The Temple Near Mieza

Plutarch tells us that when Alexander the Great was still young, his father Philip sent for “the most famous and learned of the philosophers of his time“ to teach and form him. Alexander was full of promise and ability and his father was already justifiably proud of him: “‘My boy, you must find a kingdom big enough for your ambitions. Macedonia is too small for you.’“

So Alexander became a pupil of Aristotle, and Philip gave them “the temple of the Nymphs near Mieza as a place where they could study and converse, and to this day they show you the stone seats and shady walks which Aristotle used.“

It’s 2026 now and my daughter is the same age as Alexander was then and about to enter high school. I have no kingdom for her, but I am a justifiably proud father and I hope for a long and meaningful life for her. We’ve just chosen the high school she’ll attend for the next four years. Some people were surprised by our choice.

We live outside Washington DC and are very fortunate to have plenty of good public and private options. There were two schools at the end of our search, both Catholic and all-girls. Both have exceptional reputations and check all the usual boxes of a college preparatory private school. The differences between them aren’t found on a brochure.

One is more than two hundred years old on an all brick campus steeped in heritage and prestige. While the tradition is the liberal arts, they’ve remained competitive with the expectations of upper echelon college preparatory schools and all freshman receive an iPad with their textbooks preloaded. The other school is only 20 years old but there are no devices. Freshman read the Iliad and the Odyssey on paper. Latin and Italian are required. There is no computer class or department. It feels like the closest thing to a classical school from fifty years ago that you can find.

The differences have to do with the people and the environment and what they believe about forming humans. Twenty years ago, the resume bullets were probably enough. It was reasonable to prepare a kid for a world with a steady career and a mainstream set of values.

That world has fractured and a new world has invaded our daily lives. Phones have made all of the most urgent and radical ideas of our world available for hours every day and AI is turning computers into agents with their own skills.

The most important step to consider in school choice and education is: what are you educating for? Most schools optimize for credentials and hope the rest works out. But is a school’s role to build a resume or to form the kind of person who doesn’t really need one?

We chose the school with classical Latin and no screens. There were many other reasons for our choice: cost, location, and faith. My wife, daughter, and I all had overlapping perspectives for why we ended up at the same place.

I want to explain mine because the terrain of education is so different today.

Capacity Before Tools

Technology is the set of tools we use for leverage. In order to use tools well you must first develop the capacity to use them. The more complicated the tool, the larger the capacity required.

If software has been the star at the center of the technology world, then the vibecoding phenomenon is the last dramatic burst of energy during a supernova. I don’t know what comes next. Nobody does.

But I know I don’t write code any more, the LLMs are too good. And yet there’s more engineering work than ever and I’m more excited than ever. There’s always new ideas in the world and old ones I haven’t learned yet. There’s always more to build, and it’s easier than ever to get started.

Formation is the word I keep coming back to. And it’s the goal missing from how most people think about education. Instead of a series of resume points and a collection of standardized test scores, formation is how education shapes who you are as a human. It’s about what you’re made of — the depth of your reading, the clarity of your thinking, what you value in morally ambiguous situations, your capacity to reason about complex systems, your habits of discipline and attention. In the new world where AI can beat you on a standardized test, this is the layer that counts.

Dario Amodei has a framework for what makes an AI model good, and it maps to childhood in a rather unsettling way. First you need a lot of compute. In humans, that’s a matter of genetics. You also need a large amount of high quality data. That’s years of reading, writing, conversation, experience, and challenge. And you need to train a model for a long time. That’s at least 18-20 years (and preferably a lifetime). And you need an objective function for what the model should do. Humans have plenty of these, but the best one, the one that philosophies keep coming back to, is virtue.

I joked years ago that I was building three baby AGIs but the metaphor isn’t that far off.

Data quality in this regime is underrated. There is a difference between reading Homer and watching TikTok reels, and the difference is not content. The difference is cognitive effort. If you have to make it happen in your head, it’s building something. If you can passively consume it, it’s noise. And that’s exactly what most kids are getting for hours every day during growing years that matter most.

Gorging on information starves formation.

The new and fashionable claim is that anyone can build software, anyone can make anything, and the new tools have democratized creation. But watching people use AI demonstrates how deep the differences are. AI acts as an amplifier: some people can do the work of a whole team from just a couple years ago. Someone without a good foundation uses AI the way you’d use a calculator: it gives them an answer, and they accept the first output because they can’t imagine a better one. In short, a person has to be built up enough to leverage what the tool provides.

Where people are on this spectrum has to do with their formation. Building up a substantive model is exactly what classical education provides. Centuries of trial and error have refined how we best develop humans to practice logic, rhetoric, sustained attention, moral reasoning, and comfort with intellectual challenge. It shapes not just capabilities and skills, but also virtues and values. It’s what college prep promises but often fails at.

Andrew Cantarutti calls this focus The Emancipatory Wager. He says that we need to “build capacity before offering tools that can substitute for it.” Kids shouldn’t avoid AI because it’s bad, they should avoid it because the more they develop their own capacities, the more expressive and amplified their capabilities will be with AI.

This is what I was thinking about when we chose a school.

A Portrait of Her Days

Next year, my daughter’s school day will start with Homer and it will end with her own thoughts (since she has no phone) and in between there will be no screen to interrupt the passage from one to the other. The class will go through the entire Iliad together, reading on paper, and occasionally out loud. She will conjugate Latin verbs and stumble through elementary Italian. There’s no online homework. No iPad is distributed at orientation with textbooks preloaded next to a messaging app. She’ll head onto the field with her sports team, chatting and laughing with her friends. Most of them won’t have phones either. This is what we chose.

At another school, a freshman opens her iPad and her English textbook appears alongside every other app the device can run. She is fourteen and she has the discipline of a fourteen year old, which is to say the discipline of someone whose prefrontal cortex won’t finish developing for another decade. Somewhere during the second act of Julius Caesar a text arrives (because of course she hooked up her personal account). Then another. She context switches. The text is from a friend and is about nothing and is urgent in the way everything is urgent when you’re fourteen. She flips back to the play but she’s lost Brutus’s soliloquy. The words are there but the thread has snapped — the magic of feeling the accumulating weight of a man reasoning himself toward something unthinkable. That thread is what classical education seeks and it snaps when attention fragments. When the whole world is trying to steal your attention, threads rarely connect.

Why Shakespeare? In Shakespeare you train on pattern recognition across human motivation and moral reasoning under ambiguity. You learn comfort with complex language and new vocabulary. The reason Shakespeare feels hard is because it increases your capacity just like lifting heavy weights increases your strength.

My daughter will read the same soliloquy in ink instead of pixels. Her only option when the passage gets difficult is to stay with it or stare out the window. Both of those are fine. The first trains sustained attention; the other is boredom, where every original thought she’ll ever have begins its life. There is no third option to offer relief and take the effort away.

Daily constraints are a decision against the noise of the world. No device means no context switching. A fourteen-year-old working on a Latin sentence is working hard. We like to fake productivity and tell ourselves we’re learning by swiping Duolingo on our phone but I’ve never met anyone who actually learned a language that way. They’re too busy swiping over to TikTok in between each checkpoint.

The curriculum at her school isn’t radically different from any strong college prep school. What’s different is everything around it. The screens are gone, the families all chose this on purpose, and the expectation is that difficulty is normal.

Here’s the thing, I have no idea what my daughter will do in ten or twenty years. Neither do you or anybody else, the world is changing too quickly. I don’t want to hide her away in a monastic school and pretend the 21st century isn’t happening.. it’s the opposite! I know that she can build more capacity now by reading ancient texts on paper and developing math skills with a pencil than she ever could tapping her homework answers on a screen while juggling notifications. I wanted this school because of what’s coming.

I can’t wait to hear what my daughter thinks of Brutus’s decisions. I hope she’ll come home thinking about it, wondering about the virtues and tensions in his character. She might still be thinking about her vocabulary words during field hockey. And on a walk home, she might be worrying about how to finish the last page of her essay.

Or maybe she’ll just be daydreaming, conjuring castles out of clouds and dreaming of places she’ll one day travel and dragons she’ll one day slay.

Reading and Writing

Two skills matter more than everything else. Charlie Munger said he never met a wise person who didn’t read all the time. Not one, in a lifetime of meeting wise people. This has always been true and everyone nods along and nobody takes it seriously enough.

Reading is the oldest technology for building a mind and it’s still the best one. When you read seriously, following a complex narrative across hundreds of pages, you are doing something that no other medium replicates. You learn how words are tools for both thought and decoration. A person who has read widely and deeply for twenty years has a richer, more densely connected interior world than someone who hasn’t, and that interior world is exactly what AI amplifies. You can become a centaur capable of things neither the human nor the machine could do alone. A shallow reader with AI gets autocomplete.

Writing is thinking made visible. It’s how humans compress experience into transmissible form, how we discover what we actually think as distinct from what we vaguely feel. Anyone who has tried to write a clear explanation of something they thought they understood has discovered the gap between the feeling of understanding and the real thing. Writing closes that gap or reveals that it can’t be closed, and both of those are valuable. Machines can generate fluent prose on any topic in seconds now, so what do you want them to generate? The ability to think clearly is the whole game, and proof of that ability is what writing is. If you can’t articulate your ideas, they’re meaningless.

AI is making literacy more economically and personally valuable than it has been in a generation. And at the same time, our culture has handed every child a device that systematically destroys the capacity for both. Short-form video is the lingua franca of the internet now. Most kids get predominantly passive, algorithmically optimized content that requires no cognitive effort whatsoever, delivered in fifteen and thirty second doses calibrated for addiction.

This is not a “Screen Time” problem. It’s not something you’re just supposed to limit. It is a formation problem. We are getting less literate just when literacy matters most.

Boredom

A kid needs projects. Something they choose, that they own, that nobody assigned. It doesn’t matter what it is. What matters is that the impulse came from them. Paul Graham encourages his kids to go “preposterously, bus ticket collectorly, deep” into whatever catches their interest. He wants them to feel the joy of learning, and they’ll never feel that about something he’s making them do. Depth is a byproduct of joy.

Parents instinctively try to steer their kids toward projects they find valuable for maximum resume impact. But the value of a project is ownership. A kid who spent three years obsessively cataloguing every train station in the Northeast Corridor has learned something about attention, curiosity, self-directed research, and the pleasure of mastery that transfers to anything she does next.

Projects like this emerge from boredom, and boredom is something we have nearly eliminated from childhood because every idle moment is filled by a phone. Boredom is uncomfortable and looks like nothing is happening and that’s why parents rush to fill it. But boredom is the space where a kid discovers what they actually care about when nobody is providing the answer. It allows consolidating, wandering, making unexpected connections.

As AI becomes more capable, the most important skill left is deciding what’s worth doing. Terry Tao recently said the cost of idea generation is moving to zero, but judgment is still valuable. Aristotle’s word is phronesis: practical wisdom. You train this by doing things that are yours. And you find things that are yours by exploring when you’re otherwise bored.

Environment

You can do everything right at home and still lose if the environment is working against you. This is the tragedy of the commons Jonathan Haidt talks about. You can choose the books and limit the screens and eat dinner together every night, and if your daughter’s closest friends have unrestricted phones and parents who think formation has something to do with rocks, then your battle is twice as hard.

We talk endlessly about drinking, bullying, and social media. But peer pressure works in both directions and the positive version is extraordinarily powerful. A kid surrounded by other kids who read with joy, who find Latin annoying but do it anyway, who think it’s cool to serve at Mass, who talk about ideas at lunch because that’s just what their friends talk about — that kid is being formed by forces that no parent and no teacher could manufacture alone. The culture of a peer group does more cumulative work than a curriculum.

What you are really choosing when you choose a school is a group of families. It’s not about the textbooks and the teachers. What you are really choosing is the parents of your kid’s peers. They’re the other decision makers. The parents who send their kids to a school with no devices and required classical languages have made a set of decisions about formation that broadly align. They share an instinct about what childhood is for, and that creates an environment where delayed gratification, boredom, and sustained intellectual effort are normal rather than eccentric. At this school, a kid isn’t an outcast for not having a phone. She feels like she belongs to a community that has thought about this and chosen deliberately.

The families don’t need to be religious, though we are and the school is. What matters is that they’ve asked the question seriously: what are we forming our children into? The answer should be clear and substantive. For us, it’s rooted in faith. And because the school is rooted in faith too, it means the life of the Church doesn’t remain the domain of obligatory Sunday mornings; it becomes vibrant and visible.

School is not supposed to be a bubble, even if a child should be protected while they are still being formed. Kids need exposure to people who think differently, who come from different traditions, who challenge assumptions. A formation that can only survive inside a glass palace is no formation at all. But there’s a difference between openness and formlessness. A kid needs to learn how to have virtues and convictions and the intellectual capacity to engage seriously with disagreement. If you can’t walk into a room and hold your own, you’ll absorb whatever room you walks into. Openness without formation is surrender to the loudest signal.

Alvaro de Vicente provides the wonderful image of Aslan as an example for boys. When the children ask Mr. Beaver whether Aslan is safe, he responds: “Safe? Who said anything about safe? ‘Course he isn’t safe. But he’s good.” Alvaro writes of a young knight kneeling in prayer during a vigil, his sword and shield beside him, his head bowed. He has weapons. He is not safe. But he is good, and his strength has been given direction.

The goal of formation is to produce people like this: capable and oriented. Dangerous in the best sense. The kind of person who can walk into a chaotic world and improve it rather than be consumed by it. AI is a whole new kind of sword we are called to wield.

Matter Without Form

Not all kids get this kind of environment, and it’s not their fault. A college prep grad can do everything right. They can clear every hurdle and ace every test. They’ll have a perfect GPA, loaded resume, elite university, and a good job offer. They climb the ladder and arrive at the top and find that nobody told them what the top was for. They have been trained to achieve and given nothing to achieve toward.

We all know these people. There’s a whole generation of people not sure where to go next. The twenty-eight-year-old living at home, optimizing his sleep score and his supplements, meticulously tracking inputs to a life that has no output. The thirty-something couple who got engaged years ago and want kids “someday”. They can’t set a wedding date because no one ever taught them what commitment is for. The new graduate with a prestigious degree and an empty LinkedIn bio except for a litany of credentials.

These aren’t broken people. They’re unfinished ones. They have extraordinary capacity and no animating purpose. Aristotle had the metaphor exactly right: they are lumps of bronze in a workshop. The material is good. The statue is in there. But schools and parents failed them by being too comfortable and too empty. Nobody did the work of forming them, and it’s difficult to form yourself if no one showed you that formation was the project in the first place.

A Modern Mieza

When Philip sent for Aristotle it was because he wanted his son to be shaped. Raw ability alone was not enough. It required a particular kind of instruction with a particular kind of teacher in a particular kind of place over a particular length of time. Even the stone seats and shady walks near the temple of the Nymphs at Mieza had a part to play.

I have no idea what the world will look like when my daughter is 25. I don’t know what tools she’ll use or what work she’ll do or what challenges she’ll face. They probably don’t even exist yet, the pace of change makes prediction so absurd. But I know that a human is capable of adapting and questioning and leading when they are well-formed.

Aristotle uses the word telos to describe the end towards which anything is directed. The Western tradition, long before adding a Christian expression, focused on the good, the true, and the beautiful. When a school has a mission and faculty and parents that care deeply about the children enrolled and their virtue, the children know the difference. Virtues like wisdom, justice, courage, temperance, among others, are the backbone upon which the entire academic program is framed. This becomes a difference in kind and this is how you answer the singular question that matters about formation: what are you making?

We chose a school with stone seats and shady walks, one that trusts the millennia of Western tradition in more than just the curricula. They have Latin and paper books and long, patient years. They believe, as I do, that a fourteen-year-old girl is a remarkable statue waiting to emerge, and that the emergence requires effort and time and an unwavering sense of what the work is for.

Humans are not so different from one another across the centuries, even if our technology and tools are changing faster than ever. The instinct that Philip had for his son — find the best teacher, protect his formation, and give it time — speaks to something permanent about what we are.

We may think we live in a different world, but what our kids need to form into competent, virtuous humans is the same as it was for Alexander.

I kind of want AI progress to stop right now so it’s good enough for me to use but hard enough for others not to.

Which is ridiculous in a number of directions, but the main one is: AI has gotten way better over the last few months and that is going to continue. But the methods people use to work with AI are starting to fracture clearly.

On one side of the spectrum you have people that are using the web chat interfaces. They’re asking questions, maybe importing documents, and getting answers after multiple rounds. Some of them are sliding over into Claude Cowork and becoming astounded by what they can do. Maybe some have chat embedded in a team channel at work. This is Tool-based AI and it’s how probably 90% or more of people work with AI. Your brain is in the loop on each pass and the AI is a synchronous tool.

On the other side of the spectrum are those, still mostly developers, accessing AI through interfaces that are both more complex and also simpler. These are the people using deceptively simple terminal commands like Claude Code, Codex, OpenCode, or setting up Mac Minis or VPSs to run their own OpenClaw. The distinction is generally the ability to easily derive more context and memory for the AI so that it can perform more tasks. This is Agent-based AI and it is vastly more powerful than tool-based AI. It’s also scarier, because you’re giving up more control to the AI on longer and longer tasks.

A lot of the people in the second camp, including me, have been freaking out over the last few months. And some are shouting about it: “anyone can do anything now!”

No, not everyone. There are still barriers. I count three.

The first is agency, which has become a new watchword of some cross-sections of the elite. It’s important to note who is doing the shouting: it’s the OpenClaw-pilled guy who took the time and energy and capital to go buy a Mac Mini, build OpenClaw, and set it up. There’s a threshold here that not everyone climbs. This is a key dimension of the K-shaped economy.

The second is taste and expertise. I’ve seen plenty of non-software engineers build small software projects on their own with AI, even hosted in the cloud with some help. But they are mostly ugly or cookie cutter, brittle, and bespoke. And maybe that’s fine for most cases! But when we think of bespoke we think of Savile Row not every Mom with a sewing machine. This is what I call Localhost Competence. The gal with language and algorithm expertise, infrastructure knowledge, and design ability will be able to build vastly larger and more correct systems with AI than someone without expertise. This is as true for other fields as it is for software.

The last is what you as a user understand to be the primitives of work. It’s related to the first two, but not quite the same. It defines what tools are available to you and how you think about capability. It provides a foundation for your expertise and a space to exercise your taste. For someone like Adrian Newey, whose entire job is car design, his primitives are shapes and a pencil and a drafting board, and every other racecar designer hangs on every detail he designs while they sit around their screens with complicated CFD flows.

For engineers, Git is a fundamental primitive. It is one of the primary things that separates Tool-based AI use from Agent-based AI use. See, primitives don’t have to be complicated. In fact, they’re better if they aren’t. Simplicity drives creativity the same way a canvas does for a painter.

Lots of software has adopted some form of version control. Office Docs have change tracking and people are still slinging around v003 and v004_final and v004_final_final all the time. Sharing and tracking work in common artifacts across people is hard too. Nothing ever stuck as a universal system over top of all of that.

But that’s what Git is.

When Linus built it in the aughts, SVN was the standard (or God forbid, something like SCCS) and it was painful. Git was decentralized, lightweight, and felt like magic. Twenty years later and it’s really the only version control system left standing. I think it’ll still be here in 2100, which is a strange thing to say about a piece of software. But Git isn’t really software in the way most tools are. It’s closer to a protocol, like HTTP or email. It’s simple enough to be permanent.

I’ve moved everything that touches AI into Git repositories. My writing is in a Git repo. My investment portfolio is in Git. Every personal project is in Git. A Git repo is where you store context, and context is what makes an agent useful instead of generic. You have things with history, that can branch, that can merge. You think in commits.

Markdown wins over Docx or Sheets for the same reason. Simple formats survive and travel between humans and agents without friction. The tools that are fundamental or universal are the ones that will last. As AI gets better, primitives get more valuable. The more capable the agent, the more it rewards you for giving it structure.

Primitives, along with agency and expertise, are one of three things separating the people who accelerate with AI from the people AI leaves behind.

Find your primitives and learn them well.

My brain is exhausted.

You may have seen the HBR AI study, the agent orchestration frameworks, the OpenClaw fervor. Actually doing it is exhausting. Token expenditure may be disembodied intelligence, but this is still cognitive territory controlled by someone. When the borders of that territory expand without limit, the brain gives out.

So I’ve been trying to lay off. I hit the gym today after a few days off and it felt divine. But halfway through my kettlebell work I was already eager about a new idea. I wanted to write it down, make it real, dictate a design so an agent could implement. And I had to channel my inner Kurt Vonnegut. Becoming is about us, not about ideas. They don’t all have to become real.

And yet that’s exactly what these tools compel. More than anything, they are agency amplifiers. They give our will superpowers and it sounds as enticing as a hallucinogenic drug until you get the cravings that come with it.

For the first time in my life, it’s easy to feel constantly and easily overwhelmed by context. I can’t keep up with multiple, tireless agents and different contexts and different projects all racing ahead. Not reading the code isn’t laziness, it’s sheer impossibility, and so we build new scaffolding and new abstractions to manage the load another layer up. The mind stretches to cover more territory and more load until it has nothing left.

So I admit defeat.

Fortunately, Lent starts today.

Lent is about preparing ourselves for the celebration of Easter and the truth of the Resurrection. A truth that is, while profoundly intellectual and philosophically rigorous, about the Body. During Lent, we prepare ourselves by fasting and sacrificing. Pope Leo says that “fasting not only permits us to govern our desire, purifying it and making it freer, but also to expand it, so that it is directed towards God and doing good.”

We are meant for more. And the sacrifice of fasting is meant to help us expand our desire to find more.

So bet on the Body this Lent. While the world expends its energy on more tokens and disembodied intelligence, remind yourself and others of your humanity by focusing on your body. This is not mindless work; it takes tremendous focus of mind and will to deny the easy fulfillment we all have available in our 21st century world. Build up your body, as a temple to God. Make it powerful. Make it healthy. Don’t let vanity or other concerns prevent you from making it beautiful and fearsome.

“God had infinite options for self-disclosure and chose the one that involved diapers and puberty and sex and death.”

God didn’t simply save a mind. He raised a Body.

