemma isn't really about prisoners. It's about why individually rational choices can add up to a collectively terrible outcome.
Across centuries of recorded history, a rising power challenging an established one has resolved peacefully far less often than most people assume.
Two children share one cake and one child's gain is the other's loss. Two neighbours swap mangoes for apples and both end up better off, with no new fruit. Telling these apart changes what you should do.
Systems & Complexity
We label parcels 'fragile' but have no word for the opposite — not strong, which survives shocks unchanged, but something that improves because of them. Your muscles are the everyday example.
You drive through a ten-minute jam and past no accident. The jam was real, it moved backwards down the road, and no car was it. A great deal of the world is built this way.
Almost nothing in a complex system stays still. It's either spiraling toward more of itself, or being pulled back toward balance.
Optimizing anything that isn't the bottleneck doesn't just fail to help — it can quietly make the bottleneck worse.
Cutting inventory frees a lot of cash. The original reason for cutting it was different — stock conceals late suppliers and broken machines, and removing it forces them into view.
Car plants stopped worldwide for want of cheap chips, and consumer demand had barely moved. Each tier adds a little caution to what it received, and the additions compound.
The most consequential events in history weren't hard to predict because they were complex. They were hard to predict because nobody thought to look.
Technology & Platforms
Moore's Law wasn't a discovered law of physics. It was an observed trend that the entire industry then organized itself around achieving.
The interesting story in AI isn't any single model — it's how fast the cost of intelligence is falling.
The same technology gets overestimated early and underestimated late. Both errors come from drawing a straight line through something that is slow first and rapid afterwards.
GPU datacenters look like a technology story. They're really a capital-intensity story with technology as the excuse.