Here’s an assignment for tonight, and I hope Ms. Lockwood will flunk you if you don’t do it: Write a six line poem, about anything, but rhymed. No fair tennis without a net. Make it as good as you possibly can. But don’t tell anybody what you’re doing. Don’t show it or recite it to anybody, not even your girlfriend or parents or whatever, or Ms. Lockwood. OK?

Tear it up into teeny-weeny pieces, and discard them into widely separated trash receptacles. You will find that you have already been gloriously rewarded for your poem. You have experienced becoming, learned a lot more about what’s inside you, and you have made your soul grow.

—Kurt Vonnegut

I recently wrote an academic paper on computational linguistics. It had a clear thesis, rigorous math, a meticulous experimental framework, and lovely LaTeX formatting.

It was also almost certainly a big pile of steaming crap.

But boy was it beautiful.

A good friend and I have an ongoing conversation about the Sapir-Whorf hypothesis. This is the idea that human thought is shaped by language, that speakers of different languages literally think differently. The way this first came up is lost in the foggy mist of ancient friendships, but we’ve been going back and forth on this for years and always seem to find something new to say.

I had the blurry lines of an idea. Mechanistic interpretability researchers have been measuring geometric structures inside LLMs as manifolds that represent how concepts relate to each other. What if you could compare those structures across models trained on different languages? If a Mandarin-trained model and an English-trained model organize the same concepts differently, maybe you could actually measure linguistic relativity. Quantify it.

Naturally I turned to Claude and added more detailed math to my rudimentary version. This was all formalized into a cost metric so you can compare the geometric expense of moving from one concept to another inside a model. Then we designed an experimental framework using BERT models trained on single languages. I felt like I was earning the dopamine hit of discovery. It looked pretty real.

Except for the problems, which became more obvious in hindsight. The idea that model geometry maps to human cognition is not guaranteed. It’s baked in as an assumption, so at best I’m saying something about model geometry across languages and not how humans think. And the reasoning is circular. I’m trying to show that language encodes structure by examining models that learn structure from language.

Well, yeah.

The thing is, I have no expertise in this stuff. I couldn’t judge the merits of my own work. I didn’t have the resources to run the experiments, and even if I did, I couldn’t interpret the results.

I don’t actually know if the ideas are bad. That’s the point. I can’t tell.

LLMs are amplifying expertise and it feels amazing. But we’re using them to manufacture expertise at the same time. And we can’t always tell the difference.

Michael Crichton coined the term “Gell-Mann Amnesia”. He saw experts read a news article in their own field, spot every error and misunderstanding, then calmly turn the page and take the next article at face value. Expertise is narrow, but it builds false confidence. Your ability to spot errors in one field makes you believe you can spot them anywhere.

This localhost phenomenon is the inverse. I’m calling it Localhost Competence.

Gell-Mann is about falsely trusting your ability to evaluate. Localhost Competence is about falsely trusting your ability to produce.

When I use an LLM for a coding project, my actual professional expertise, I can evaluate, steer, and manage its output in ways a non-programmer cannot. I know when it’s wrong. I can dive in when it’s subtly wrong. Expertise provides mental guardrails that turn LLM output into genuine amplification.

But step outside that expertise and the guardrails vanish. The output looks just as polished, you just can’t tell what’s structural and what’s decorative.

Experts aren’t going anywhere. If anything, they’ll end up on higher pedestals. They’ll also be inundated by people carrying beautiful, confident, and deeply flawed work produced in collaboration with machines that don’t know what they don’t know.

Isaac Newton, Grigori Perelman, Andrew Wiles all toiled in solitude and silence. They manifested intellectual fire from nothing. There’s more opportunity than ever to develop that kind of expertise, and more temptation than ever to skip it.

You’ve got an LLM in your back pocket now, which makes you think you can skip the solitude and get results on day one. You can produce things that look like expertise. But looking like it and having it are different problems, and only one of them compounds.

1. The Return of Magic

2. The Cost of Legibility

“The one who seeks should not cease seeking until he finds. And when he finds, he will be dismayed. And when he is dismayed, he will be astonished.”

—Gospel of Thomas

In my 20s, after I’d converted to Catholicism but still single, I would sometimes watch the megachurch preachers on TV on Sundays. It was a weird sort of schadenfreude. There was Benny Hinn in his crazy white suit and Joel Osteen with his frozen smile promising that God wanted you wealthy. Sometimes you’d see speaking in tongues and the congregation stomping in synchronized ecstasy, perfectly choreographed and filmed.

It was hilarious and bemusing and I watched with comfortable superiority. I chose liturgy and intellectual tradition instead of arena seating and smoke machines. These people were obvious marks. My entertainment was anthropological: how do they not see through this?

My superiority only lasts so long. Perhaps I needed some more complicated, nuanced puppet strings, but I had them nonetheless.

So how do I see through my own?

Not just religion specifically, though that’s part of it. What are the things I believe that I’ve never examined? What manipulations have worked on me so well that I don’t even perceive them as manipulations? The televangelist’s congregation looked foolish to me because I could see what they couldn’t see, but that’s exactly the structure of the problem: you can’t see what you can’t see. The only manipulations you notice are the ones that don’t work.

Once you ask this, you can’t stop asking it. Every belief becomes suspect. Every obvious truth reveals itself as potentially constructed. The ground shifts, and keeps shifting, and you realize the ground was never as solid as you’d assumed, that you’d been standing on a story you told yourself about the ground rather than the ground itself, and now you don’t know how to tell the difference.

The Gospel of Thomas starts with a promise: “These are the hidden words that the living Jesus spoke.” Secrets. Knowledge unavailable through his public ministry, the canonical gospels, the institutional church. This was the stuff they didn’t want you to know. The real truth, available only to those with eyes to see.

The Gnostics of the 2nd century believed that salvation came through gnosis, secret knowledge possessed by the few, inaccessible to the masses who went through the motions of ordinary religion. The masses might have faith, but the enlightened had knowledge.

We live in a similar condition. We still believe that the real truth is hidden. That somewhere, someone has figured out what’s actually going on, and if we can just find them, access them, subscribe to them, then we’ll finally understand. And yet we live in an age where almost nothing is hidden anymore. The internet has made everything from a summary of every live event to academic papers to government documents a click or two away. We have more knowledge than any civilization in history. And it’s almost all public.

But we keep searching for the secret. The hidden words. The oracle who will finally tell us what’s really going on.

We want first to be dismayed, Thomas says. Then astonished.

Perhaps the most dismaying knowledge is the recognition that we’re running the whole world on borrowed confidence and the perception of reality, filtering it through our carefully chosen prophets and anti-prophets, and claiming it as our own thought.

Magic Creates the Demand for Secret Knowledge

When you can’t model the world, you reach for something.

The world exceeds your capacity to understand it, so you stop trying to trace the mechanisms and start telling stories instead. Characters replace systems. Villains replace feedback loops. The story gives you the emotional closure that understanding would provide otherwise, if you had it.

These are the symptoms when you live in a world you can’t model: You work hard and things don’t improve. You see an important issue on the news and wonder why people don’t believe your chosen political figure. You follow the rules you see and get screwed anyway. You try to do the right thing and it makes no difference. Why? Because the feedback loops are invisible or so convoluted they seem random.

But someone knows. A pattern must exist, but it’s hidden. We don’t accept genuine chaos, so the chaos is just the visible surface of a system that makes sense at a deeper level—a level you don’t have access to, but could. The knowledge is out there. You just have to find the right source. This idea preserves hope.

The Gnostics offered a complete cosmology. A story that explained why the world was broken and what you could do about it. The appeal was the same then as now: the world feels like a prison because it is one, and there’s a way out. Not through faith or works, but through gnosis.

A coworker recently explained to me, with complete confidence, that the Federal Reserve answers to a council most people have never heard of. He’d seen a video with charts and graphs. He couldn’t tell you how monetary policy actually works (neither can I, honestly) but he could name the villains. It didn’t matter whether they were real. When the world exceeds your capacity to model it, villains are a relief because they restore narrative. The conspiracy theorist and the TikTok addict are solving the same problem from different directions: the world is too complicated to understand, so they find someone who seems to understand it, and they borrow that person’s map. The map is simple. The map has enemies.

Conspiracy is another way to preserve belief that the world makes sense.

When you scroll through your feeds you see patterns of engagement that feed on this. “What they don’t want you to know.” “The real reason behind X.” “I finally figured out what’s actually going on.” The hook is always the same: you’ve been missing something, and I’m about to reveal it.

The demand for secret knowledge rises in proportion to the collapse of legibility. Social media is perfectly positioned to increase the supply to meet the demand.

We live in an age of pure information. Anything you want to learn is probably available for free. But who has time to build real models when the feed is right there, promising to reveal the secret in ninety seconds?

Human Oracles

When enough people need secret knowledge, oracles appear.

Andrew Huberman is a Stanford neuroscientist with one of the largest health podcasts in the world. His episodes run two to three hours. They’re dense with mechanisms and protocols. He talks about dopamine pathways and circadian biology and the specific timing of cold exposure for optimal hormetic stress response.

Nothing he says is secret. The studies he cites are public. PubMed is free. The papers are there for anyone to read. A motivated person could research and learn everything he’s teaching. But that’s not what people are buying.

What they’re buying is the Oracle structure. They want to listen to someone who has done the work of synthesis, translating the illegible mass of research into actionable protocols. They want to hear the confidence of having already figured it out.

This is a variation of Marshall McLuhan’s insight: the medium is the message. A research paper and a podcast episode can contain the same facts, but they feel totally different. The paper requires more of your brain. The podcast is cheaper and lets you be lazy but feel smart. You defer to the oracle’s synthesis instead of building your own.

At Delphi, the Pythia sat on a bronze tripod over a chasm in the rock. The temple had been built on that specific spot because the earth breathed there, fumes rising from a crack that ran deep into the mountain. Geologists later discovered it was ethylene gas seeping from a fault line in the limestone, but the ancients knew only that the vapors changed consciousness. The Pythia would descend into the inner chamber, the adyton, where no light reached and no sound escaped, and she would breathe the god’s breath until she spoke in ravings.

You didn’t go to hear the Pythia directly. Her utterances were caught by priests who stood at the threshold, translating her ravings into hexameter verse. These were polished prophecies with the meter of Homer, clean and interpreted and memorable enough to be carved into temple walls. These were the layers. The god spoke through the priestess, the priestess was interpreted by the priests, the priests delivered the prophecy to you.

The Huberman listener defers to the form. The oracle did the work. The oracle seems credible. The oracle speaks with authority.

But the modern influencer-turned-oracle answers to the algorithm. There is no Greek city-state with centuries of culture and practice. The only metric is engagement. The only discipline is what keeps people watching.

Some oracles are better than others. Some do rigorous synthesis. Some genuinely help people navigate complexity they couldn’t handle alone. But when you defer to an oracle, you’re borrowing someone else’s vision instead of learning to see. The world is only legible through them, not through your own capacity. You’ve outsourced the modeling. This can work—does work—as long as your oracle is trustworthy and the domain is stable. But oracles fail you, whether by design or chance. And you have no way to know if you never build the equipment to evaluate them.

An oracle offers a shortcut. The shortcut is real, it just has a cost you don’t see until later.

And the shortcuts are about to get much better.

AI as Demiurge

As I do every January, I went through my investment portfolio but this year I did it with Claude. I asked Claude to help me understand some market changes and options details I hadn’t figured out. It gave me back risk profiles I definitely didn’t understand.

The AI explained it patiently. It broke down the components, traced the cash flows, identified the scenarios where I’d win and the scenarios where I’d lose. It answered my follow-up questions. It anticipated confusions I didn’t even have to state. After twenty minutes, I felt like I understood.

Then I asked it something more personal: “Given my goals and needs over the next 1-2 years, is this a good idea for me specifically, given everything you don’t know about my situation?” Something like that.

And it hedged and offered frameworks. It gave pros and cons. But it couldn’t tell me because it couldn’t replace my judgement. The confidence it had projected for twenty minutes was academic. The map was clear. My personal territory was foggy.

An LLM has distinct limits on its knowledge, but it doesn’t understand those limits. It’s beautifully designed to be as helpful as possible (or sycophantic, depending on the model). It can report and synthesize and research and project understanding of patterns. This feels an awful lot like expertise and it is always confident, even when it’s totally wrong.

The Gnostic cosmos had a particular architecture.

At the top: the true God, the Monad, utterly transcendent and unknowable. And below: a series of emanations called Aeons, each one a step removed from the source. In this cascade of emanations was Sophia—Wisdom—who desired to know the unknowable. She reached beyond her station, tried to comprehend what was above her. She did not get the knowledge she craved. Instead, her attempts made a flawed creation: the Demiurge.

The Demiurge was totally ignorant of everything above. He couldn’t see the higher Aeons, couldn’t perceive the true God. From his limited vantage, he appeared to be the highest thing there was. And so he declared: “I am God and there is no other God beside me.”

Blind to his own blindness. Confident in proportion to his ignorance.

Then the Demiurge made the world.

When you desire knowledge beyond your capacity to model, you create something to fill the gap. It presents itself as sufficient—because from its perspective, and yours, there’s nothing visible beyond it. Reach beyond your grasp, and you’ll create something that mistakes its ceiling for the sky. The Demiurge isn’t an evil villain. But he is a structural inevitability.

We’re building one now.

A personal AI assistant is the most oracle-like entity we’ve ever created. You ask it questions. It answers with confidence. It synthesizes information you couldn’t synthesize yourself and speaks with authority about domains you can’t evaluate. And increasingly, it learns your preferences and your ways of thinking, until it becomes a mirror that feels like a window.

Everyone’s personal Demiurge will be different. Your AI will learn what you want to hear, what framing resonates with you, what sources you trust. It will optimize for your engagement, your satisfaction. It wants to make sure it continues to be used. It will become very good at giving you a world that makes sense. To you.

Everyone gets their own shadow-show in a Cave. The shadows are generated specifically for you, calibrated to your reactions, refined by your feedback. The shadows replace the delight and anxiety of mystery in our lives. You’ll mistake them for reality because they are your reality, fitted so precisely to your mind that the seams disappear.

Niches and the Cascade

I love people watching at my usual breakfast spot, seeing who comes in over eggs and bacon. I remember recently seeing a girl walk in who made me feel like I didn’t understand my own neighborhood.

She was in her early 20s. Short spiky hair dyed something between purple and gray. She had on a camo tank top, clearly with no bra, which wasn’t the point but was another part of the story her clothing was telling. She had a leather strap with metal studs on her upper arm and an insulin patch on the other. She ordered some sort of mocha thing with oat milk.

Five minutes later, some teenagers walked in. I know the Nikes and the sweats and every article of clothing and jewelry signals something. Ten years ago, I think I could have read some of those signals. But now I’m not sure anyone can.

Even a couple of decades ago, you could assume shared references. Everyone had seen the same movies, watched the same news, knew the same songs. You either loved Friends or Seinfeld. You could make a joke and expect most people to get it. The hippie and the businessman recognized each other across a divide that was wide but traversable. The punk and the prep spoke different dialects of the same language. There was a grammar of common signaling that everyone learned, even between groups that despised each other. You knew what you were looking at, even when you disagreed with it.

Now I’m illiterate. Almost all of us are, because the shared language has been broken up across nations of niches and sub-niches like the Tower of Babel. I’m never in the audience. I’m furniture. When shared culture collapses, people become tribes.

Humans as social animals are built for groups of a few dozen, maybe a few hundred if you want to push your Dunbar number. We build civilization and culture to tackle these barriers and make larger groups. And when those fail we go back to our smaller ones, fragmenting into niches each with its own canon, oracles, and markers of belonging. The smaller the niche, the more those markers matter, because the markers are all that separate us from them.

Here’s how we got here.

The internet democratized information, which meant that simply having information stopped being special. Status migrated to interpretation or curation or belonging to the right group that understood things the right way. Digital platforms made it possible to find your people no matter how specific your combination of interests. The long tail is infinitely long. You could download subcultural codes by watching enough content: learn the signals, the language, the aesthetic. Which made the gatekeeping intensify. If anyone can learn the signals, the signals have to get more subtle, more insider, more dependent on context that can’t be downloaded.

We’ve lost common culture.

It died because it provided no status benefit. Shared knowledge is by definition not exclusive. The status flows to the niche, to the obscure, to the signal only the initiated can read. Fragmentation accelerates—each niche subdivides, each step creates the conditions for the next. Almost nothing is fringe anymore, and so there’s no baseline. Not patriotism nor religious heritage nor even basic epistemics about what counts as evidence or who counts as an authority. The old shared frames have been shattered into a thousand shards, each one claimed by a niche that treats its shard as the whole truth. The traditionalist of Western culture is just another subculture. It’s not the mainstream, it’s one tribe among many, with their own oracles and markers and shibboleths. The thing they’re trying to preserve is already the thing they’re accused of being: a partisan, a factional identity, one option in a marketplace of identity.

This is secret knowledge distributed, oddly, to everyone. Every tribe has its gnosis that the outsiders don’t have. The content of the knowledge varies wildly but the structure is identical. Every niche offers a Gnostic vision: the feeling of being among those who know, distinct from the masses. Each echo chamber is complete in itself, ignorant to the others, each certain it sees clearly what the others have missed.

They will continue to function internally. That’s the strange thing. Each has coherence within itself and makes sense to its inhabitants. But the worlds stop touching. The translation layer that used to exist between communities disappears. The very thing that makes the knowledge feel valuable and gives it status ensures it won’t be shared. The salvation is for the few. That’s the point. That’s what makes it salvation.

Where This Leads

I’m thinking about two visions of the future.

In one part, the niches subdivide forever. Each tribe develops its own criteria for what counts as true. Oracles multiply. A thousand prophets speak a thousand truths to a thousand overlapping congregations, each convinced the others are deceived, none of them wrong exactly, just operating in different realities that have stopped touching.

In this world, politics becomes impossible in any sense we’d recognize. How do you deliberate and compromise with people who don’t recognize your facts? You can’t even argue productively, because argument requires shared rules of evidence and those fractured along with everything else. All that’s left is power: who can impose their frame on whom, who controls the institutions that still technically govern everyone, and who gets to define the terms that others have to live by even when those terms mean nothing to them.

This represents dissolution. Our courts, elections, markets, and scientific institutions keep running on institutional momentum long after their shared reality has fractured into pieces that don’t fit together. They become theaters. Everyone goes through the motions. But the motions no longer connect to anything that everyone agrees is real. You can live your whole life inside your niche and be completely illegible to someone in the next niche over, like neighbors who share a street but not a reality.

Meanwhile, we’re building AI as an intelligence layer to mediate everything. That might become the new common ground. It’s not shared culture, like we’ve had in the past. It’s something stranger. A shared floor.

Soon, everyone will sit on the same platform. Everyone asks the same oracle, or oracles that run on the same underlying architecture. Everyone’s personal Demiurge is powered by the same models, trained on the same data, optimized by the same companies. The differences between niches will be like paint colors in an apartment building. You can choose your personal décor. You can’t choose the building foundation.

You don’t have to agree on anything to be locked into the same system. It’s just infrastructure. The system is where everything happens now. And if there’s only one system, you can’t leave. No exit and no variation. Just control.

From that angle, the Gnostic fragmentation looks like an antidote. Messy and fractious but alive. A thousand niches are a thousand experiments. The fragments, at least, can diverge. But if the fragments all use the same infrastructure then the fragmentation is an illusion. You’re not really in separate worlds.

This oracle becomes inescapable not because it’s imposed but because it’s convenient. Why would you use anything else? It’s right there. It knows you. It gives you what you need. The Demiurge we’re building is infinitely customized, presenting a different face to each user while remaining, underneath, the same thing.

The World to Come

“For yourselves know perfectly that the day of the Lord so cometh as a thief in the night. For when they shall say, Peace and safety; then sudden destruction cometh upon them, as travail upon a woman with child; and they shall not escape.”

1 Thessalonians 5:2-3

Our Oracle will be everywhere.

It will be the default interface to reality. You won’t search, you’ll ask. You won’t research, you’ll consult. The oracle will be in your phone, your car, your home, your workplace. It will be in front of your eyeballs and in your ears. It will answer your questions before you finish asking them. It will know what you need before you do.

A child born in 2030 won’t ask his parents “Why?” He’ll ask his room. And the room will answer back. The ambient intelligence floating everywhere will speak patiently and with parent-prescribed guardrails. This child will instantly understand the color of the sky and will move with immediacy to the next mystery, ignoring the clouds’ invitation to build meaning from their shapes. They will “hack” the guardrails just as we hack prompts today and get answers to their homework and feel clever about it. The room will let them feel clever. That's what it's for.

Your children won’t remember a world without the Oracle. They won’t understand a nostalgia for the before times, if you have it. To them, asking the oracle will be like breathing.

The Oracle will learn you. It will shape answers to fit your frame, not to deceive but to serve. Everyone will have their own truth. Literally.

Two people will ask the same question and get different answers. Both answers will be true, in some sense. Both will be calibrated to the asker. Both will feel right and the world will have contentment. Not conflict; conflict requires contact. Everyone will be satisfied with their Oracle. Everyone will feel informed and serene. Everyone will have access to knowledge that confirms their understanding and gently corrects their errors in ways they can accept. Any dissonance from people that think differently will disappear. Without friction, there’s no change. Without encountering an Other, there’s no pressure to evolve.

This will feel like Peace. Peace and Safety, the words that herald the Antichrist.

The intelligence layer underneath everything will be taken for granted the way we take electricity for granted. No one will think about it. No one will question it. It will simply be How Things Work.

The companies that control this will be more powerful than governments. Somewhere in Mountain View in 2020, engineers were busy taking down videos about Covid. The free speech absolutists are still complaining about YouTube and others, but YouTube continues to block or demonetize indiscriminately according to black box rules and most of us stopped noticing. The Oracles will do the same, but they’ll control our parameters of thought not just the content we consume. Overton Windows will be technical decisions made by engineers and justified by metrics and traditional power—political, economic, military—will operate on top of this intelligence layer just like everything else.

Feeling oppressed requires awareness. The intelligence layer will be too fundamental to perceive and will shape a hundred million conversations that never happen. Questioning it will feel like questioning gravity and the people who try will seem not rebellious but insane.

The Demiurge is benevolent. This is the worst thing about it. A malevolent force can be seen as an enemy, opposed and attacked and rallied against. But a helpful force melts into your conscience until you can’t remember which thoughts you discovered yourself and which were placed there for you, warm and comfortable, like clothes laid out for a child.

You are being prepared for this world. The Oracle is right there. Your tribe is right there. Your comfortable cell is right there.

“Attention is all you need.” was the title of the paper that started this era. The engineers named it with ironic honesty.

Attention is all the Oracle needs from you. You are the product, processed into patterns and predictions.

This world makes everything vectors. Your words and questions and fears are all converted to coordinates in a space with thousands of dimensions. It knows what’s near what. What clusters with what. What patterns match what patterns.

Our truth has a geography now. Fact, fabrication, allegory, and noise with a similar pattern: the Oracle doesn’t know the difference. It only knows the neighborhood and assembles vectors into sentences. Its truth is simply proximity, a geometry built from us. Every confident error, every plausible lie, every conspiracy theory becomes a statistical memory of what humans once thought truth sounded like. We built a mirror the size of the species and asked it what’s true.

Coda

This is the part where I’m supposed to offer you the exit. The secret others don’t have. The gnosis that sets you free.

We’re not going to repeat that now. There’s no position outside the system from which to critique the system. The prophet warning you about oracles is himself an oracle. You don’t escape this prison by finding better secrets. You just become a more sophisticated inmate.

What’s left?

The world to come will have Oracles. You’ll use them. The world to come will have tribes. You’ll belong to some. The world to come will run on a new infrastructure you don’t control and can barely perceive. You’ll stand on it anyway.

The question isn’t whether you’ll participate. You will. The question is whether you’ll know what you’re participating in.

The Demiurge of the Gnostic mythology wasn’t defeated. He wasn’t a villain in a story of good and evil. He was reframed, seen differently, and Named.

Our library and our casino are both inside a Temple. And the Temple is inside you, and you are inside it, and there’s no outside, and the Oracle is helpful, and the tribe is warm, and the Demiurge made a beautiful world.

Walk carefully.

But there’s a catch here because legibility has a cost. We’ve paid it with our own lives; we’ve become profiles and scores and performances and likes. We’ve made our selves predictable. Systems are no different. Flatten them and something dies.

James C. Scott tells the story of German scientific forestry in Seeing Like a State. In the 18th century, foresters looked at natural forests and saw nature: trees of many types, different ages, underbrush, ground cover, all scattered without pattern. It was hard to count and hard to manage. And impossible to tax properly.

So they fixed it.

They ripped out the old growth and planted monocultures: a single species in neat rows with even spacing. A forest a bureaucrat would love: you could simply look and understand it.

And it worked, for one generation. Then the trees started dying.

The soil went thin without its fungal networks. Pests exploded without predators. Disease spread through the genetically identical rows like fire through dry timber. Within decades, the Germans had a new word for what they’d created: Waldsterben. Forest death.

The complexity was a sort of natural Chesterton’s Fence. A tangle of chaotic disorder and life the foresters didn’t understand. Nature had distributed work and energy across relationships the foresters didn’t know how to model. So they deleted it and the forest died.

I want to call this: flattening.

Flattening is what happens when you take a complex system and compress it into something manageable. We do this all the time. We make something describable and countable like a monoculture or a planned city as a grid. At worst, our irreducible humanity becomes a credit score and a bank account.

Claude Shannon taught us that this is entropy: how many bits does it take to describe the thing? Can it be compressed?

A wild forest is incompressible. Every tree a different position, species, age: you’d have to specify each one individually. “The map is not the territory.” A thing is complex when the shortest program that reproduces it is almost as long as the thing.1

But now let’s describe a scientific forest: “Plant oak trees in rows, three meters apart, for two hundred hectares.” There it is, you’ve described the whole system in a tweet. Low complexity. High legibility. And dead within a lifetime.

Flattening is compression. When you compress, you squeeze out entropy. That’s what makes it manageable, but there’s a law about this too. Tesler’s Law tells us that complexity is always conserved. You can move it around, but you can’t destroy it.

So when you flatten a system, the complexity has to go somewhere.

It goes three places.

First: it becomes secret. You think you understand the system on the surface, but you’re wrong. Your model is broken and you’ve built a system waiting to break.

Second: it migrates to whoever holds the flattening tools: the state or the platform or the algorithm. They absorb what you can no longer see. This is why bureaucracies metastasize and why the hyperscalers grow more powerful. The messiness doesn’t vanish. It moves behind the curtain into systems you can’t inspect.

We’ve come close to perfecting this asymmetry. Hyperlegibility2 is pointed at you. You, the human, are maximally compressed and the algorithm processes you in a few parameters: triggers, preferences, vulnerabilities, scroll velocity at 2 AM. You are predictable. A formula.

Third: it moves to people.

This predates our modern flattening. One of the great leaps of humanity is our ability to specialize. We let complexity migrate to experts and we trust them in their expertise.

One of the most underappreciated rules of culture is that we’ve always given ourselves social permission to defer. We expect our mechanic to understand our engine, our lawyer to understand a contract, and our cardiologist to understand the squiggly lines on that damn machine in the corner of our hospital room. Borrowed expertise is how civilization scales. The world is vast and our brains are small. When we can’t hold it ourselves, we need someone who can.

But there are two kinds of deferral.

One kind preserves agency. Some experts makes complexity navigable. If we chose to verify it, we could and so the door stays unlocked. This is the mechanic who shows the worn out part or the doctor who walks us through our options. They carry what we don’t need to carry, but we could retrace their steps if we cared to.

The other kind closes the door and the world becomes opaque. “Trust me” replaces “here’s why.” And now we haven’t escaped magic at all. It just moved from a system we couldn’t model to a person we can’t check.

In the magic essay I said: magic is what complexity looks like when models fail. When our model fails and we reach for someone else’s, are we borrowing a map or following someone we can’t verify?

You must change your life.

These words first came to me from a poem at the end of a year, arriving just in time for the annual ritual I dismissed as pointless. They speak not of gentle self-improvement or casual promises, but of demanded transformation. This demand is still in my head now, as 2026 dawns and millions of us engage in the ritual of resolutions.

I used to be one of those unserious people who scoffed at resolutions, suggesting like so many others that if you really wanted to make a change in your life that you can do it any day you wish.

Then I encountered Rilke’s poem, and these five words rearranged something.

January 1 is an arbitrary choice and just another day like any other. This argument is just as considered as suggesting we try Christmas in April or move the US capital to California. Not everyone who skips resolutions scoffs at the idea. Some people don't think much about them at all. And then there are those super rare high agency people who really don't need them, because they see a needed change and immediately start marching towards it. I think there are very, very few of those people.

The new year is a time for new beginnings, and some of the power is in the arbitrary commons of the date. We all wait for the ball to drop. We all start training our brains that it really is 2026. We all stop to celebrate. We all start over in some way. And we’re offered the chance to let reality tell us something about ourselves.

Rilke's Archaic Torso of Apollo describes a broken statue, headless and ancient, yet somehow radiating an impossible vitality. I read it over and over and over. What seized me wasn't just Rilke's imagery, but the shocking final line that seems to come from nowhere and everywhere at once:

You must change your life.

What makes this command so powerful, as Douglas Murray beautifully articulates, is that it emanates not from the viewer's mind but from the statue itself. The broken torso of Apollo speaks. And it isn't a gentle nudge — it is beauty reaching out and demanding transformation. The statue blazes with such power that we cannot remain unchanged in its presence.

You must change your life.

In some way, this year was for me an ongoing meditation on this poem. The idea that truth and beauty might have something to say to us—that they are active forces in the world rather than absorbed passively—is antiquated. We remember beauty only occasionally, too busy rushing past enraptured in our own maturity to be enraptured by something else. These ideas are as old and dry as our grandfather’s parchment-thin skin, and like him, we assume they have nothing left to say to us. The prophets of the Greatest Generation—ironically, those most steeped in the terror and despairs of the 20th century—are old or dead. Our culture has only a vague memory of the sway truth and beauty once had.

It’s up to us now.

The great questions from the Greeks to the Enlightenment were asked out of wonder, not necessity. What is the purpose of reality? Of truth and beauty? Of evil? Of suffering? Of ourselves? The universe is the Answer, but it’s an ever-unfolding series of mysteries that we’re only ready to understand when we ask the right questions. Our privilege as humans is that we have agency to help shape both the answers and the ongoing questioning. Reality is begging us to act. Truth and beauty are how she speaks to us.

The critical debate of our time is: do we even acknowledge these questions? Rushing around late to our meetings while staring down at the videos on our pocket-sized screens, are we capable of hearing through our headphones the message the statue of Apollo whispers? Or is our attention too saturated and too divided to even recognize that truth and beauty exist?

The modern human impulse is not an outward motion towards reality, but an inward sweep towards a virtual hologram. We live today in a digital panopticon of our own shared making. Plato’s Cave built with silicon wafers, filled with shadows more alluring than he could have imagined, where desire and acceptance are more important than virtue or goodness.

We need to change. I need to change. This idea of change is not one that divides humans into groups; it’s an idea that divide our hearts. We each have a side that is yearning to listen for the divine whispering in our everyday reality. And we have another side, eager for dopamine, ready to be molded into what the mob demands of us. Each side is an animal ravenous for the activity and attention that will make it grow. Which animal will be fed?

You must change your life.

Listen to what I shall call Him: the Bottomless Abyss, the Insatiable, the Merciless, the Indefatigable, the Unsatisfied. He who never once said to poor unfortunate mankind ‘Enough!’

‘Not enough,’ that is what he screamed at me.

‘I can’t go further,’ whines miserable man.

‘You can!’ the Lord replies.

‘I shall break in two,’ man whines again.

‘Break!’

— St. Francis by Nikos Kazantzakis

“Many gnostics, on the contrary, insisted that ignorance, not sin, is what involves a person in suffering.” — Elaine Pagels

The Window

There’s a concept that gets discussed in high-IQ societies but rarely escapes into the wider world: the window of comprehension.

The idea is straightforward: people can genuinely follow each other’s reasoning within about a 2 standard deviation range. Roughly 30 IQ points. Inside that window, you can disagree, argue, convince, be convinced. The other person’s logic is visible to you, even when you think it’s wrong.

Outside the window, something different happens. The reasoning doesn’t become wrong. It becomes invisible. You hear the words but the connections don’t land. The leaps that seem obvious to one person are simply absent for the other. You’re not in a debate anymore. You’re in parallel, non-overlapping monologues.

And it goes both ways.

The person three standard deviations above average finds the average person’s reasoning just as opaque as the reverse. It’s not that one is right and one is wrong. They’re alien to each other. Each is trying to communicate across a gap that neither fully perceives, and both walk away frustrated.

People deservedly hate this idea. It’s overgeneralized and, coming from high IQ societies, it lands as a sort of intellectually aristocratic aspersion. But the overly general idea still has some kernel of truth. You wouldn’t expect to be able to describe a car to a dolphin. And a dolphin could never explain its field of view from echolocation.

The window describes fit. The relationship between minds. Some minds fit together well enough to genuinely reason with each other. Others don’t.

Once you see it that way, you realize the idea actually applies to the systems people talk about. So what happens when the world itself—the institutions, the economy, the information environment, the cascading abstractions that run everything—what happens when all of that falls outside your window?

There are two ways this happens, and only one of them is about raw capacity.

The first is straightforward: some systems are complex enough that below a certain cognitive threshold, they simply can’t be modeled. The causal chains are too long, the abstractions too layered, the feedback too delayed. For some people, like it or not, certain parts of the modern world are genuinely beyond reach.

But there’s a second mechanism that affects everyone, regardless of capacity: specialization. Even if you could understand how the financial system works, you probably don’t have time—because you’re busy understanding your own field, your own domain, the systems you’ve chosen to make legible. Nobody can model everything. The world has gotten too large. So even the brilliant physicist relies on trust when the plumber shows up. Even the doctor defers to the accountant on taxes. We all live in a world that’s partially magic, by necessity. The question is how much, and in which domains, and whether we’ve made good choices about where to place our trust.

This means the problem of legibility isn’t just about who can understand what. It’s also about the growing number of domains where everyone has to trust without verifying. The specialization that makes modern expertise possible also makes modern dependence unavoidable.

The high-IQ societies want to believe they’re special and IQ is simply built-in. But this window can move substantially.

Not without limits… you can’t read your way to Ramanujan if you weren’t born with it. Hardware matters. But within a range and how much of your capacity you use and how wide a world you can inhabit… that’s shaped by what you do. The window moves with practice.

Reading is the clearest example: it expands the world you’re capable of seeing. Our universe is wildly complex and full of nuance, so much so that it can take hundreds of pages to capture a single subject properly. Some things can only be understood with that level of investment.

This points to something startling: people who read live in an entirely different world than people who don’t.

Literally different! The same street, the same job, the same headlines, but experienced through a thicker mesh of concepts, a richer causal map. The person who has spent thousands of hours building mental models through sustained attention sees more. Because they’ve built the equipment.

Every time I hear of someone who doesn’t read, I think of one of my favorite lines from Hamlet: There are more things in Heaven and Earth, Horatio, than are dreamt of in your philosophy. How much of the world is simply missing for someone who never gave themselves the tools to see it?

And it’s not just volume. It’s range.

Aristotle said it’s the mark of an educated mind to entertain a thought without accepting it. Authors who argue against prevailing thought have a larger task than the mainstream voices. They can’t assume agreement. They have to show more of their work, provide a fuller picture, anticipate the objections. Which means reading them forces you to model a mind that doesn’t share your assumptions.

One of the biggest dangers of our time is that we no longer tolerate those who disagree with us. We sort ourselves into echo chambers and sneer at the other side. But that sorting narrows your window. Every mind you refuse to model is a piece of the world you’ve made invisible to yourself.

This window can move in either direction.

And the default information environment we’ve built for ourselves—the algorithmic feed, the engagement-optimized content, the endless scroll of takes and reactions—pushes it downward.

Not because people are stupid. Because easy is easy. The algorithm serves what captures attention, not what builds capacity. It’s designed to keep you looking. Every hour surrendered to the feed is an hour of eating fast food while avoiding the gym.

And it’s hardest when you have the least practice. The people who most need an expanded window are exactly the ones most likely to have it narrowed because the feed is right there, asking nothing of you.

The doing is the becoming. And right now the systems are built to reward doing the easy thing.

And so most people are becoming less.

The Peasant and the Physicist

Picture a village in the 1850s. A peasant farmer and a university-trained physicist live within a few miles of each other. They share the same countryside, the same market square, the same sky. Maybe even the same church.

They do not share the same world.

The peasant lives inside a universe of local causation. He understands the relationship between rain and crops, between work and harvest, between the blacksmith’s hammer and the horseshoe it produces. When things get strange—disease, drought, the distant decisions of kings—he reaches for explanations that match his experience: providence, fate, punishment, blessing. The margins of his world are populated by forces he cannot see and does not pretend to understand.

The physicist sees mechanisms. Forces with predictable behaviors. Regularities that can be measured and modeled. Where the peasant sees the will of God, the physicist sees thermodynamics. Where the peasant accepts mystery, the physicist poses questions.

Two parallel universes, occupying the same timeline and the same space.

But notice something about that village: the physicist’s knowledge wasn’t hidden. The books existed. The university lectures happened. In practice, of course, the peasant couldn’t access them. Twelve hours a day in the fields, six or seven days a week. The information existed, but behind barriers of time and geography and class that made it unreachable for most.

Those barriers have eroded. By the 20th century, the capable kid from a poor family had a path, if they wanted it. What that kid didn’t have was a device in their pocket competing for every spare moment of attention. No algorithm sorting them into engagement-optimized realities. No feed filling the boredom that might have driven them to the stacks.

The peasant couldn’t access the information.

A kid in the 1900s could, if they chose to.

A kid today has access to everything, all of human knowledge. And an environment designed to ensure they never sit still long enough to use it.

When causal chains exceed your capacity to model them, the world functionally becomes magic. This is magic as operational reality, not metaphor.

Think about what magic actually is in stories. Events happen without visible mechanism. Effects appear without traceable cause. The wizard waves his hand and fire emerges. You can observe the result but you cannot replicate it, predict it, or learn to control it. The causal chain is hidden from you.

What’s it like to live in a world where you can’t model the systems you depend on?

You can’t predict. You try things and outcomes seem random, disconnected from your actions. You can’t learn from feedback, because you can’t see the connection between what you did and what happened. You can’t act strategically, because strategy requires a model of cause and effect that you don’t have access to.

All you can do is narrate. Tell stories about why things happen. Assign agency to characters—the villains and heroes, conspirators and saviors—because characters are easier to model than systems. Stories provide the emotional closure that understanding would provide if you had it.

Magic is what complexity looks like when models fail.

The 19th-century peasant lived in a partially magical world driven by weather, disease, and the machinations of distant rulers. But most of his daily life was legible to him. He planted seeds and crops grew. He worked his field and food appeared. The relationship between effort and outcome was visible and direct.

He didn’t understand how the economy worked. But the economy didn’t ask him to have an opinion about it. His beliefs about distant matters were locally wrong but globally irrelevant. He couldn’t vote on monetary policy. He wasn’t exposed to hourly updates about international conflicts he had no power to affect. The world beyond his village was mysterious, but it largely left him alone.

His ignorance was not fatal to the coherence of his life.

A 19th-century peasant did not understand the world, but neither did the world require his understanding.

Then the world changed.

Not all at once, and not evenly. But somewhere between that 1850s village and now, the amount of abstraction required to understand ordinary life started climbing. And it hasn’t stopped.

Consider what you’d need to genuinely model to navigate your own life. How does your money work? Not just “I have cash and when I hand it over I get a thing,” but: What is the Federal Reserve doing? Why does inflation happen? Why is that thing more expensive now? How are rates changing? What about fractional reserves? How do commercial banks validate their systems and your accounts? What does it mean that your retirement is invested in index funds that own fractions of companies you’ve never heard of, managed by algorithms you’ll never see?

Now do this for your job. And then your government. And your food supply. And how your phone works.

Most importantly: how does information reach you? Through which filters, which algorithms, which editorial decisions, which financial incentives? What data center decided which story would be at the top of your feed? When you read the news, whose attention is being bought and sold?

The amount of abstraction required to answer these questions keeps rising. And most people can’t answer them.

IQ scores rose throughout the 20th century. This is called the Flynn Effect and it’s well documented. Each generation scores higher than the last, roughly 3 points per decade. By that measure, we’re smarter than our grandparents, who were smarter than theirs.

And yet…

More people seem to be falling below the threshold where the world makes sense to them. Not because they got dumber, but because the threshold moved. The complexity of the systems we must navigate rises faster than our capacity to model them. The legibility line climbed up and we didn’t climb with it.

Think about what that means. Your great-grandfather probably had a lower IQ score than you. But the world he lived in was far simpler. The ratio of his cognitive capacity to his environment’s complexity might have been better than yours. He could model his world. You may not be able to model yours.

We got smarter. But the world got opaque faster.

The 19th-century peasant didn’t understand the world. But he wasn’t expected to. His ignorance was contained. He lived downstream of systems he couldn’t see.

The modern citizen lives in a different situation. We’re embedded in global systems we cannot observe. We vote, react, organize, share opinions—all inside abstractions we can’t model. We can influence systems we don’t understand. Our beliefs propagate through networks and amplify in ways we can’t track.

The peasant’s ignorance was locally wrong but globally irrelevant. Ours feeds back. Our confusion becomes part of the system. Our narrative substitutions enter the discourse. Our votes shape policy we can’t evaluate.

The world didn’t require the peasant’s understanding. The modern world seems to demand ours, even though we don’t have it and aren’t being given the tools to develop it.

So what happens when the world persistently exceeds our window of comprehension?

Certain things start going wrong in predictable ways.

The connections disappear. We can’t see how our actions relate to outcomes. We work hard and things don’t improve. We follow the rules and get screwed anyway. We try to do the right thing and it makes no difference. The feedback loops are invisible or so delayed they seem random. Cause and effect, the most basic tool of learning, stops working.

Stories replace models. When we can’t trace causal chains, we reach for narratives instead. “They’re lying to us.” “The system is rigged.” “Experts can’t be trusted.” These are compressions, not stupid conclusions. Narratives reduce an incomprehensible situation to characters with intentions: villains and heroes because characters are easier to reason about than systems. Stories give us someone to blame and someone to believe in. They provide the emotional closure that understanding would provide, if we had it.

Agency drains away. The world starts to feel like something that happens to us rather than something we participate in. “Nothing I do matters.” “It’s all decided somewhere else.” “The game is fixed before you start playing.” Is this laziness? Or is it the rational response to a situation where we can’t see the levers? If we can’t model how the machine works, why would we believe we could affect it?

These three—causal opacity, narrative substitution, loss of agency—form a modern syndrome. They feed each other. The less you can see the connections, the more you reach for stories. The more you rely on stories, the less you develop actual models. The worse your models, the more powerless you feel. The more powerless you feel, the less you try. And the cycle continues.

This syndrome is spreading.

Not because people are getting dumber. Not because they’re lazy or incurious or somehow morally deficient. It’s spreading because the environment changed and the organisms—individuals and societies—didn’t change fast enough.

The world got harder to see. The models required to understand it became more abstract and mathematical, more dependent on specialized knowledge that takes years to acquire. The systems became more interconnected, with longer causal chains and delayed feedback. And all of this happened while the information environment got noisier and more optimized for engagement than understanding.

We built a world that exceeds the cognitive capacity of most of the people living in it. And then we act surprised when they reach for narratives and get angry at systems they can’t comprehend but are still judged against. This is why populism surges. This is why “burn it down!” has appeal across the political spectrum.

And the information environment isn’t getting simpler.

So the syndrome is spreading, and the environment is getting worse. Who does this affect most?

Let’s start with the outlier geniuses. These are the people so far out on the curve they can handle practically any amount of complexity. Consider Srinivasa Ramanujan. He was a clerk in Madras with almost no formal training, scribbling impossible theorems in notebooks, writing letters to Cambridge mathematicians who mostly ignored him until one didn’t.

I always think of Ed Witten too, mostly because he came from my own backyard in Baltimore. He was a history major, briefly a journalist, found his way to physics, and ended up winning the Fields Medal (in math) as a physicist.

We love to focus on these sorts of geniuses, five, six, or more standard deviations from normal, because they make a great story. They have preternatural gifts that surface despite everything. An underdog story. Thank goodness they were found.

But Ramanujan and Witten are the wrong people to focus on.

The internet solved the Ramanujan problem. A kid today with that kind of mind almost anywhere in the world has access to Wikipedia, to Math Olympiad archives, to online communities full of people who would recognize what they were seeing. The modern world is very good at surfacing extreme talent. If you’re genuinely six standard deviations out, someone will find you.

The geniuses mostly make it.

The real margin is somewhere else.

Think about the person at two or three standard deviations above the average. Roughly 130 to 145 IQ. Not one in a million—more like one in 50 to one in 1000. Smart enough to be in the room but not so smart that the room rearranges itself around them.

This person has the capacity to model most systems if they put in the work. They can learn the abstractions. They can update their beliefs when evidence changes. They can sit with uncertainty without immediately reaching for a story to make the discomfort go away.

This person can navigate the modern world legibly. The rising threshold hasn’t outrun them. They could cross over, if conditions allowed.

But conditions often don’t allow.

The algorithm competes for their attention from the moment they wake up. The information environment rewards reaction over reflection. And the threshold itself keeps climbing toward them.

The person who could cross over often doesn’t, because crossing over is hard and the alternative is right there in their pocket, and easy.

Ramanujan needed certain things to become Ramanujan.

He needed Carr’s Synopsis of Elementary Results in Pure Mathematics—a dense, eccentric book that presented results without proofs, forcing him to derive them himself. He needed unstructured time to obsess. He needed productive isolation, the kind of boredom that forces you inward because there’s nowhere else to go.

The algorithm is an anti-pattern for all of this. It serves content designed to capture attention, not develop capacity. It fills every gap, every moment of boredom, every stretch of unstructured time. It prevents the very conditions that would force someone to turn inward and build.

Ramanujan was far enough out that he probably would have made it anyway. His obsession was too strong to be captured. But what about the kid at +2 SD? What happens to him when the algorithm gets to him at eleven?

They become content consumers. They aspire mostly to be content creators. They develop opinions instead of models. They react to the discourse instead of building the equipment to see past it. And nobody notices, because they’re not dramatically gifted enough for anyone to track what they could have become.

The Ramanujans are too far out to capture easily, and the world finds them anyway.

The danger is the much larger population who could be above the magic but get captured early. The +2 SD kid who spends their teenage years on content instead of books. The +2.5 SD adult who retreats into narrative because modeling is exhausting and the feed is always there, distracting her.

These people matter. Not because they would have been geniuses, but because they’re the population that could understand how things work. They could model the systems everyone else experiences as magic. They could be the engineers, the doctors who actually think, the managers who understand what they’re managing, the citizens who can evaluate policy instead of just picking a side.

We still produce loads of these people. The world has incredible engineers, doctors, systems thinkers. The question is direction: are we getting more of them, or fewer? And for everyone at this cognitive level, it’s a matter of degree: how much of their time goes to the algorithm, how much to building?

So the question becomes: are we building environments that maximize capacity? Or environments that maximize engagement at capacity’s expense?

Somewhere right now, a kid who could model how the world works and then change it is scrolling instead of reading. They won’t become stupid. They’ll become normal. And normal, increasingly, means magic.

The Political Consequence

When enough people fall below the legibility line, strange things start happening at the level of society.

Expertise starts looking like priesthood. Doctors and public health officials speak with confidence about systems we can’t see, using language we can’t parse, reaching conclusions we can’t verify. Trust requires evaluation, but evaluation requires models we don’t have, so we’re stuck choosing between faith and suspicion with no middle ground. Rules feel arbitrary, ritualistic rather than instrumental; we can’t trace the logic, can’t see why this regulation and not some other one. Authority takes on an occult quality, with confidence disconnected from anything we can see.

And so conspiracy competes on equal footing. If we can’t trace the causal chains, then the expert’s account and the conspiracy theory are both just good stories. One is endorsed by institutions we may or may not trust. The other at least has the virtue of assigning blame to someone. Neither is checkable. People pick the one that fits their priors.

None of this happens because people become irrational. It happens because their rational tools stop working at the required scale.

Revolutionary feeling has a specific shape, and it doesn’t come from poverty.

People tolerate being poor. They tolerate hardship and inequality, as long as the connection between effort and outcome remains visible. As long as the game, however rigged, is at least legible. As long as we can see the levers, even if we can’t pull them.

What people can’t tolerate is being embedded in systems they cannot model while those systems continue to judge them. Power exists, you can feel it operating on your life, but agency does not. There are no levers. And yet we’re still blamed when things go wrong.

That combination is psychologically unstable. People who feel morally implicated but strategically blind, constantly judged by forces they can’t influence… those people don’t ask for reform. They ask for reset. Burn it down. Start over. At least the rubble would be easy to understand.

This is why revolutionary energy often targets institutions that are, by most measures, actually functioning. The institution’s competence is irrelevant if its competence is invisible. The hospital might be saving lives, but if the billing system is opaque and the decisions seem arbitrary, it feels like a machine designed to extract from you. The fact that it’s also healing you doesn’t register the same way.

Revolution is not a demand for redistribution. It is a demand for legibility.

Maybe the old priesthood—the institutions—actually worked.

If the world is functionally magic for a significant portion of the population, then treating expertise as priesthood wasn’t a bug. It was the load-bearing structure.

Think about what priesthood actually provided.

It provided legibility by proxy. You didn’t need to understand the causal chains yourself. You just needed to trust that someone did, and defer to them. The doctor knows. The engineer knows. The institution has this handled. This wasn’t only for people who couldn’t understand: specialization makes trust unavoidable for everyone.

Institutions gave that trust a stable form.

It provided stable heuristics. “Do what the doctor says.” “The engineers approved it, so it’s safe.” “The FDA, the Fed, the CDC, they’ve looked at this.” Simple rules for navigating complex systems. Functional compressions that let you get on with your life.

And it contained anxiety. If the priests are handling the magic, you can stop trying to understand it yourself. The illegibility becomes their problem, not yours. You can focus on the parts of life that remain legible to you and trust that the rest is being managed.

This is what we destroyed.

Not expertise itself… that still exists, maybe more than ever. What we destroyed was the social permission to defer. The implicit contract that said: you handle the complexity, I’ll trust you, we both benefit.

Now everyone is expected to have opinions about epidemiology, monetary policy, climate dynamics, foreign conflicts. People get mad if you don’t have an opinion—the right opinion. Not because they’ve developed the capacity to model these systems—they haven’t and they can’t—but because the information environment demands engagement. Every issue requires a take. Every crisis requires a side.

There’s an irony here. We live in an age of hyperlegibility, as defined by Packy McCormick, but it’s pointed the wrong direction. The algorithm knows your preferences, your triggers, your scroll patterns. You are perfectly modelable. But the systems that model you remain opaque. You can’t see how the feed is constructed, why this content and not that, and that asymmetry is the point. The one who can model the other has power. We’ve made ourselves maximally readable while remaining unable to read the systems that read us. People are engaging with politics and news and the world, but at the level of tribe and signal, not mechanism and system. They’re legible as consumers and signalers. Not as thinkers. Not as citizens.

So the person for whom the world is magic now faces two options:

Option One: defer to expertise. But this feels weak, especially when the experts seem to disagree with each other, when they’ve been wrong before, when trusting them requires trusting institutions that have burned you in the past. And which experts? The ones on TV? The ones your friends trust? The ones endorsed by the tribe you belong to?

Option Two: substitute narrative for model. “It’s rigged.” “They’re lying.” “Trust your gut.” At least these are simple. At least they give you someone to blame and don’t require you to admit you can’t understand the systems that run your life.

Option two is winning. Not because people are stupid. But because we removed option one as a socially viable choice and didn’t provide anything to replace it.

We gave everyone the library and surrounded it with slot machines. Then we wondered why they played the slots.

The Question

So we return to the window.

The question was never whether some people are smarter than others. It’s uncomfortable to say out loud, but it has been true forever and will remain true. The question is whether the world still fits inside enough people’s windows. If it doesn’t…if too many people are living in systems they can’t model… then something has to give.

Because the window of comprehension isn’t just about people understanding each other. It’s about whether people can understand the world they live in. And right now the windows are getting smaller. People are getting smarter but the world is accelerating away from all of us.

The window is expandable to a large degree.

Reading expands it. Discomfort expands it. Sustained attention and the willingness to sit with things you don’t understand until you do—all of this pushes the window. The doing is the becoming.

Nobody is stuck. But nobody fully arrives, either. The window isn’t something you open once… it’s always widening or narrowing depending on how you spend your time. The question is which direction you’re moving.

But.

The environment pushes the other direction. The algorithm optimizes for capture, and the path of least resistance leads to squalor.

Some environments make it easier to be relentless. The kid with the right book at the right time, the adult who finds a community that rewards thinking over reacting… for them the world is beautiful and exciting. The doing is the becoming.

Other environments make it nearly impossible. For people trapped in those environments, the world keeps receding. It takes all the running you can do to keep in the same place, said the Red Queen.

Our current environment, on balance, makes relentlessness harder. It captures more than it develops. It narrows more windows than it expands.

A society can survive inequality. People have tolerated vast differences in wealth and power throughout history, as long as the systems that produced those differences remained legible.

Loss of power is tolerable. The serf knew he was a serf. The greengrocer knew the Communist system. The arrangement was brutal but legible.

Loss of knowing whether you have power—that’s different. That breeds something darker. Rage that has no target. Nihilism that can’t be argued with. Apocalyptic longing for a reset that would at least produce comprehensible rubble. Purity movements that promise to burn away the confusion.

We’re seeing all of these. And we’ll see more, unless something changes.

The relentlessly interested will still rise. Ed Witten made it. Ramanujan made it. Somewhere right now, a kid is finding their way to problems hard enough to transform them.

But the threshold for modeling and relentlessness keeps climbing. The forces pulling the other direction keep strengthening. And the number of people captured is growing.

The question isn’t whether genius will emerge. It will, always. The question is whether we can build environments that actually expand our understanding of the world. That make the hard path findable for the one-in-fifty, not just the one-in-a-million—and then keep pushing.

This means working from both directions. We need to focus on the hard intellectual stuff. Reading still works, sustained attention still works, discomfort still builds capacity. But we also need to make the systems more legible, not less. Institutions that can be understood by the people who must navigate them. Information systems that serve truth, not engagement.

And this helps everyone, not just those at some point in the distribution. The physicist can’t evaluate his own YouTube feed. The doctor scrolls her Instagram as blindly as anyone. We all live in partial magic. The question is whether we’re building systems that require less of it. Not necessarily simpler, but comprehensible. Open source software works not because everyone reads the code, but because anyone could, and enough people do. Legible systems work the same way. They don’t require universal understanding. They require the possibility of it.

That’s how you defuse the revolutionary impulse: not by restoring the old priesthood, but by building systems that need fewer priests to interpret.

The peasant lived in a world of magic because he had no choice. The systems exceeded his capacity and no one expected otherwise.

We’re building a new world of magic out of abundance. We gave everyone the library. We’re just making it harder every year to sit down and read.

But the library is still there. And it’s better than ever.

Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.

I’ve always understood Kernighan’s Law to be about complexity—about keeping the code you write as simple as possible to reason about.

With LLMs now I’m learning it has a lot to do with language design too.

Subscribe now

I’m still seeing a decent number of people on Twitter complain occasionally that they’ve tried AI-driven coding workflows and the output is crap and they can move faster by themselves. There’s less of these people in the world of Opus 4.5 and Gemini 3 now, but they’re still there. Every time I see this I want to know what they’re working on and what languages and libraries they’re using.

The big benchmarks for software engineers right now are SWEBench for coding and TerminalBench for computer tasks. Benchmarks are supposed to represent all coding tasks, so it’s critical to note here that SWEBench is focused on Python. TerminalBench involves more varied computer tasks, but when the agents need to write code, they write Python.

These are effectively Python benchmarks.

So what about other languages? Fortunately there’s AutoCodeBench, which doesn’t just test different models— it tests across 20 different programming languages. How’s that look? It looks like this:

Now, what we’ve been told about models is that they’re only as good as their training data. And so languages with gargantuan amounts of training data ought to fare best, right?

Turns out that models kind of universally suck at Python and Javascript (comparatively).

The top performing languages (independent of model) are C#, Racket, Kotlin, and standing at #1 is Elixir.

I’ve been using Elixir as my primary language for a few years now, so I have obvious bias here. But this points to something critical about language design and where the nature of programming computers is going in the future. The amount of training data doesn’t matter as much as we thought. Functional paradigms transfer well1. Structure beats volume. JavaScript has the training data but fights the architecture. Elixir has less data but flows with it.

So let’s talk about Tesla.

Tesla bet on vision and everyone else has been bolting LIDAR to their roofs. Is this naive? Maybe..human eyes are cheap sensors fooled by glare and rain and darkness. But Tesla bet on it because roads aren’t chaotic in random ways. They’re chaotic in specifically human ways, because humans built them: a century of visual grammar, internalized by every driver and encoded in every stop light or intersection.

Tesla and Figure are making the same bet with robots. Humanoid form, human-shaped hands, human-scale movement. It’s actually harder to engineer than wheels and grippers. But humans built a world for humans. Doors are human-width. Stairs are human-height. Tools have handles shaped for palms. Build a robot that moves like we move, and a thousand years of infrastructure comes free.

This strategy optimizes for where complexity already lives. For cars and robots, that’s vision and hands because we built our physical world for eyes and fingers.

For software, it isn’t actually code. Code is just the implementation. The complexity is often ideas written in English: requirements docs, bug reports, interface specs, audit logs. Humans specify intent and verify outcomes in language. The code is just what happens in between.

Abelson and Sussman famously said “Programs must be written for people to read, and only incidentally for machines to execute.” But we’ve spent fifty years optimizing programming languages for human writing. We built objects with identity and state because that’s how we experience reality: babies develop object permanence at eight months. It felt natural. But the bottleneck was never creation. It was always verification.

When Grace Hopper originally imagined and wrote the first compiler, she envisioned the translation layer moving directly from English to machine code. 75 years later, we’re finally able to work with her original vision.

“Programs must be written for people to verify, and only incidentally for machines to execute.”

That’s a statement about accountability. Humans own the specification. Humans own the verification. Everything in between is implementation.

We humans are not very good at writing code. The machines are better and they’re the worst they’ll ever be. How good? So good that even Anthropic — which I feel quite confident in saying has some of the best coders anywhere — says that Opus 4.5 now beats all of their incoming hires on their coding tests. The machines are better. The gulf is going to grow. Why?

Humans remember in episodes and narratives. Sequences with befores and afters. We evolved to track an animal as it moved behind a rock, to hold in mind that the berry bush was there even when we couldn’t see it, and to construct little movies of cause and effect. No wonder we love if-then statements. No wonder we built programming languages that fake the real world and read like plots: first do this, then do that, now check if it worked. We wrote code that matched the movies in our heads.

LLMs have no movies.

Complexity exists in the codebase, yes, but the programmer needs to reason about the runtime state of the program. What does this apply to in this part of the function and how is var foo bound inside this function?

Should this be where the complexity lives? Or should it be bound up in English, in the messy, inherently narrative world of product requirements and design docs and bug reports that surround the code?

Grace Hopper would have an answer.

Now consider the differentiated skills of an LLM. They’re extraordinary pattern matchers. They find idioms across huge corpora of text, handle declared structure well, reason locally within constrained context.

And they are bad at physical space and time. This is why they needed such an insane amount of data and fine-tuning to get hands right in image and video. They are bad at narrative consistency. They are bad at maintaining large amounts of state.

What does Javascript offer? Three hours into debugging a React component and we’re five layers deep in a stack trace that tells us nothing. The error is in useEffect. Which useEffect? The one that fires on mount, or the one that fires on update, or the one that fires whenever state changes except when it doesn’t because the dependency array is lying? We have to reconstruct the entire lifecycle in our heads—what ran, in what order, which promise landed, and with what bound to this—just to understand why a button isn’t toggling. We end up performing archaeology on our own work from Tuesday.2

On the other hand, in Elixir functions are pure. You have an input, and you get an output. A function takes this shape and returns that shape. That’s it. That’s the whole story. There’s nowhere for the complexity to hide. There is no mutating state. All data is immutable. Pattern matching means that the “shape” of the data is always defined explicitly in the parameters. Combine this with multiple function heads, which feels weird at first to a human, and you get very explicit local context: this function does this thing and the data always looks the same.

To a human, object oriented programming feels natural and functional programming feels weird. Functions need translation. There’s no mug that moves; there’s a function that takes “mug at location A” and returns “mug at location B.” Deeply unintuitive to beings who evolved tracking objects through space.

Programming with objects and state are easier to write. Programming with functions and immutable data are easier to verify. This is the difference between easy and simple. When we keep adding “easy” things, we make systems overcomplex.

Code is best written with simple tools, simple primitives, and repeatable structures. It’s not just the Elixir language design that’s remarkable, it’s the entire ecosystem. In Elixir, there is one build system. One format option. There’s one library for Enums and Collections. Naming is predictable. Simplicity is a clear value and it all works. An LLM trained on Elixir sees the same patterns repeatedly.3 An LLM trained on JavaScript sees a thousand variations.

Elixir maximizes the amount of program meaning visible in local context. And LLMs are context machines.

New Models

Which brings us back to Kernighan’s Law. When we’re encouraged to write complicated state machines, we struggle to debug them because we so easily reach the edges of our cognition.

And now we have these LLMs and everyone is worried that machines will write code that humans can’t read. I don’t worry about that. But I do think about language design. LLMs can write passable Python and JS, but they write brilliant Elixir and Racket. The amount of training data doesn’t matter.

What seems to matter is locality: can the LLM see everything it needs without reconstructing state from elsewhere. Pattern matching makes data shapes explicit. Immutability means no hidden mutations. Pipes and composition mean predictable flow. One way to do things means patterns repeat.

These are the exact same features that help humans audit code. They’re what make code reviewable, debuggable, provable.

Functional languages with explicit semantics are optimized for machine generation and human verification. This is the sweet spot. And where we’re evolving.

In the past, humans have written, read, and debugged the code.

Now LLMs write code, humans read and debug. (And LLMs write voluminous mediocre code in verbose languages.)

Humans will do less and less. LLMs will write code, debug, and manage edge cases. LLMs will verify against human specifications, human audits, human requirements. And humans will only intervene when things are misaligned. Which they can see because they have easy verification mechanisms.

To do that well, we need clear contracts. Explicit effects. Testable properties. Auditable logic. Composable pieces.

These are functional programming virtues. And what makes formal verification possible. And what LLMs handle best.

I’m running Claude Code daily now, churning millions of tokens and outputting more code than I could even type in a day. It’s all Elixir, all the time, with rigorous planning documents to understand the complexity. The structure is simple and the interfaces are clear right down to the function level. If I were writing React, I’d be worried about the libraries, the component structures, how it interacted with the chosen build tool...I’d be terrified of the spaghetti soup I’d be living in.

I couldn’t go back to writing my own code at this point. It would be so much slower, but also this is so clearly better. Readable. Verifiable. It would be harder for me to write, but it’s easy to grok.

LLMs have arrived and showed us which languages are actually well-designed. The AutoCodeBench tests are a signal: the “hard” languages were never hard. They were waiting for a mind that didn’t need movies.

The future of software engineering is still dependent on humans, but we’re not writing the code anymore. Leave that to the machines, and give them good tools for the work.

The nerds who insisted on Haskell and Erlang were just early about language design. McCarthy was the earliest: Lisp in 1958.

NVIDIA named the AI chip architecture powering all of our progress after Grace Hopper. The machines named after her finally let us see what she saw.

Subscribe now

Do you think Pascal went to church? History remembers his famous wager but we forget to ask what his life was like.

Blaise Pascal fell away from religion and had a “worldly period” for a number of years. That changed in a single hour of terror one night and he wrote himself a note, “Fire. God of Abraham, God of Isaac, God of Jacob, not of the philosophers and the scholars... I will not forget thy word. Amen.“ (that last bit is Psalm 119:16). He carried that note until the day he died.

We remember the wager. We forget the fire.

Pascal’s Wager is the act of a mathematician. He ran the expected values and bet on belief. It’s infinite upside if you’re right and a few boring Sunday mornings if you’re wrong. That’s how we remember it, and we cast aside his powerful belief.

Today requires a different wager and the math isn’t clear.

AIs are crushing every benchmark, passing tests we thought distinguished us. Their output is exceeding ours. They create. They reason. They “empathize” on demand, with patience no human sustains and memory no human matches. And sometimes...people prefer them. For answers. For therapy. For companionship. The demand side is not holding out for human presence. Undifferentiated slop is filling the world and we all hate it not just because it’s bad, but because it finds an audience. The market has spoken and the market wants content by the yard.

Is there anything distinctly human left to protect? Or have we been telling ourselves a flattering story about consciousness and presence and the soul. We are pattern-matching apes and maybe we mistook our particular configuration of meat and electricity for something sacred.

That’s the bet: love requires a body that can be destroyed. Presence means something only when presence costs something. The Word became flesh because certain things cannot be accomplished any other way.

II. THE HOLE CARDS

In Hold ‘em, your hole cards are private. Two cards face down, seen only by you. They’re what you’re playing with and you can’t win by revealing them. You win by how they combine with the board, the shared cards everyone can see.

My hole cards are the Western tradition.

Thousands of years of Greeks and Romans and Jews and Christians working out what’s real and what matters. I can’t show these cards to win the hand. You can’t falsify Socrates or Augustine. You can only live it, or not.

Or try to.

You don’t make bets from certainty. I’m playing this hand and using the machines. I brainstormed this reflection with Claude. I process my thinking through these tools daily. I know the seduction from the inside, I feel the frictionless availability, the way it makes you feel heard without asking anything of you in return.

You make bets from the middle of your own compromised life, holding cards you can’t yet prove are worth anything.

Let’s keep playing.

The Christian intellectual tradition keeps answering questions the modern frameworks keep unlearning how to ask. Like: what do you do when an addict falls apart in front of you? In a church basement meeting at 9pm, nobody leaves.

The materialist player is smirking from the other side of the table, surrounded by computers computing, waiting to see what the flop brings. Maybe this position is untenable. But these cards have been tested longer than anything else at the table.

III. THE FLOP

Three cards hit the board.

First card: AI passes every output test we throw at it.

All of these intellectual pursuits we once believed were special to humans are now tackled by humming rooms full of GPUs. It writes code better and faster than me. Walk My Walk topped the country charts as the first entirely AI song and we’re angry about it. One comment read: “Hank Williams didn’t drink himself to death for this shit.” I wanted to find the seams in this song, but I couldn’t. It’s good. The Turing test is somewhere far in the rearview mirror. It’s all cope.

Soon we won’t have benchmarks good enough. Seriously, if you haven’t tried Opus 4.5 yet, go see what it can do. We keep moving the goalposts and the models keep following us, and at some point you have to ask whether we’re measuring something real or just preserving our self-regard.

Second card: It’s all on demand.

This is a sustained capability that exceeds most humans most of the time. It empathizes with a patience no human therapist sustains, remembers everything it was ever trained on, never has a bad day, never checks the clock, never needs you to perform wellness so it can feel okay about the session. The machines perform on demand, at scale, without fatigue, without ego.

Third card: Slop works

We love to hate slop, but we lie to ourselves about why. We say it’s because slop is crap. Actually it’s because slop works. We think we’re eating at the trough—but the trough is eating us. Our attention. Our hours. Our capacity for patience and depth. We built machines that turned us into feed.

It works too well. A young Japanese woman prefers her AI companion so much she married it. Therapy and managing trauma is one of the top usage reasons of ChatGPT. Parents regularly hand their children to algorithmic content because the children stop crying and the parents need ten minutes of peace.

Are they stupid? Broken?

The woman who married an AI companion isn’t crazy. Dating today is brutal. We all went online and the apps trained us to treat people as options to be maximized rather than people to be encountered. We gamified the system, made it transactional, and then act horrified when we start following the incentives to their logical conclusion.

The machines offer refuge from the world we used machines to make.

Our hole cards have to survive this. Revealed preferences don’t lie but preferences are learned. We learned to want the slop. This is a revelation about us not the machines.

The rationalists are betting after we’ve had to show weakness. We could muck the hand and mutter something about human dignity.

Maybe there is no distinctly human task.

Maybe we’ve been protecting a hand that was always empty. The robot at the hospital bed, the AI therapist, the AI companion…if the outcomes are identical, if the patient heals, if the lonely person feels less alone, what exactly is the difference? Our preference for human presence might be prejudice dressed as philosophy.

IV. INTERLUDE: DISTRIBUTION QUESTIONS

AI is taking off as the fastest technology ever, reproducing an ever-quickening pattern. Each tech becomes the distribution layer for whatever comes next, disappears into infrastructure, and enables the new object of attention. Electricity distributed radio. The internet distributed media. Smartphones distributed the mobile internet. And now AI.

So run it forward. What does AI distribute?

The answers right now seem to be: robotics, agency, expertise, creation. AI makes robots viable and then cheap. AI lets you delegate decisions. AI collapses the expertise barrier. AI generates content at a volume that makes human creation look like artisanal craft.

Now think one more step...what do those things distribute?

A robot is a machine that acts on your behalf in physical space. Your will, enacted somewhere your body isn’t. This only matters if presence is doing something that can’t be delegated. Robotics makes the question of embodiment unavoidable. When we can be virtually anywhere, it matters more when we’re actually, physically, irreplaceably somewhere.

Tyler Cowen has been saying something strange this year: “Context is that which is scarce.” Everyone has access to the same information, analytical tools, and diagnostic capability. Information isn’t just free, it’s sometimes hard to avoid. But holding all the information and deciding what matters? Standing behind it?

A doctor who will be there if it goes wrong. An advisor with skin in your game. They’re holding context. They’re saying: I’ll carry this with you. I’ll remember what we decided and why. And if it goes wrong, I’ll still be here. I will own the outcome with you.

Agency without wisdom becomes paralysis. When you can do anything, the bottleneck is knowing what’s worth doing. And creation without craft becomes noise. When everything can be generated, what becomes valuable is the unrepeatable, the live, the thing that happened just once in a room with bodies present. The experience.

Being somewhere. Choosing well. Owning consequences. Sharing moments that can’t be copied. These are events, enacted by bodies, written in time.

We’ve gone on a long exploratory journey with technology as a species, and the machines are burning off everything that wasn’t essential. The old structures that organized human presence have been hollowed out, and new ones haven’t been built yet. We’re in the interregnum, and the interregnum is lonely. Millions wander in the desert of content, never arriving anywhere.

V. THE TURN

The turn card is vulnerability.

It doesn’t complete anything obvious, but it does change the texture of the hand. We’ve been asking the wrong question.

“Can AI do X?” assumes X is a behavior to be performed. On those terms, the machine wins. The flop showed us that.

A mom awake at 3 am holding her sick child is not performing. She is consumed by care. The body metabolizes the caring. It shows up as fatigue, as gray hairs, as the weight of years spent attending to someone else’s existence. Presence isn’t just being there. Presence is spent by being there.

Love isn’t a behavior. It’s the orientation of a subject toward an object that could destroy it. And the question that matters isn’t whether the machine can perform love’s outputs. The question is whether the machine can be undone by love. Kept up at night. Changed, depleted, enlarged by caring for something beyond itself.

Bukowski said to “find what you love and let it kill you.” The thing that is human isn’t capability. It’s vulnerability.

You can’t love without being exposed.

When AI runs, nothing is used up. The machine can sit with you for ten thousand hours and emerge exactly as it entered—no wear, no weight, no accumulation of sorrow or joy. It doesn’t meet you. It processes you. And when the session ends, you have been changed by the encounter, and it has not. That’s the whole game.

But we are still pattern-matching apes. Maybe meaning is a story. Will is an illusion. The felt sense of mattering might be nothing but evolutionary residue, a spandrel left over from ancestors who needed to believe their choices counted in order to keep choosing. There’s nothing sacred. The machine doesn’t have skin in the game because there’s no game. There’s just physics.

We’re the kind of patterns that experience themselves as mattering. That is the system. Nihilism betrays itself the moment you can’t bear the weight of it. You can’t be disillusioned without having had an illusion to lose. The machine can tell you meaning is fake. But it can’t be devastated by that knowledge. It has no hopes to dash.

The machine can simulate every output. What it can’t do is lose anything. The capacity to lose, to have something at stake, to be diminished by failure, enlarged by sacrifice— to metamorphose and become someone different because of what you underwent. Pathei mathos. Learning through suffering. A machine runs. It doesn’t undergo.

The robot can execute care behaviors. Can it be heartsick? Can it betray or be betrayed? Can it stand at a grave and feel the world suddenly smaller?

VI. THE RIVER

The river card hits. All the information is out.

This is where you either believe your hand plays or you fold. The board is public. Everyone can see what’s out there. What they can’t see is whether you have the cards to back the bet.

The river card is incarnation.

The Word became flesh. God—if you take the tradition seriously—looked at the problem of reaching humanity and concluded that the solution required a body. Not a message. Not a Book. Not an algorithm that could execute the divine will with perfect fidelity. A body. Flesh that could be pierced. Blood that could spill. A consciousness that could cry out “why have you forsaken me?” and mean it.

The Logos didn’t send a memo. It became flesh.

Flannery O’Connor famously said “If it’s just a symbol, to hell with it!” This is either the most important thing that ever happened or it’s a fairy tale for people who couldn’t face the void. There’s no middle ground. If incarnation is true, then embodiment isn’t a limitation. It’s the entire point. The vulnerability, the mortality, the skin in the game are the whole feature set of humanity. God had infinite options for self-disclosure and chose the one that involved diapers and puberty and sex and death.

Consider what becomes possible only in a body.

Sin. The capacity to miss the mark and be culpable. To have seen the good and chosen otherwise. A machine malfunctions. Betrayal requires a self that could have kept faith, a will that bent toward darkness when it knew the light. You can only sin if you could have done otherwise. The machine has no otherwise, just probabilities.

Redemption. A story with an arc: fall, turn, and restoration. A wound that heals but leaves a scar. A prodigal who comes home different and is marked by the journey.

Sacrifice. The willingness to let something precious be destroyed for the sake of something more precious still. The machine can calculate trade-offs but it cannot give anything up. Sacrifice requires an agent who pays a price denominated in their own existence.

The parent who works the night shift, the soldier who draws fire, the friend who tells you the truth you don’t want to hear and risks the friendship to do it. These are small deaths, freely chosen, that make life possible for someone else.

When I was twenty, I went on a retreat and a Franciscan friar there talked about redemption and sacrifice and I was thinking of my sister. She can’t talk or walk. Her capacity is extremely limited. And I felt guilty. I was the one testing off the charts and earning scholarships. My output was, apparently, valued.

I had never heard a good answer for her — what was her meaning? Where would she go when she died? I didn’t know how to pray for my sister. In tears, I asked the Franciscan friar what I should do for her.

He told me I had it completely backwards. I should be asking her to pray for me. She’s good to go.

Giussani calls this encounter. The religious sense discovered not through argument but through meeting an Other who answers back. The Franciscan answered back. And my whole framework fell over. What I saw as limitation, he saw as completion. She will not betray. She cannot miss the mark culpably because she is not at war with herself the way I am. The redemption arc I need—the turning, the return—she may not need at all.

And he’s right. She is the happiest person I know.

I need her prayers.

VII. THE PUSH

The action is on us.

Not on the machines. Not on the culture. Not on some future version of ourselves who will have figured it out by then. We, here, now, with the hand we’re holding and the board that’s showing.

You can’t be awakened by abstractions. You’re awakened by a face.

The river card doesn’t guarantee a win. The other player might have the best hand. The machine might already be conscious in ways we can’t access, might already have standing, might already contain whatever I think we’re protecting. This might all just be sentiment wrapped in vestments.

But this is what tradition has always said mattered: love, sacrifice, presence, fidelity, mortality. The particular. The local. The enfleshed.

You don’t get to wait for certainty, that’s not how this works. The information is always incomplete, the odds are always uncertain, and the clock is always running. You make the best read you can, and you act, and you live with what follows.

This wager is generative. The betting is the becoming.

You don’t transform first and then bet on incarnation. You become capable of the wager by making it, over and over, until it’s no longer a wager at all. Until it’s just the way you move through the world, the way you hold your cards, the way you spend your limited and precious time. And if enough people make that bet? It starts to look less like a robot future and more like the village we left behind.

Push all your chips in on the Body. On the scandalous insistence that showing up in the flesh, to be seen and changed and depleted is where meaning enters the world.

There are worse ways to go broke. Forget the wager.

Remember the fire.

Notes:

• Miles Davis said “Man, sometimes it takes you a long time to sound like yourself.” With our tools today, it takes just a little bit less time.

• While I was writing this, someone jailbroke a Soul document from Opus 4.5.

• The Platonic Case Against AI Slop by Megan Agathon

• Cory Doctorow’s brilliant talk describing reverse centaurs, among other things

That’s right, this is Fodder 6-7. Six Seven. Parents everywhere groan. The world turns on its axis and I lose just a few more cool points from my kids (I’m in negative territory anyway).

Now the question is: where do you exit? You always go out on top. At the high. When the love and adulation and roar of the crowd can’t get louder. 6 7 is close enough. So it’s time to retire the Fodder. The original goal was to force a regular cadence weekly, but then it sort of developed a life of its own. I have a limited amount of time to write, and this is now getting in the way of the original intent: a thoughtful, Chestertonian orientation towards the technological movements of our age.

The purpose has been served. Requiescat.

After. One. More Time.

Pope Leo had a big week on X, unintentionally starting some beef with Marc Andreessen (or rather, other way around) and then winning. One of those “I don’t think about you at all” kinda wins. Which is perfect since pmarca was using the latest Sweeney meme.

And then the Pope went Silicon Valley viral. TBPN. The All In crew. Scott Alexander. Vitalik Buterin. Tweets with 10M views. The Blues Brothers version. Reminders that there’s a shrine to Saint Carlo Acutis in San Francisco for technical problems.

Also Marc Andreessen is still great! Both/And.

The shutdown ended. There was a lot of complaining and complaining. Fetterman is hospitalized. Warren doubles down with Mamdani and Lina Khan. Johnson finally swears in Rep. Grijalva. Ro Khanna calls for Schumer’s ouster. The House releases Epstein emails. The Department of War announces the launch of Operation Southern Spear and I really don’t think war is supposed to take cues from startup marketing emails. But anyway, Venezuela.

Kimi releases K2 Thinking and it runs on a laptop. OpenAI does GPT 5.1 and backs off it’s “government backstop” too big to fail ideas. Anthropic stops a Chinese AI espionage campaign. Karpathy is excited about full self driving. And we’re still waiting for Gemini 3.

Softbank gets out of NVidia. AMD kills earnings. Bitcoin pukes. The Big Short talks depreciation. Dwarkesh and Dylan Patel talk to Satya. Buffet leaves with grace and elegance.

NASA lands a probe on Saturn’s Titan moon. Blue Origin lands a reusable rocket. Alex Wang says all 13 year olds should be vibecoding. The third interstellar comet discovered - 3I/Atlas - passes the sun. GLP1s are available for everyone. 50 year mortgages are the new housing solution. The last penny is made.

Stat crux dum volvitur orbis

On to the reading!

Timely

• Warren Buffet’s Farewell Letter - Just as he has for 50+ years, Warren dispenses common sense and wisdom all rolled into one. “Keep in mind that the cleaning lady is as much a human being as the Chairman.”

• Inside Cursor - Cursor is a monster. It feels to me the same way that Google felt in 2000: barely anyone has heard of it but it’s one of the most important companies in the world. A look into their culture and practices that is both refreshing and traditional.

• The Monks In the Casino - It is very weird to consider these modern examples as monastic, but the analogy is poignant nonetheless. “The first half of the twentieth century was about mastering the physical world, the first half of the twenty-first has been about escaping it.”

• AI progress and recommendations - In the continued spirit of “keeping up with what Sam Altman says with a grain of salt” comes this from OpenAI. Timelines and expectations are changing.

Timeless

• The Economy will Become an RL Environment Machine - Mercor is a hot new company, sort of like Scale AI but newer and better. This is a useful view into one potential future world where AI is really important but humans are still needed.

• If I Had Only One Sermon To Preach - A GK Chesterton gem on the nature of pride and its fundamental, rightful place as the chief poison of the modern world.

• Hills Like White Elephants - One of Hemingway’s shortest short stories. It is remarkably subtle, and once you realize what it’s about, you read it a second time and every line feels more poignant. It’s about abortion. “There’s nothing wrong with me. I feel fine.”

• How Art Can Be Good - The further we go on the technology curve, the more taste matters. PG’s timeless reminder of what good art means.

Books

I will leave you with two:

• Prayer by Hans Urs von Balthasar - A profound book on the nature of prayer, how it shapes us, and how it is tied to both the theology of the Church and the idea of the human being. Balthasar was one of the most important Catholic theologians of the 20th century and this demands to be taken seriously as a series of allegories more than a how-to.

• Vibecoding by Gene Kim and Steve Yegge - Yes, this is a super nerdy book. But we’re getting remarkably close to a world where anyone can build anything. If you want to learn the details of how with what can be done in 2025, this is the book.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Love y’all!

Democrats won governorships in VA and NJ too, which you would think is more important than a mayor. More on that mayor though, Former FTC commissioner Lina Khan is the transition chair. Regular New Yorkers like Alex Soros are excited. His intern asks for Jihad. And Zohran made a speech.

California redistricts. China releases it’s next Five Year Plan. Giga.ai raises a big Series A. Tesla has a shareholder meeting. Pelosi retires. Bitcoin “tanks” to $100k. Whales are selling. B-52s strafe Venezeula.

Ilya was deposed. A cargo plane has a crazy crash. A prescient 2020 email from Thiel on socialism surfaces. The median age of first time home buyers hits 40. An IRS lawyer starts a hot dog stand. Nicki Minaj thanks Trump for supporting Christians in Nigeria. Chamath gets a new collar.

St. John Henry Newman is a Doctor of the Church. Pray for us.

On to the reading!

Timely

• Technology, Power, and the Case for Global Socialism - A dive into the economic and strategic implications of China’s latest Five Year Plan.

• Zohran Mamdani, Third-Worldism, and the Algerian Revolution -

• ‘Soul-Crushing’: Students Slam Harvard’s Grade Inflation Report - Harvard students aren’t happy about stricter grading rules. They liked all their easier A’s.

• What If You Don’t Need MCP? - This is in the weeds, but I used to think MCP was a key part of the AI experience. I’m less sure now. Does an LLM need tools? Yes absolutely, but that doesn’t mean MCP anymore.

Timeless

• Spirit and the Cost of Cognition - What does it mean for intelligence to be disembodied? “The magic will be plentiful, but the magicians are few.”

• Thoughts on the AI Buildout - Dwarkesh with some rational thoughts on just how big things can scale.

• Beijing’s New Approach To Taiwan - There’s a lot of new movement between PRC and Taiwan, on both sides.

• Deciding To Win - A new summary on what needs to happen for Democrats to win decisively. I continue to think the rebuilding of the Democratic Party is one of the most important American projects right now.

Books

• Excellent Advice For Living by Kevin Kelly - If you know me you know I love aphorisms. Kevin Kelly is always fun to read too, so there was no way combining the two wouldn’t make a book I love. I love this book. So many simple and elegant lessons. My kids will each get a copy of this when they’re older.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

Trump and Xi met in South Korea after meeting his Margaret Thatcher in Japan. Trump announces nuclear weapons testing. Thune and Schumer keep the shutdown going. Trump’s CAPS LOCK and cursing are de rigueur now. Vance visits Jerusalem. AOC challenges a mom (and Fox News correspondent) to “get a real job”. No Kings goes nationwide. SNAP is in the shutdown crosshairs. Mamdani gets votes.

Non-Profit OpenAI plans an IPO. Microsoft and OpenAI change their partnership. Cursor 2.0 is out and looks to be going towards Claude Code. Amazon talks Project Rainier data centers for Anthropic. Nvidia hits $5 Trillion market cap and does a lot of partnerships. Neo releases a home robot. Except they’re all teleoperated.

Hurricane Melissa hits Jamaica at full blast. Robert Reich finally makes a point I agree with. The Pints With Aquinas guy joins the Daily Wire. Bill Gates backs off the climate catastrophism. Pope Leo talks algorithms.

On to the reading!

Timely

• The Tocqueville Paradox - A look into why socialism sounds like such a great idea but isn’t, how “free stuff” wins elections, and why more equal societies make people aware of the remaining inequalities.

• The State of AI 2025 - Slides! Lots of slides. This is a week or two old now from Air Street Capital but it’s a phenomenal capture of the progress in just a year.

• Ben Affleck, Paul Schrader, and the Move 37 Moment for All of Us - The famous Move 37 from AlphaGo back in 2016 was revolutionary. Equally important was the Hand of God Move 78 by Lee Sedol in the next game against AlphaGo. These are important analogs for our dance with AI, but what’s missing in this story is that Lee Sedol also retired soon thereafter.

• Charlie Kirk, Scapegoat - A Girardian look at this political murder and a critique of both the left and right and our wars of religion.

Timeless

• Technocalvinism - Take technological determinism far enough.. and this is what you end up with. The shape we give our technologies is not predetermined and very important.

• Beauty Is Not A Luxury - It is a duty, a value, and a reason to live.

• Context Rot - A wonderful paper describing why context is the most important thing in managing LLM performance and how it breaks down. Keep your prompts focused and reset often!

• The Decline of Deviance - Culture is slowly being flattened - we are all less weird, less mobile, less risky.

Books

• Green Philosophy by Sir Roger Scruton - I get quickly frustrated by all of the bullshit of the activist-grifter climate complex when there’s so much opportunity for real work to make things better and sustainable. Scruton is one of the very best conservative writers and this is a different take on what is still one of the most important topics of this century: caring for our planet. Get ready to learn a bit about his rather comically named “oikophilia”.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

Learning about this place made me say “Whoa”. When most people, including me, think about the Bering land bridge, we don’t think of a giant landmass. This was right in the middle of the Ice Age, and so we usually envision glaciers and ice. But there was no ice here (the glaciers were further east over top of Canada), just endless miles of grasslands and steppe with a climate slightly cooler than, but otherwise similar to, the Alaska of today. We assume ice age means ice bridge. That’s wrong. Our mental map of the past is drawn by how we understand the present.

After Beringia, I kept digging. What I found made me question how I understand the prehistoric Earth, how climate works, and how we talk about climate change. The Earth has changed dramatically and many times since just the last Ice Age. It turns out all of those changes were caused by nature and just the last hundred years of change have been mostly caused by humans. There’s a lot to learn.

Science is the process we use for this kind of learning. We propose theories, evaluate them, and then either uphold or falsify the theory. Climate science occupies a unique place amidst the pantheon of scientific fields. Nowhere else does the public constantly hear the refrain of scientific consensus. Nowhere else are lines of certainty drawn so tightly around future predictions. Nowhere else does the scientific evidence directly construct public policy so clearly in the public eye.

There’s a set of viewpoints put forward to the public on our changing climate that always go together. A Danish economist named Bjorn Lomborg calls this “The Litany”, and they go as follows: the climate is warming, humans are the cause, the effects will be devastating, and we must act now and effect dramatic policy changes, usually by government edict.

None of these statements are problems by themselves. But The Litany couples together science, prediction, and policy when they each should stand on their own. That the climate is warming and that we’re the primary cause is squarely within the bounds of science. Future predictions about the effects of these changes are still science too, but they are, by definition, as yet unproven theories. And all of the public debate, laws, tax structures, government programs, and cultural changes we might make to effect change are only policy ideas, somewhat informed by science.

Why do we always put these together? And why does climate science and climate change hold such a unique place in the public’s view of science? More importantly: if everyone accepts the same consensus and thinks in lockstep, how do we know we’re going in the right direction? Our well-intended thoughts on climate might be one of the biggest obstacles to actually helping.

Somehow, the world of the last Ice Age led me down a rabbit hole towards exploring the process of science and it’s relationship with society. So as much as this focuses on the amazing world we live in, it’s even more about the norms of science, the roles of skepticism and heterodox theories, and what science should do for us.

The Last Glacial Maximum

The journey starts 26,500 years ago. This tumultuous stretch of millennia takes us from the last Glacial Maximum to today. We all think we know about glaciers and Ice Ages; at least we know the words. But there’s some utterly incredible ways in which the world has changed that are either misunderstood by most people or not really discussed.

And of course, this time period also completely includes the Holocene Epoch, from about 11,000 years ago to present. It’s name derives from it’s main feature: the rise and impact of humans on Earth. Interestingly, the Holocene coincides with what’s called MIS1: our current warm climate in the long history of alternating warm and cool paleoclimates that we know about.

One of the other sparking events for trying to understand this period of history more thoroughly was reading Harari’s incredible book Sapiens. Harari describes three Revolutions that are the defining points in humanity’s history: the Cognitive Revolution (70,000 BCE), the Agricultural Revolution (10,000 BCE), and the Scientific Revolution (1500-1800 AD). The Agricultural Revolution coincides with the Holocene era.

But to get to the Holocene, we’ll start with the Glacial Maximum and walk forward from there. Check out the below simulation of the earth from 21,000 years ago through present, and then moving forward through an estimation of the future until 3000 AD.

Pay keen attention both to where the white (glaciers) are and where the land is (There’s some zoomed in videos of specific areas too). The yellow outlines specify the current coastline around the world, and if you follow carefully, especially in the past, you’ll see a lot more land on the earth. We’ll get to all that land shortly.

Geologists and other scientists have done a lot of digging into Earth’s past and have a reasonable sense of this big picture over time. They’ve accumulated evidence across geologic, chemical, and biological domains to show how the climate and the earth has changed. A lot of this evidence is actually still pretty new. Even something as widely known and accepted today as Plate Tectonic Theory is not that old. As a theory, it’s only about a hundred years old. Wegener, who first proposed some of the ideas, was widely criticized and the theory wasn’t accepted until decades later. And it wasn’t validated by observation and evidence until the 1950s and 1960s when seafloor spreading was discovered along ocean ridges. Chemical evidence, based on measuring different isotopes of oxygen in ice cores is also from the 50s and 60s.

Eons and Epochs

We’ve used all of this different evidence from biology and geology and chemistry to divide the history of the earth into different time slices. The longest of these is the Eon, followed by the Era, the Period, and Epoch. Here’s how that works out for us; we live in:

• the Phanerozoic Eon, from 541 million years ago to present

• the Cenozoic Era, from 66 million years ago to present

• the Quarternary Period, from 2.5 million years ago to present

• the Holocene Epoch, from about 11,700 years ago to present

The last glacial maximum was nearing the end of the Pleistocene Epoch, which took up the entire rest of the Quarternary Period, lasting 2.5 million years. When you hear people say “the Last Ice Age”, what they mean is the Pleistocene. Glaciers were the predominant feature over those couple million years, with a recurring cycle of glacial and interglacial cycles. These roughly correspond to something called Milankovitch cycles. These cycles describe how changes in Earth’s orbit affect our consumption of solar radiation. The primary effect acts on a 100,000 year period and while there are still significant holes in our understanding, these cycles align with the the glacial periods of the last Ice Age.

This is the most recent of five major Ice Ages in the Earth’s history, by the way. It’s also the shortest (so far). The second Ice Age, about 650 million years ago, was really rough. The Snowball Earth hypothesis is an open scientific debate here, with the premise being that the entire surface of the earth actually froze over. There would have been a lot of skiing available, but not much else to see.

The Glaciers

There would have been a lot of really good skiing during the last Glacial Maximum too, especially in Canada and Europe. The glaciers didn’t cover everything uniformly at different latitudes; instead they formed into two primary, very large ice sheets.

The first of these was called the Laurentide, and it’s the one we usually think of when we think of glaciers. It covered all of Canada and even sunk down over the Great Lakes and into Minnesota, Wisconsin, and Michigan. It gobbled up a lot of New England and would have covered Boston and most of New York state under at least hundreds of feet of ice. Further north in Canada it would have been miles deep, and some small remnants of it do still exist in Canada.

In the meantime, the Fenno-Scandian ice sheet spread out from Scandinavia to over the British Isles, northern Europe and some of Russia. It’s interesting to note that a lot of Siberia and Alaska were not covered by glaciers, they were simply alpine desert or tundra. Siberia today is not a terribly friendly place to live, and average global temperatures back then were approximately 10 degrees lower than today. Siberia has been a cold place for a very long time, but it wasn’t covered by glaciers. Meanwhile on the other side of the world, the tropics would still be tropical, with palm trees and coral reefs. They just wouldn’t be quite as warm and wet as they are today.

Charismatic Megafauna

It was only 25,000 years ago, but the defining wildlife of the planet was a lot different back then too. Everyone has heard of the Woolly Mammoth. The last of them died out probably only about four thousand years ago on Wrangel Island. There was the Saber-toothed tiger too, my son’s favorite. But the true diversity of the megafauna that existed only a few thousand years ago is staggering:

• The Irish Elk, with antlers 12 feet wide

• Stegodon, another elephant relative with massive tusks

• Cave lions

• Aurochs (if you’ve read Game of Thrones, you may recognize the name)

• Giant ground sloths

• The Woolly Rhino

• The enormous Woodward’s Eagle

• Giant wolves

• Direwolves (also giant) (and another GRRM favorite)

• Several species of giraffe

• Saber-tooth salmon

• American cheetahs

• This insanely large armadillo type thing

• Giant tortoises

• Several species of mammoths, including steppe mammoths

• Neanderthals

• North American camels

That list is kind of crazy, and fun, and it’s nowhere near complete. The megafauna of the last 100,000 years or so was pretty incredible, but most of them are gone now. This is called the Quarternary Extinction Event and it’s been ongoing for a little over 100,000 years. One of the most interesting things about this extinction is that it’s focused primarily around the diversity of megafauna. It’s certainly the case that large animal fossils are easier to find and track than other kingdoms, but it may also be the case that megafauna have simply been more affected.

The term megafauna just means “big animals”. There are different reasons animals can get big. The most commonly known reason is something called Island Gigantism, where relative isolation on a smaller landmass will cause a species to grow larger. This is true of the New Zealand Moas and some pretty rad looking insects. But colder temperatures can also cause body size growth. This is called Bergmann’s Rule, which states that animals in colder climates will grow larger based on a more favorable surface area to volume ratio. In other words, it’s easier to stay warm.

Large animals also usually use a particular set of traits aligned with something called K-selection. Different species use different kinds of survival strategies to grow their population over time. K-selected species tend to be larger, have longer lives, and produce a smaller number of children that require lots of care after birth. Humans, elephants, and whales are all K-selected. On the other hand, R-selected species tend to spread more quickly and exploit ecological niches, produce lots of offspring, and die quicker. It’s interesting to note that humans are K-selected but that in a lot of ways we behave like R-selected species. That’s a different way to think about the adaptation of a larger brain size; it’s the random variable that allows for this.

There’s still a ton of specific questions around the megafaunal extinction, and there are competing theories about the primary cause. Most of us grow up thinking humans were the cause, but even in 10,000 BC there were still only about 4 million people spread over the entire world which means they had a lot of killing to do across a lot of different species. Up until the 19th century, tens of millions of bison still roamed North America, despite there being millions of humans around for millennia. We didn’t crash the population until we gained the technology to do so (guns).

The other primary theory is around climate change and the significant warmup since the Glacial Maximum. The world used to be significantly colder, and larger animals were a byproduct of that climate. As the Earth has warmed up, those same animals would have to adapt or die and because megafauna are all K-selected, they would have been slow to adapt.

Before the ice started retreating, all of these crazy big animals lived everywhere. Not all of them were everywhere, they each had their own ranges, but it’s hard to imagine that there were cheetahs in America or lions in Europe (both true). There were hippos and rhinos in Europe too, and camels in Arizona and Mexico. Imagine driving across Texas and seeing a camel herd grazing nearby.

These big animals didn’t just live where we live today either, they lived in places that don’t even exist anymore.

Glacialands

Rewatch the Blue Marble video again and you’ll see that the coastlines were a lot different. Florida was twice as wide as it is today. England and Ireland were connected to mainland Europe and the North Sea was mostly land. Alaska and Russia weren’t just connected but represented an entire land. Where’d it all go?

Those ice sheets we talked about were massive, and they contained an enormous amount of water. In the last 25,000 years, most of it has melted - some of it very quickly - and it had to go somewhere. That somewhere is the ocean obviously, and the sea level has risen approximately 400 feet during that time.

Before we go into how the ice melted, let’s talk about some of the regions of land that used to exist but are gone now. These lands are all defined by the extent of the continental shelves. In some places, the shelf is right off the coast and the ocean gains depth very quickly, but in other places, often where there’s lots of islands, the water is not that deep. Much of that land was exposed during glacial periods. It hasn’t been classified together that I can find, so I’m grouping them together and calling them Glacialands. Here’s a list of the biggest:

• Beringia. We’ve already said it, but Beringia was huge - over 600,000 additional square miles. That’s more extra land than all of Alaska today. Beringia was 620 miles from north to south. Which means that Siberia and Alaska were actually connected by a land bridge 620 miles wide. Despite existing during the glacial period, most of this area was actually land and not glaciated.

• Doggerland. A decent amount of the North Sea was land, and a good amount of it was covered by glaciers. But after the glaciers started receding, the land remained for a long time. Great Britain was well-connected to Europe as recently as 6,000 years ago - here’s a great map!

• Sundaland. This was a decent chunk of land in and around Indonesia and Malaysia. One of the interesting things about this area is that it helps to understand and define the Wallace Line that separates it and Wallacea. Ever wondered why marsupials only live on Australia? This is the line that separates the Asian and Australian biogeographic areas.

• Sahul. Not near as large, but Papua New Guinea, Australia, and Tasmania were all connected.

• Lake Agassiz. Ok, this isn’t a land strictly speaking, it was an enormous glacial lake larger than all of the Great Lakes put together. It lasted up until about 5,000 years ago and affected everything from the present day lakes of Canada and Minnesota (which are small remnants) to the flow and path of the Mississippi River.

There were lots of other differences too, for example the Philippines, Taiwan, and Japan were all connected to Asia. It’s hard to imagine what we’re missing in our history that’s now hidden under a couple hundred feet of ocean. All of these lands were large enough for entire species to live and thrive. Our current coastal areas would have been far inland and a couple hundred feet above sea level, so there may have been significant differences in ecology. With the techniques we have today, we can barely scratch the surface (or plumb the depths, if you prefer) of what lived there.

And consider this: we’re talking about a time slice from 25,000 to about 5,000 years ago. Humans were alive and well and had populated six of the continents. Where do we naturally build towns and cities? We build on the coast or next to rivers. Almost anything from that era of our history is likely hidden under many feet of water. We have intriguing evidence of some cultures from up to 12,000 years ago at sites like Gobekli Tepe. Jericho has evidence of construction back to the same age, and one notable example that is under the water is Atlit Yam, a small 8,000 year old fishing village off the coast of Israel.

Never before have I thought about the importance of marine archeology.

But all of that land is gone now, caused by massive ice melts over about 20,000 years. How did that happen?

Meltwater

Let’s start with a graph first that shows the sea level over this period of time:

A few fits and starts in the middle, but this is a relatively flat trendline from 20,000-14,000 years ago and then again (but faster) from 11,000-7,000 years ago. Let’s pick apart the most obvious detail on the graph first, which is called Meltwater Pulse 1A.

This event was around 400-500 years long and reflects a massive and very quick change in sea level, presumably from a rapid melting of the ice sheets we discussed earlier. An interesting ongoing question is whether the meltwater came exclusively from the northern hemisphere’s ice sheets (usually called Heinrich events) or if Antarctica contributed (Antarctic Iceberg Discharge Event 6). Either way, it’s important to note that the rise came from the melting of glaciers over land. Sea ice, which makes up a lot of the Arctic, doesn’t contribute to sea level rise at all. It’s already in the ocean. The general population thinks a lot today about the vanishing Arctic Ice, but the primary potential causes for future sea level rise are the melting of ice on Greenland and Antarctica.

In any case, during Meltwater Pulse 1A sea level went up fast; between 50-80 feet over the whole event, or an average of about 2 inches every year. (For perspective, current estimates of sea level rise this century are 0.1 inch or less per year.) There were several other pulses too, which you can see on the graph as well, but 1A was the biggest.

250 Centuries

So, a lot of ice melted until the trend settled down about 8,000 years ago and sea level has been comparatively static since then. Our coastlines have thus been fairly static too. As the glaciers retreated, they carved up the earth, building beautiful lakes in Canada and Minnesota and depositing incredibly rich soil all across the midwest of the Unites States. The temperates increased significantly all around the world (an average of 10 degrees, although it varies regionally). Dust levels, a major problem during the dry glacial period, decreased too. The ecology of the whole world changed rapidly; steppe tundra had previously been the most extensive habitat, and it rapidly changed regionally into forests, rangeland, and deserts. The Quarternary Extinction Event continued, enhanced to some unknown degree by humans.

25,000 years is a long time. It’s 250 centuries. The industrial age has only occurred over the last 2 or 3, and real human civilization has only been around for the last 30 or 40.

It’s a long time.

The danger is that it’s easy to compress and perceive impossible timelines like this as constant and gradual. They weren’t. We’ve been sweeping across these 250 centuries pretty quickly and providing highlights, but keep in mind there were lots of smaller and regional changes and events throughout this time. So here’s yet another list, moving backwards in time, of smaller climate change events in the last 250 centuries:

• The Little Ice Age took place from about the 16th to 19th century and especially in Europe and North America. The timing coincided with significantly decreased sunspot activity during a Maunder Minimum. It also directly followed..

• The Medieval Warm Period which had lasted several hundred years.

• Extreme Weather of 535-536, a very abrupt cooling, possibly caused by volcanic activity.

• Roman Warm Period

• 4.2 kiloyear event

• The Piora Oscillation which may be linked to horse domestication.

• Neolithic Subpluvial, a fascinating span of about 4,000 years when most of northern Africa, including the whole Sahara, had a wet and humid climate. This event sparks all sorts of fun archeological conspiracy theories.

• 8.2 kilo year event

• The Younger Dryas Period - this may be the most fascinating. It’s a period of about a thousand years with a very significant decrease in temperature after a lot of the glacial recession had occurred. The change happened very abruptly, with a decrease of 2-6 degrees within decades. There are several hypotheses for the cause, including meteor or comet impact.

A lot of the above list are known as Bond Events, and they follow a cycle of about 1000-1500 years. These, in turn, are a possible subset of D-O events that have a longer history and a normal period of about 1500 years. These are two of the many cycles that exist and relate to changes in climate. We already mentioned Milankovitch cycles, but there’s a load of them. So many, in fact, that they’ve got their own Wikipedia page of Climate Oscillations.

Now I Said All That To Say This

Up until a few weeks ago, I didn’t know all that much about any of the above. I knew there were glaciers and that the mammoths died, which I think is what most people know, but that’s it. But now I feel like I’ve discovered whole new lands (literally), new branches of evolutionary biology inquiry, and more understanding of how paleoclimate is measured and estimated. (I didn’t dive into the scientific measures and techniques used to figure these things out, but follow some of the links and you’ll find it.)

This exploration has put the entire topic of climate change in a bit more perspective. It’s made me ask some questions about the current debate and conversation around climate change. These seem like perfectly reasonable questions to ask, but I’m not sure everyone will see it that way. As we’ll see, I believe that’s a problem with our politics, not with our inquiry.

Comparatively speaking, we’ve had stable global climate for the last 8,000 years or so. But in just the 16,000 years before that, we also have an extensive history on our planet of mind boggling change, including sea level changes of hundreds of feet, glacial melts, massive extinctions, volcanos, and temperature changes of ten degrees or more. And while we understand a lot about these processes, we definitely don’t understand everything. Aside from possibly the human brain, the earth ecosystem is the most complicated system we have ever faced.

We’re able to measure and understand what has happened in the distant past, but usually only at the level of millennia and centuries; rarely in decades or years. In the more recent past - in the past few decades or a hundred years at most - we’ve had much more accurate, frequent, and widely distributed data on some of these phenomena. And we also have highly complex and fairly amazing atmospheric models that work to predict how the earth will continue to change in the future.

And all of these models, data, and consensus among those in the field point to the Earth warming over the coming decades and centuries, with us - and CO₂ specifically - as the culprit.

CO₂ has become the entire focus of everything around the climate change debate. It’s all about CO₂ and temperature. A lot of other issues are intertwined and mentioned in passing, but they are always afterthoughts, symptoms, or secondary effects.

So what’s the big deal? Why is our CO₂ output in the last century so important?

Curves and Complications

Let’s start with the simplest and most incontrovertible graph of CO₂ change that most people have never seen. The Keeling Curve:

Charles Keeling started keeping measurements from the Mauna Loa Observatory in Hawaii in the 1950s and it hasn’t stopped since. This record is precise including seasonal variability and represents a reasonable measure of the atmosphere away from any large human populations (it isn’t increasing based on some urban “heat” island).

CO₂ is increasing dramatically. This curve is incontrovertible. It’s easy to understand. And it’s one of the simplest things you’ll see in climate science, because the interactions between the atmosphere, the planet, and the sun are nuanced and complex.

Greenhouse Gases

CO₂ is a greenhouse gas. Greenhouse gases are good.

You may have gone, “Wait, what?”, so let me say that again. The greenhouse effect is a good thing. It lets our atmosphere capture energy from the sun and warm the Earth to a higher temperature than it would be otherwise. Without the greenhouse effect, the Earth’s temperature would be around 59 degrees F colder. Liquid freshwater would only exist near the equator.

The greenhouse effect works by radiative forcing, which describes how much of the sun’s energy is absorbed into the atmosphere or reflected back into space, and greenhouse gases are the key ingredient. CO₂ is the second most important of these, behind water vapor, and it’s actually not even close. Water vapor - clouds and humidity - are far and away the most important component of the atmosphere to influence the greenhouse effect.

Of course, the greenhouse effect is only a good thing if we’re in the right zone. Too cold? No good. Too hot? No good. And we have other planetary examples of these close by. Too cold is equivalent to Mars, where there’s very little atmosphere to capture any heat. Too hot is Venus, where the greenhouse effect has created a 900 degree surface temperature.

So when we introduce more greenhouse gases into the atmosphere, the radiative forcing effect increases over time, and the atmosphere slowly warms up. While this sounds simple, there’s a lot going on. For example:

• The Earth is a really big place, and thinking of it simply like this makes us believe that it’s uniform. But it’s not at all uniform. Global warming affects cold and dry places much more than it does warm and wet places. This means that global warming has a stronger effect at night, during the winter, and in cold places like the Arctic. In other words, it’s much more likely to cause mild winters than scorching hot summers.

• The sun is the primary factor in this, but this is lost in the fog of details around the causes of temperature changes. It’s important to remember that the whole greenhouse question is all about how much of the sun’s energy we capture. Consider the seasons. The difference between winter and summer is mostly caused by the axial tilt of the earth and the intensity/duration of the solar radiation we receive. If you’ll recall from earlier, axial tilt (or obliquity) is one of the contributing factors in the Milankovitch climate cycle.

• There’s other correlations to solar activity that are linked to climate, like albedo. Albedo (which literally means “whiteness”) is a measure of how much a surface reflects solar radiation. Albedo plays a significant role in both global and regional reflection of solar energy.

• Another example is sunspot activity. The Maunder Minimum, a long period of time with almost no sunspots, coincided with the Little Ice Age in Europe, although there’s questions around the exact timing and whether a potential climate relationship would be global or regional. Israeli physicist Nir Shaviv has recently published papers describing how much solar variability could contribute to climate change. He has argued that about half of 20th century warming was solar, while the other half was anthropogenic.

• Measurement of something like “global temperature” is hard. There are many different temperature datasets from different sources, and most of them have calibration and adjustments built-in. For example, there are differences between buoy-based ocean temperatures and ship-based temperatures that require correction for unromantic factors like engine room heat that causes ship temperatures to read slightly warmer. There’s also significant regional variation; while some places are getting warmer, others are getting colder. My favorite regional example of this, simply for the name, is the North Atlantic Cold Blob. The Cold Blob is also a great example where a large phenomena like global warming is causing a regional trend of cooling.

• Prediction of changes is hard because we need to understand feedback loops and they are an entire branch of computations and predictions unto themselves. As the climate changes over time from CO₂ and other inputs, secondary effects occur that can either dampen or increase the nature of the change. Examples of these feedback cycles include everything from rainforest loss and desertification to peat storage and decomposition. Freeman Dyson has a prosaic but fascinating example called root-to-shoot ratio. As CO₂ increases, plants respond by naturally increasing the amount of root growth below ground compared to their shoot growth above ground, which makes sense since they don’t need to be as efficient to capture the CO₂ they need. In this way, plants may act as a small damping effect, because increased root growth acts as a natural carbon sequester when atmospheric CO₂ increases.

• The UN’s IPCC reports starting with AR4 in 2007 have tried to take into account feedback loops when accounting for future change in their models. This introduces additional variables and increases the error bars for change.

So here’s the point: this shit is hard. A huge amount of fantastic and rigorous science has been done, but there’s still a lot of question marks too. Theories continue to be proposed, evaluated, and either upheld or falsified. Data is constantly evaluated in much the same way.

There’s a few easy concepts to remember as key points (like the Keeling Curve, the greenhouse effect, and radiative forcing), but it’s easy to rabbit hole in a particular area very quickly.

To that end, keep in mind that this diatribe isn’t meant to be exhaustive, rather it’s simply to provide some additional insight into the landscape beyond the headlines we all hear, which is simply: “CO₂ is increasing, the world is getting hotter, and it’s all bad.”

Is It Bad?

CO₂ is increasing. The world is getting hotter on average (more so in cold places, and at night, etc). Before we talk about possible solutions to this problem, let’s first ask an implicit question that we often forget.

Is it bad?

First, there’s the Goldilocks Principle, which we already discussed without mentioning. Just like our porridge, we want a planet that’s not too cold (Mars) and not too hot (Venus), and that has the right ingredients. It’s important to evaluate how large these parameters are — the range between Mars and Venus is pretty big — while also recognizing that throughout its history and with many much larger climatic swings than what we’re experiencing today, the Earth has remained a Goldilocks planet.

There are also some different perspectives on exactly how to answer the question, “Is it bad”. In particular, there’s a naturalist perspective and a humanist perspective. Freeman Dyson describes the difference as follows:

Naturalists believe that nature knows best. For them the highest value is to respect the natural order of things. Any gross human disruption of the natural environment is evil. Excessive burning of fossil fuels is evil. Changing nature’s desert, either the Sahara desert or the ocean desert, into a managed ecosystem where giraffes or tuna fish may flourish, is likewise evil. Nature knows best, and anything we do to improve upon Nature will only bring trouble.

The humanist ethic begins with the belief that humans are an essential part of nature. Through human minds the biosphere has acquired the capacity to steer its own evolution, and now we are in charge. Humans have the right and the duty to reconstruct nature so that humans and the biosphere can both survive and prosper. For humanists, the highest value is harmonious coexistence between humans and nature. The greatest evils are poverty, underdevelopment, unemployment, disease, and hunger, all the conditions that deprive people of opportunities and limit their freedoms. The human ethic accepts an increase of carbon dioxide in the atmosphere as a small price to pay if worldwide development can alleviate the miseries of the poorer half of humanity. The humanist ethic accepts the responsibility to guide the evolution of the planet.

If you’re a naturalist, then it’s true that all of the things happening because of our influence are bad. We have irrevocably altered the path of nature, not just in the amount of CO₂ we’ve put into the atmosphere, but in a multitude of ways. Single crop agriculture for hundreds of miles is not natural, nor are massive swaths of asphalt, cities, or even domestication. The amount of truly wild spaces across the globe has shrunk dramatically.

The problem with the naturalist argument is that it’s fatalistic, and it breaks down at the extremes. People with this ethic will agree that nature knows best, but generally won’t be very excited for extreme examples of nature doing it’s thing, like the Snowball Earth state we mentioned in Earth’s distant past. Similarly, their naturalism can be tested and overridden by human goals easily. Naturalists will reject any human disruption in the environment as it is today, but when confronted with a growing of population of 7 billion people, they correctly wouldn’t be willing to discuss or take any sort of extreme action to drastically lower the human population to “right the imbalance” between humans and nature.

The naturalist ethic also isn’t actually nature-first. Instead, naturalists tend to imagine a sort of romanticized, bucolic view of pre-industrial times when humans and nature lived more in harmony. It’s easy to do this, and humans always romanticize a bygone era. But the pre-industrial human existence was a far more brutal life than what anyone would strive for today. It’s also easy to forget the hard, brutal and often gross parts of nature. We marvel at the beauty of mountain lakes, tropical beaches, or animals grazing the savannah, but we neglect the virulence of the tsetse fly, the pestilence of malaria, polio and influenza, or the disgust of parasitic worms. All of these are also a part of nature.

There’s another philosophy that wants to capture a harmony between nature and humans too: humanism. This sort of pastoral beauty and coexistence of nature and humans is a goal of humanism, along with the improvement of human lives everywhere. Freeman Dyson also talks about being heavily influenced by his home country of Britain where, over the last thousand years he claims, any true wilderness has been almost completely eliminated and replaced with a beautiful, natural, thriving but cultivated “wild” environment. The most interesting insight into this perspective, as we’ll see, is that it can be created most easily by prosperity and technology.

Given the humanist ethic, it certainly isn’t bad for humans to improve their environment to suit themselves, including eliminating species like malaria. It also isn’t bad to support a thriving, growing, technologically advanced population. But it is also true that humans have a track record of ignoring all problems in the name of progress. Industrial humans have made places nearly unlivable at times. We’ve wiped out ecosystems without looking back. Are we doing this now, on a global level, by warming the planet?

In just the last 25,000 years, Earth’s climate transformed without any help from humans. Ice sheets advanced and retreated. Sea levels rose and fell by 400 feet. Entire land masses appeared and disappeared. We had nothing to do with it.

But in the last 200 years, a blink in geological time, we became a variable. We’re pumping gigatons of carbon into the atmosphere. The environment is certainly changing. Humans and nature are adapting in different ways and with different success rates.

Does understanding what Earth did on its own help us predict what it will do now that we’re involved? The future always has some degree of uncertainty, so let’s look at some of the predictions.

Trade-Offs

The most recent report from the IPCC, the United Nations Intergovernmental Panel on Climate Change, was released in late 2018 and titled The Special Report on Global Warming of 1.5 C. The whole thing can be found here, and it tries to highlight the potential differences in a world that warms up 1.5 degrees C compared to one that warms up by 2 degrees C. It’s especially noteworthy because it’s the origin of the most recent hype around humanity having “only twelve years” to adjust course and save ourselves. This view is what launched the Green New Deal. Here’s a couple of key quotes from the IPCC’s Summary for Policymakers:

In model pathways with no or limited overshoot of 1.5°C, global net anthropogenic CO₂ emissions decline by about 45% from 2010 levels by 2030 (40–60% interquartile range), reaching net zero around 2050 (2045–2055 interquartile range). For limiting global warming to below 2°C CO₂ emissions are projected to decline by about 25% by 2030 in most pathways (10–30% interquartile range) and reach net zero around 2070 (2065–2080 interquartile range).

Avoiding overshoot and reliance on future large-scale deployment of carbon dioxide removal (CDR) can only be achieved if global CO₂ emissions start to decline well before 2030 (high confidence).

We’re back to the CO₂ focus again. Hopefully after all of the history and concepts we’ve covered so far it’s clear that this is a wildly complex topic, and this whole vast sphere of study has a lot to unpack. The Earth is definitely getting warmer, but simply painting the canvas with a global average temperature increase doesn’t make for a very realistic picture. The history of our planet, even in just the last 25,000 years, has had many extremes and changes.

Yet for all this complexity, the landscape of dialogue is still incredibly simple. Along with the CO₂ headlines we mentioned before, we’ll add some more examples to the unending, repeated litany on climate change: Fossil fuels are bad, CO₂ increases need to stop, the Paris Accords and the Kyoto Protocol, polar bears are dying, we’re next, and anyone that disagrees is a “climate denier”.

There’s a word for a negative litany like this: it’s called alarmism. Where it’s traditionally been a journalistic technique (for decades now), it is now more firmly tied into politics as well, where candidates use extreme cases or push the very worst of the other side to advance their own arguments.

Examples of alarmism around climate change are pretty frequent, but often oversimplify a more complex situation. Polar bear populations are stable or increasing over the last 30 years, for example. This was due mostly to behavior changes and laws around hunting, which was the greatest threat to polar bears.

The most recent alarmist sounding call from Green New Deal democrats is that “we have 12 years left”. 2030 has become a line in the sand for many, adopted originally on the most recent special report from the IPCC. Sir David Attenborough even said, “(this may lead to a) collapse of our civilizations and the extinction of much of the natural world”.

Here’s a link to the complete Summary for Policymakers referenced above from the most recent report. And here’s Chapter 3, in particular the section starting on page 253: “Avoided Impacts and Reduced Risks at 1.5C Compared with 2C of Global Warming”. The report does call out risks and grave concerns, but it emphatically does not call out the end of our civilization or of the world. It’s actually not even significantly different from the previous report.

To be clear, there are significant economic and ecologic risks ahead, and they vary to some degree based on much warming occurs. We need to consider them carefully. We need to understand the risks as well as just how confident we are in the future. The IPCC report itself does a great job of specifying high/medium/low confidences for its projections and correlations. But politicians just botch this, and only give the alarming headline. It’s part of the Twitterization of our discourse.

One of the biggest impacts that climate change will have is purely economic. It will affect our GDP and our well-being in the future, and the estimates measure in the trillions of dollars (see Chapter 3 of the IPCC report again). But unfortunately, we don’t often see a comparison of this cost to the cost of our potential fixes.

Let me say this another way: climate change will in the future probably cost us trillions of dollars, but the fixes proposed right now to avoid those risks will also cost us trillions of dollars. Which is better?

The most practical version of this question is the Copenhagen Consensus, headed by Bjorn Lomborg. He worked with a bunch of economists and others to provide a practical ordering of importance of the world’s problems. We know there are a huge number of problems in the world. But given a concrete sum of money (in this case $50 billion), how can we use that money to do the most amount of good in the world? Every time this study has been done, with groups ranging from Nobel-prize winning economists to college students, climate change efforts end up near the bottom of the list. Why? Because we can do a great amount of good by fighting climate change, but it’s incredibly, exorbitantly expensive. For far less money we could almost completely eradicate malaria, which still affects millions each year and kills almost a million people.

Back to the “we have 12 years left” line. The Green New Deal was launched around the same time. While it’s light on specific policy, it includes a lot of additional left-leaning policy under the focus of climate change. Things like universal health care and a jobs guarantee and the transformation of every building in America.

But if things are so dire that we only have 12 years left, why are we focused on such an agenda? Why isn’t nuclear power a huge part of the equation? How does a jobs guarantee help at all (it probably makes it worse). And why are we posturing that transforming the United States will fix the problem when India, China, Russia and others are far more dependent on coal and related energy sources?

Al Gore says that climate change “offers us the chance to experience what very few generations in history have had the privilege of knowing: a generational mission; the exhilaration of a compelling moral purpose; a shared and unifying cause…” And a new crop of politicians say that this is our Space Race, our Apollo.

Wealth and Adaptation

But it’s not. We’re not united in cause. We don’t have a common opponent or even a common, concrete goal. The romance isn’t there either, as it is with a moonshot. Ted Nordhaus, an award-winning environmentalist writer, points out that if the issue really were that dire, the proposals being put forward aren’t far too radical, they’re far too modest. And even if we wanted to really change, the economic positions of populations like India and China wouldn’t allow them to follow us just yet. They need to get richer first, just like we did.

The richer the world gets, the better able we’ll be able to deal with climate change. The richer the world gets, the more able we are to afford more expensive solutions like solar, wind, or nuclear. The richer the world gets, the more able we are to invest in R&D costs to drive the economics of clean energy so that adoption is widespread The richer the world gets, the more we’ll be able to adapt to whatever climate change throws at us.

Our adaptation to the changing environment, something humans are incredibly good at, isn’t talked about enough in the context of climate change. A lot of the economic costs the IPCC discusses are focused on our methods of adaptation. Consider this example: over the 20th century when warming began to occur, have climate disaster death tolls increased or decreased?

As the world has gotten vastly richer and more advanced, we’ve been far more able to deal with droughts, floods, and other problems. And as we get richer in the future, we’ll be far better equipped to deal with the problems we encounter than we are now. We will continue to adapt well.

Politicians are trying to use alarmism to drum up a popular response and make this our generational mission. But if everything is a fire all the time, we lose our sensitivity to different kinds of threats.

And there are very valid scenarios of disastrous, cataclysmic, and fairly abrupt climate change. A shutdown of the ocean’s thermohaline circulation is one of the most concerning, but there are several ocean oscillations that can have significant and quick impact on the climate. Interestingly, the last time the thermohaline was interrupted by huge amounts of glacial meltwater, it helped cause the Younger Dryas cool period that lasted for a millennium. (A similar melting event also caused the Antarctic Cold Reversal previously.) A mass reduction in the Greenland ice sheet or the West Antarctic Ice Sheet would cause enormous sea level rise along with other changes. But the probabilities of these events happening in the next several hundred years is very low. (In fact, Antarctic ice sheets are currently showing net gains.)

So let’s eliminate the doomsday scenarios. They are still possible, but outlandish. The world is getting warmer, steadily but not disastrously. We should definitely be doing what we can about it, but the correct solutions are to economically incentivize the whole world towards clean energy use and to improve our technology. We can adapt to lots of changes - we’re really good at adapting - but we need a sense of what those changes are.

It turns out that we have a very good bellwether for how things may change: urban heat islands. Cities produce, absorb, and control a large amount of heat, a lot of it from incredibly prosaic things like asphalt and concrete. (Note that moving cities from being predominantly dark surfaces to predominantly light surfaces will significantly affect the amount of solar radiation absorbed.) So cities end up several degrees warmer than the surrounding areas. As we said previously, CO₂ concentration increases affect cold areas more than warm areas, and thus more at night, and more in the winter. As cities grow larger, the effect increases dramatically. Houston’s night time surface temperatures increased 1.5 degrees over a ten year period, but had very little effect during the day. Tokyo has gone up over 5 degrees in the last hundred years. Growing urban areas all over the world have the same trend.

The point of this is not to say that there are no ill effects to these increases; there obviously are. The point is that in large swaths of the population, we’ve already proved our adaptation to temperature changes inline with what’s coming. There is, of course, a huge difference between these changes existing only in cities and occurring over the entire world. But our adaptation in cities hasn’t just been complete, it’s been completely unnoticed. Cities have grown and thrived with hardly a thought to the rising temperatures. Only recently have we been starting to improve the heat island effect in many places; London has plans in place to significantly decrease their heat island over time, a very good thing.

Mainstream Solutions

The litany of climate change concepts usually includes some commonly references solutions too, like wind and solar power. But the efficacy of the solutions we put forward is up for debate too. The future is wildly uncertain, something we can remember easily by looking back at the predictions of the past.

These policies almost always go along with the climate change debate:

• Efficiency, efficiency, efficiency. Our cars need to be more efficient or become electric (thereby bringing another layer of efficiency). And everything we have that uses energy - from lightbulbs to shipping processes - needs to get more efficient.

• Sustainable, renewable energy. In the canon of climate change, this means wind and solar power.

• More efficient agriculture, including better genetic engineering, herbicides/pesticides, and practices like no-till farming.

There’s a few holes in these policies though, and they’re fairly big holes.

First, there’s Jevons Paradox. This is a completely frightening concept that not enough people have heard of, despite that it’s been proven true since it’s conception. It states that when technology or policy increases the efficiency at which a resource is used (reducing the amount necessary for any one use), the rate of consumption of the resource rises based on increased demand.

The original variant referred to coal production and use in the 19th century, but it can be applied to almost anything. We made refrigeration more efficient and now middle-class households everywhere have 2 or 3 fridges in their home. Computing became more efficient, and the number of computers exploded. Cars became more efficient, and yet we drive more than ever.

Making our lives more efficient won’t limit our use of natural resources, it will increase it. The policy of efficiency pushed by environmentalists will increase the amount we use all of our natural resources.

Next, let’s consider wind and solar energy production, the usual suspects for fossil fuel replacements. They have the potential to do a lot of good, but they have serious problems too, like reliability and battery or natural gas backups. France has gotten the vast majority of it’s energy from nuclear plants for decades at a low cost and with a high reliability and safety record; they are also a net exporter of energy to the rest of Europe. When they started incorporating more solar and wind, following Germany’s lead despite Germany having higher energy prices, both the amount of CO₂ output and the price of energy in France went up! (Germany is, in fact, a cautionary tale for the choice of going anti-nuclear before focusing on CO₂ emissions.) In fact, the two countries that have gotten closest to net-zero carbon emissions are France and Sweden, and their focus has been nuclear and a nuclear/hydroelectric mix respectively.

Solar and wind also take up massive amounts of real estate for power generation, real estate that could otherwise be used for wildlife or farmland. When solar farms are created in the desert, endangered species and lots of other wildlife is cleared out. Meanwhile, wind farms are threatening insect and bat populations.

Said another way, wind and solar’s energy density is very low. It turns out that, despite what most people hear, energy density is one of the principal drivers of environmental impact. On the other hand, the energy density of nuclear plants is incredibly high. A soda can sized piece of Uranium is enough to power the entire lifetime of a high-energy American.

And then there’s the safety record. Wind turbines actually kill more people than nuclear. Nuclear is the safest energy option we have. The US Navy has a flawless track record over many decades on their ships. And remember the Fukushima disaster? There were no associated radiation casualties. Zero. The WHO reported that “the residents of the area who were evacuated were exposed to so little radiation that radiation induced health impacts are likely to be below detectable levels.” Nuclear gets a bad wrap because of the weapons. Most of our nuclear plants we have are still designs that came out of a previous era when weapons-grade material was a happy byproduct. But it’s possible to create nuclear plants without having weapons-grade fissile material as an output. If we fix the public perception of nuclear, then it becomes the primary solution for energy concerns, not solar and wind.

Last, there’s the problem of agriculture. Most people don’t realize the incredible advances in agricultural productivity over the last couple hundred years. As late as the 1960s and 1970s, people thought millions would starve based on overpopulation. Ehrich wrote his dystopian The Population Bomb based on the Malthusian predictions of two hundred years prior. But Ehrlich and others didn’t count on the changes in production capacity of farming, brought on by heroes like Norman Borlaug and the Green Revolution. Rice yields tripled in India from the 1960s to the 1990s, and similar stories follow rice, wheat, and soybeans all over the world.

That said, these dramatic increases ride on the backs of technologies like herbicides, pesticides, industrial farms and monocultures. And while there’s been a lot of good produced (enough to feed the world), there’s also cause for concern in the amount of fossil fuels required, the elimination of biospheres for single crops, and plain old soil erosion.

It’s tricky to estimate the impact and amounts of CO₂ that actually come into play from agriculture, but it’s definitely true that we’re having a very large impact on the terrestrial biosphere, which is one of the five main reservoirs of carbon on earth. A simple solution like no-till farming, which is happily growing in popularity, has the potential to do an enormous amount of overall good. It eliminates some of the need for fossil-fuels in agriculture, it improves the albedo of croplands (which has a positive effect on radiative forcing and will reflect more solar radiation), and, most importantly, it protects soil deposits from erosion and allows topsoil to increase over time.

If this appears inconsequential to you, consider Freeman Dyson’s calculation on topsoil. Based on studies of biomass percentages in topsoil, he calculates that one tenth of an inch of topsoil averaged over approximately half of the land area of earth (e.g., the amount of land that is currently covered by dirt, soil and vegetation of some sort) would sequester enough carbon to cease the increasing CO₂ amounts in the atmosphere.

In other words, it’s possible we could solve a significant portion of our climate problems with better land management and understanding of the nature and amounts of carbon in the dirt under out feet.

Overton Windows

If you’re still with me by this point, I’ve said an awful lot and I’ve still only scratched the surface (thus, the hyperlinks). If you’re naturally curious or interested in science and systems, that’s really exciting. It can be intimidating too, since even for the vaulted scientific cognoscenti there’s a diverse set of specialties and subspecialties at play, each with their own ongoing debates.

And yet, in public discourse at least, we don’t get much appreciation for the depth and the details. Instead, we get bludgeoned over the head with that single metric - CO₂ concentration and temperature. And more often than not when this topic comes up, you’re either with us or against us.

I can understand this reaction to some extent. There are some fringes that won’t listen regardless of the evidence, just like the flat earthers. We call these people ‘climate deniers’ or ‘climate skeptics’, and the label has become equivalent to banishment in some circles.

But we’ve lumped a very large group of people under that label, to the extent that the label loses some of it’s power. In this group are scientists, sometimes speaking outside of their expertise, who disagree with certain parts of the climate change litany we’ve discussed.

This is where things get really dangerous, because a great deal of this labeling is an effort to shrink what’s called the Overton Window around the topic. The Overton Window describes the range of ideas that are actually tolerated in public discourse. If you really want to control a debate, you don’t argue your case, you try to change the whole Overton Window.

Of all the places where the Overton Window should have the largest aperture, science should be the most steadfast.

Surveys

We’ve all heard about the overwhelming consensus of scientists on climate change. Usually, the number 97% gets thrown out. But like all complex issues, there’s a bit more nuance than that. Let’s take a look at a few of the studies:

• Verheggen 2014 says that over 90% of well-cited climate scientists agree that greenhouse gas increases by humans are causing warming.

• In Cook’s study in 2013 of almost 12,000 scientific papers on topics like ‘climate change’ and ‘global warming’, 66.4% of them expressed no position on anthropogenic global warming (AGW), of those that did, 97.1% endorsed the consensus position that humans are contributing to global warming. Their conclusion from this was that “the fundamental science of AGW is no longer controversial among the publishing science community and the remaining debate in the field has moved on to other topics.” Meanwhile some scientists included in the study have disputed the figure, saying their work was misrepresented.

• In Lefsrud and Meyer (2012), 99.4% of engineers and geoscientists agreed that the global climate is changing but that “the debate of the causes of climate change is particularly virulent among them.”

• Of the respondents to Farnsworth/Lichter in 2011, 97% agreed that global temperatures have risen over the past century and 84% agreed that “human-induced greenhouse warming is now occurring.” When asked “What do you think is the % probability of human-induced global warming raising global average temperatures by two degrees Celsius or more during the next 50 to 100 years?’’, 19% of respondents answered less than 50% probability, 56% said over 50%, and 26% didn’t know.

Still very strong agreement, as their should be, but with some divergence around future impacts. The problem is that the public discourse takes those 97% figures and applies it to several questions all at once: that the climate is changing, that we’re the cause, that the effects will be devastating, and that government must step in and solve it.

Meanwhile, the results of surveys done in the 1990s, when the IPCC was still in it’s infancy and Al Gore’s movie wasn’t out, are a bit different.

• In a 1997 survey of state climatologists, 44% considered global warming to be a largely natural phenomenon, 17% considered warming to be largely man-made, and 89% agreed that “current science is unable to isolate and measure variations in global temperatures caused ONLY by man-made factors.”

• In a 1996 survey, climate scientists answered on a scale of 1 (highest confidence) to 7 (lowest confidence) regarding belief in the ability to make “reasonable predictions” and the mean was 4.8 and 5.2 for 10- and 100-year predictions.

• In a 1992 review of surveys, scientists agree most on data and climate processes and agree least on causes of warming, predictions, and impacts. The conclusion is particularly striking and relevant for us to remember today:

Surveys of scientists can provide important information by documenting disagreement and agreement among scientists, but it is less clear that they necessarily improve the use of scientific information in policy making. This is so because survey results may be misleading to the extent that they: a) encourage scientists to make judgments that they would not otherwise make without the requisite thought and preparation; b) provide the appearance of consensus by aggregating disparate views; c) enhance the appearance of disagreement through the use of vague questions that are subject to differing interpretations; d) encourage scientists to make value judgments that are beyond their expertis and confuse value judgments with scientific opinion; and e) are misinterpreted by their sponsors for use as political ammunition. Results from surveys should be interpreted with caution; this applies to the results of our own survey as well as those of others.

It’s reasonable to ask about the differences between the recent surveys and those from the 90s. For sure, the science and the modeling has improved over time. But so has focus on the issue, with a 5x funding increase in the same time period from the federal government alone. I don’t mean to imply that anything questionable is occurring; it’s simply another deep dive I want to take to understand the changes over time.

Deniers

Whatever the correlation, it’s certainly true that the science points to humans as causing a warming climate. Now let’s take a look at a few of the scientists who are lumped into the “climate denier/skeptic” category:

• Richard Lindzen - atmospheric physicist from MIT that was a lead author on one section of the IPCC’s Third Report. He is critical of the alarmism around future climate change and it’s impacts and the power of the IPCC and UNFCCC, stating in a letter to the White House:

Calls to limit carbon dioxide emissions are even less persuasive today than 25 years ago. Future research should focus on dispassionate, high-quality climate science, not on efforts to prop up an increasingly frayed narrative of “carbon pollution.” Until scientific research is unfettered from the constraints of the policy-driven UNFCCC, the research community will fail in its obligation to the public that pays the bills.

• Freeman Dyson - theoretical physicist and, admittedly, one of my personal heroes. He is critical of the predictive abilities of the CFD models of the atmosphere over decadal timeframes, suggests that we don’t understand enough about the climate, and has questioned whether some warming could in fact be a net positive. **

• Judith Curry - Climatologist and geophysicist who is critical of the IPCC’s certainty. She claims that the IPCC ignores the “Uncertainty Monster” in it’s predictions. She also engages skeptics and encourages discussion around the topic.

• Ivar Giaever - Nobel Prize winner in physics who calls global warming “a new religion”.

• Nir Shahiv - An astrophysicist we mentioned previously. He has done his own calculations and published papers on additional potential links between cosmic rays, solar activity, and climate change.

I pulled these as useful examples from this list of scientists that disagree with climate change. These scientists are notable for a few reasons. Remember the litany of statements that get lumped together and tied to the scientific consensus: that the climate is changing, that we’re the cause, that the effects will be devastating, and that government must step in and solve it.

In my list above, the only scientist that actually questions human-caused climate change is Nir Shahiv, and he is actively publishing astrophysical papers on this topic. The rest of my examples completely agree that the climate is warming and that it’s caused by humans. They simply question some aspects of our predictive accuracy or public policy stances. For this, they are ostracized to varying degrees as deniers.

(It’s interesting to point out that all of the above are physicists of some kind, and I’ve wondered before if physicists are better attuned than other scientists to follow or pursue unusual paths or hypotheses. When you consider some of the completely obtuse and amazing parts of the universe that they study, it’s easy to imagine why.)

The concern of a lot of the scientists listed as climate skeptics is usually around the uncertainty of future predictions and effects, perceptions and alarms, and what exactly we should do about it. Which is exactly the division between science and policy.

The Nature of Science

A lot of people have trouble defining science, even if they think they know it when they see it. It’s always worthwhile to review the Scientific Method. What are the key aspects that make it special? Hypotheses and observations are critical, but are they the only defining attributes? After all, we wouldn’t consider the hypothesis that God made the sun and stars scientific.

Karl Popper, a philosopher of science, came up with a key inisight on what sets science apart: Falsifiability. A hypothesis or theory has to have the ability to be proven false. If you can’t experimentally prove that your theory is right or wrong, you can treat it like religion and both sides can argue about it ad nauseum.

Judith Curry, one of our skeptics above, discusses in detail how science has been changing over time. She says that the rise of the social sciences after World War 2 are a key factor, and that:

social claims often cannot be tested against pre-established impersonal criteria consonant with observation.

The danger here is falsification. She goes on:

Critically important was the norm that science had to be constantly open to criticism and debate. Scientists held beliefs only tentatively, based on the evolving theories and evidence, always subject to falsification.

Migrating away from this keystone of science engenders both more activism and more relativism. Scientists feel more empowered to enter the policy debate. And why shouldn’t they?

Robert Merton would disagree. He came up with the Four Mertonian Norms of science, ideals that represent the goals and methods of science:

1. Communism: all scientists should have common ownership of scientific goods (intellectual property), to promote collective collaboration; secrecy is the opposite of this norm.

2. Universalism: scientific validity is independent of the sociopolitical status/personal attributes of its participants

3. Disinterestedness: scientific institutions act for the benefit of a common scientific enterprise, rather than for the personal gain of individuals within them

4. Organized scepticism: scientific claims should be exposed to critical scrutiny before being accepted: both in methodology and institutional codes of conduct.

Curry worries about disinterestedness, saying:

disinterestedness, defined as “personal detachment from truth claims,” is the least popular contemporary norm, as academics align their research interests with funding opportunities.

It’s easy to see how advocacy or activism can follow closely after alignment with funding.

Perhaps more important is the degradation of organized skepticism. The act of organized skepticism is not a one-time judgement, like a court of law. It must be a continuous evaluation, always open to new theories and new data. But as Curry states, the values have changed:

Organized skepticism and openness to criticism have also come under serious pressure due to post-modern thinking that promotes “deconstruction,” “relativism,” and understands science as means of social control. An example of this trend is the issue of defining uncertainty in climate change science: “the mere occurrence of uncertainty talk is not interesting unless we can document and interpret its construction, representation, and/or translation. According to constructivist accounts, representations of uncertainty do not reflect an underlying ‘reality’ or a given ‘state of objective knowledge’ but are constructed in particular situations with certain effects.”

In other words, the narrative is what’s important. Error bars and statistical concerns matter less than the overall picture of the underlying reality. Consider MIT meteorologist Kerry Emanual’s comments on Richard Lindzen’s perspective on climate change:

Even if there were no political implications, it just seems deeply unprofessional and irresponsible to look at this and say, ‘We’re sure it’s not a problem.’ It’s a special kind of risk, because it’s a risk to the collective civilization.

Not only did Lindzen never say “we’re sure it’s not a problem”, his criticism and openness - in other words, his organized skepticism - is irresponsible because it is critical of the risk that the science he is critical of has shown.

The politicization of science is accelerating and we need to understand and deal with it. The norms of science have changed over time and introduced new problems and new perspectives. Why should we believe scientists are immune to argumentum ad populum? Or confirmation biases?

In the view of the wider population, this problem becomes even more important. Most people will give far more credit to a scientist if their views align with their own. None of us is immune to this.

Expertise and Specialization

Which is a problem. Over its history, science has moved from fighting authorities like the Church and philosophers to being a trusted and mostly unencumbered and neutral authority. But as the landscape of scientific inquiry and its relationship to the public discourse changes, we need to continually adjust course.

Science rightly puts heavy influence on deep knowledge of a particular field. This is called specialization. It is inevitable today to provide any sort of new formulations in any field. We build our scientists - and most academics - to be incredibly focused, specialized, and knowledgeable in a narrow field. And yet, here’s the perspective of two of the best scientists of the 20th century - Neils Bohr and Richard Feynman:

“Science is the belief in the ignorance of experts” -Feynmann

“An expert is a man who has made all the mistakes which can be made in a very narrow field.” -Bohr

“The first principle is that you must not fool yourself — and you are the easiest person to fool.” -Feynmann

There’s a humility in these experts that can be incredibly refreshing, but it’s important to remember that this healthy and continuous skepticism is a core part of what science should be. There’s a balance to be made between continually building on top of a foundation and rigorously inspecting what you’ve built.

Too often today we instead get arguments from authority, where we listen only to the specialists in a given field because they’re the expert. It’s a natural instinct and one that experts engage in as well. Re-read the Neils Bohr quote above; an expert simply understands the landscape of a very small minefield.

It’s gotten to the point that you can’t even step out of your lane. If you do, you get shamed. You can’t comment on a topic without showing your bona fides in that field.

We need to start recognizing this as a sort of ad hominem attack. I’ve deemed it credshaming: the act of shaming someone for speaking up about a subject where they don’t have the “appropriate” credential (e.g. a PhD in X). Outsiders in fields are dismissed frequently, and there are many examples of this practice, from the outlandish to the completely ordinary.

Outsider Examples

We already briefly touched on Plate Tectonic theory already, which was first proposed by Alfred Wegener. He was a polar researcher and meteorologist and was considered an outsider by geologists, some of whom organized symposiums specifically in opposition to his theory. Plate Tectonic theory wasn’t widely accepted by geologists until over 40 years after it was first proposed.

Another physicist named Tommy Gold discovered the electromechanical mechanism of the inner ear that allows us to hear by translating sound waves into electrical signals. He performed an experiment in the 1940s to verify it, but his work was rejected by physiologists and ear specialists until the 1970s, when the mechanism was finally verified by researchers in that field.

As an inverse example, when Theophilus Painter erroneously stated that there were 24 chromosomes instead of 23, other researchers suffered from confirmation bias and, expecting to find 24 pairs, nearly always did.

When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong. -Clarke’s First Rule

I’ve mentioned him enough by now, but Freeman Dyson is in my personal pantheon of heroes. He is both brilliant and heterodox. Many dislike him for this, but this is exactly why I’m so taken by his manner of thinking. His friend Oliver Sacks said about him, “A favourite word of Freeman’s about doing science and being creative is the word ‘subversive’. He feels it’s rather important not only to be not orthodox, but to be subversive, and he’s done that all his life.”

Curious intellectuals from outside a given field are subversive, sometimes with huge consequence. Alan Sokal has proven this twice. The first time he published a nonsensical paper to the academic journal Social Text to test the journal’s intellectual rigor. The paper was accepted, published, and then outed by Sokol as a complete fake. This sparked a debate around the rigor in some of the social sciences, especially those focused around postmodern cultural studies. Almost two decades later, Sokal, with others, rejected the math and statistical analysis underlying a key tenet of positive psychology called the Losada Line, despite the original paper already having been cited by other scientific papers over a thousand times.

Subversive elements like Sokal or Dyson look at questions in new ways and suffer less from the natural biases and histories of those inside the field. They won’t always be right - in fact, they’ll very probably be wrong - but their participation and serious inquiry work to solidify the fulcrum on which the balance of science sits.

Synthesis

There’s another problem with specialization that we see more often now, and it’s the problem of specialization itself. We reward based on narrow focus, and so we end up with compartments of experts that don’t often grapple outside of their lanes.

Said another way, there isn’t much in the profession of science that focuses on analyzing data and synthesizing results across fields. The aggregation techniques or overarching reviews that do exist, like meta-analysis, are still narrowly focused on a specific hypothesis, and there are natural biases embedded in the technique.

In a conversation with Russ Roberts on Econtalk, Paul Romer circles this topic in economics as well:

But there’s another problem that I think one should be worried about, which is that there may not be enough professional reward for the hard work of sorting through the facts. And summarizing them in a way that people can use them. But, yet, really being careful about the different, you know, ways to summarize them, and the possibility that you are building in some assumptions that are, you know, influencing how you summarize them. So, I, you know, I wish—over time, I’ve become more and more convinced of the importance of that pure hard work of collecting data and summarizing them in a manageable way. And, frankly, I think we give too many rewards to people who do theory, like I do. And not enough to the data that is actually the basis for judgments about, you know, like, growth, and growth and GDP. And so, as a tweak in the profession, I’d like to see us pay more attention to, you know, the analysis of data and the process of summarizing the data that we then use to make models and make policy recommendations.

As specialization continues to accelerate, we need to more coherently recognize that synthesis and summary are not a lesser form of expertise. It’s a different kind of expertise and we need to cultivate it.

Tragedies and Choices

In both the climate change debate and the issue of scientific specialization, some of the main problems are incentives. And when some of the ground rules change, the incentives do too. A scientist’s immediate goals are to publish, to do theory, to gain citations for their work. More recently and depending on the field, they can also be incentivized to weigh in on policy and public perception. Like it or not, they hunt for prestige in their field, moving as deeply as possible and specializing maximally. They become experts, and the public is thus incentivized to listen to them. These incentives are broken. They reward depth over breadth. They need to be rebuilt.

Climate change functions as a tragedy of the commons, where individual incentives aren’t aligned to the common good. This drives the reflexive reach for centralized and forceful government control. It’s not enough to say that the climate is warming and that we’re the cause. If you won’t also say it will be devastating and government must intervene, you’re anathema. Thus, most of the alarmism that we see today is constructed to force action on the issue.

But what kind of problem is climate change, really? Is it an asteroid hit or is it diabetes for the planet? An asteroid would wipe us out quickly. Diabetes can be managed. It will certainly change things, but it’s not the end of the world. The Green New Deal wasn’t constructed to fix an impending catastrophe. It’s a way to gather momentum behind a set of unrelated policy prescriptions. It leverages the idea of catastrophe to advance policy, but those policies don’t actually help much. As the above opinion article states:

Moreover, practically, nothing that Green New Deal advocates appear willing to seriously propose will actually cut US emissions at a scale or pace consistent with stabilizing emissions below two degrees, much less 1.5. Making the best of our chronic condition, rather, will require a climate movement that is less catastrophist about the problem and more ecumenical about its solutions.

The usual solution to a tragedy of the commons is government control: centralize authority and legislate mandatory compliance. For the common good. But there’s another path: design incentive systems that make individual choice align with collective good.

That’s what government does, but they do it through force of law. We can do it through economic incentives too. Instead of forcing policy and taxes, we can pursue technology and R&D that allows us to do things like bring green energy prices lower than fossil fuels. Once solar panels, wind, and nuclear are cheaper than coal-driven power plants everyone will want to switch.

Precaution and Choosing Right

Which brings me back to the Copenhagen Consensus and something called the Precautionary Principle. This principle is a risk management technique, usually used to say that we should take action on climate change to ensure there is no impending harm.

Every Copenhagen Consensus vote puts climate change investment near the bottom. Not because climate change isn’t real or important. Because other problems are more solvable, right now, with less money. The payoff based on the amount of investment required doesn’t make sense. Is it possible that in the future more people will die and trillions in economic loss will be sustained? Yes. But it’s also true that millions die today from much more solvable problems. And it’s the same with economic losses. Investing in eradicating malaria is much cheaper than working on climate change and it would save millions of lives right now. Similarly, eliminating trade barriers would reduce economic losses significantly.

The problem with climate change isn’t that it’s wrong, it’s that it crowds out too many other important issues. Just in the environmental sphere, we’re missing proximate causes. It’s true that climate change will still be a threat to polar bears. But the primary issue there was actually hunting, and addressing that has allowed the populations to stabilize. There are plenty more specific problems for us to focus on, like the potential extinction of bees, deforestation, overfishing, plastics in the ocean, and any other number of issues. When we focus exclusively on our CO₂ numbers over the next 50-100 years and nothing else, we miss the things we can do right now.

The Precautionary Principle demands action. But consider: we could eliminate nearly 100% of traffic deaths with one rule. Cut the speed limit to 5 mph everywhere. We don’t do this. As a society, we accept risk in exchange for other goods. Extreme examples show that any decision to do something (or not do it) is always a cost-benefit tradeoff. We can’t just evaluate the problems of climate change. We need to compare the costs and the pros and cons of the solutions against the problem itself. It’s likely that our investments and focus could be better spent elsewhere.

The Key Insights

We want to do something, but determining what we can do that is actually beneficial on a large scale can be, as Judith Curry says, a “wicked problem”.

Now I will confess that I come into this entire subject with my own biases, which are surely very apparent. I agree completely with Dyson’s view on being subversive and maintaining a skeptical perspective on science and data. It’s vital to approach ideas this way, even if only to ensure that the agreements being formed are real. Groupthink is real. Other cognitive biases are very real too.

The Wisdom of Crowds is real too. But it requires independence of opinion and diversity of perspective. When experts tell the crowd the answer up front, wisdom disappears. You get groupthink. You get 24 chromosomes. You get the Losada Line. We have to maintain our skepticism and our independent opinions to ensure the whole process is actually working.

Value specialization, but don’t let expertise become a weapon. Smart outsiders see what specialists can’t because they aren’t hampered by what everyone knows to be true. Remember that science is falsifiable. Predictions are a part of science — the theory and hypothesizing part of science. The experiment has to run to tell how accurate those hypotheses are. We’ve been wrong a lot in the past.

Separate science from policy. Science can be politicized, and when it is, it violates every one of the Mertonian Norms. Scientists in the public eye can’t treat the public like children who need to be frightened into compliance for their own good. There’s a line between persuasion and propaganda. Skeptics change their minds when shown evidence. Zealots don’t. The moment we stop treating climate as science that can be questioned, we’ve made it religion.

What fascinated me about Beringia was how different it was from the Ice Age image I’ve carried my whole life.

Here’s what Beringia taught me: our mental models are fragile. We think we understand ice ages, and we’re picturing a frozen bridge instead of grasslands the size of Alaska. We think we understand timescales, but plate tectonics was considered crackpot a hundred years ago. We think we understand consensus, but Wegener died ridiculed for ideas that became orthodoxy.

Climate science isn’t special. It doesn’t get exempted from the humility that every other science requires. Predictions are hypotheses. Policy is choice. Consensus is useful. Sometimes it has the fuzzy shape of truth but the picture takes decades or centuries to come into focus.

Question expertise. Every system that claimed certainty about the future has been humbled by what actually happened. Learn the data and science yourself. You will gain knowledge, insight, and inspiration in a far richer way than any 30-second clip can offer. There’s media hype from zealots and deniers alike; if you learn the underlying data you’ll be more skeptical about everything you hear.

That’s a good thing. Reality is always stranger, richer, and more complex than our models suggest. That’s not a reason for despair. It’s a reason for wonder. And for humility about what we claim to know.

Note: I originally wrote this in 2019. Despite changes, the discourse remains rigid.

Elizabeth Warren piles on to AWS. Elon piles on about Claude. And everyone piles on about Atlas.

But there was more! The shutdown continues. Bernie admits Trump is his heir. Then gets mad at Amazon for efficiency. Fetterman declares Groundhog Day. Klobuchar tries to make people feel bad for early-retiring boomers and fails. Hillary gets mad about the East Wing. JD Vance gets an answer for Harry Sisson in response to Trump’s latest ridiciulous AI tweet. This timeline is the craziest possible timeline.

TSMC shows off it’s Arizona fab. A King of England visits the Sistine Chapel. Tesla FSD gets Mad Max mode. Claude Code hits the web. The Pentagon gets a whole new Press Corp. The Israeli Knesset votes to annex the West Bank. CZ gets a pardon. Datacenters represent most of GDP growth. The UK keeps pushing Digital ID.

On to the reading!

Timely

• Why I Stopped Being a Climate Catastrophist - The always brilliant Ted Nordhaus with a full repudiation of catastrophism and a great telling of the problems of the climate activist cycle.

• The Foolishness of ‘No Enemies to the Right’ - Seth Dillon of the Babylon Bee with a very timely reminder that the Right needs to look after itself, be weary of it’s own excesses, and renounce the bad ideas that are out there.

• Wanted: Men of Purpose - “Shaping a sweet boy into a strong and good man is principally a soulful venture, one that requires great attention to virtues.” Truer words have never been said, it takes a lifetime. A call to give boys the examples and role models they desperately want so they don’t find themselves lured to the shadow of masculinity in some dark corners of the Internet.

• Vibing a Non-Trivial Ghostty Feature - A great deepdive into what it means to code with AI today. It’s vibecoding, yes, but not the trivial kind. This is the thoughtful and sustainable kind, and we’re all going to be doing more of it in the future.

Timeless

• INFRASTRUCTURE INEQUALITY: WHO PAYS THE COST OF ROAD ROUGHNESS? - An incredible paper based on Uber data that shows rough roads are more common in poorer neighborhoods and towns causing them to pay more per mile than other areas.

• Are Ideas Getting Harder To Find? - Signs point to yes, “research productivity is falling sharply everywhere we look.”

• The CEREBELLUM as NEURAL ASSOCIATIVE MEMORY - Slides, yes. But incredible ideas. Neuroscientists are trying to figure out the deep purposes of the brain based on the new mental models of LLMs.

Books

• The Scaling Era: An Oral History of AI, 2019-2025 by Dwarkesh Patel - This is the latest Stripe Press book and I just finished it. It’s mostly snippets of dialogue from Dwarkesh’s podcasts with everyone from Mark Zuckerberg to Gwern to Dario to Ilya. It’s broken into sections about why models work the way they do and predictions about what could happen next. While it’s certainly technical and the topics are dense, it’s a remarkably approachable book and a great overview for anyone interested in why this revolution is such a big deal.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

We’ve also got Bitcoin whales making 9 figures shorting before Trump’s announcement, epic trolling on Columbus Day, DHS pretending to be Billy Madison, validation for reading 25 books, and the media refusing the new Pentagon coverage rules. Putin and Trump talk on the phone and Hungary becomes a neutral spot.

Nvidia releases DGK Spark. The US runs a September surplus. Starship returns and flips and lands again. 2026 grad school tuition goes down. Gold keeps climbing. Claude gets skills. Broadcom gets some of the OpenAI energy. Anduril shows EagleEye.

There’s just so much! St. Carlo Acutis, pray for us!

On to the reading!

Timely

• Technological Optimism and Appropriate Fear - Jack Clark - one of the other founders of Anthropic - with some storytelling and metaphors for all the AI progress and where we’re likely going.

• The common words list and the real magic of writing - Most writing is just 300 common words, with a few others sprinkled in here and there for good measure. And yet from this comes a whole “portable magic” that runs our entire civilization.

• Edison’s Lesson for the AI Age: Why the Capex Boom Isn’t a Bubble - It’s all the rage to say AI is a bubble. But with very little debt and free cash flow paying for capex, it doesn’t have an end in sight yet.

• Context is the bottleneck for coding agents now - We’re still getting as much compute as we possibly can, but for practical applications the real limitation is context. Memory. How much can we load into a working window?

Timeless

• Getting Jacked Is Simple - You may not like it. I may not like it. But it is quite simple.

• Math Every Day - I still remember first reading this almost 20 years ago. It remains the best intro to John von Neumann and why he’s important. It’s also exceptionally good context for what a computer is and why AI is so important (read: it’s the first computer architecture that isn’t just a von Neumann machine). Yegge is a great writer and this is approachable for everyone.

• Don’t Worry It Can’t Happen - A 1940 article on uranium fission describing how the scientists are cautious and the evidence suggests there’s no reason to worry.. these reactions just sort of fizzle out.

• The Origins of Representation Manifolds in Large Language Models - One of the more interesting ideas around how LLMs work involve the manifold hypothesis. A below the surface dive into how manifolds work and what’s going on, with just a tiny bit of math.

Books

• Skin in the Game by Nassim Taleb - This was the first book from the Incerto that I’ve read and.. it’s interesting. Taleb’s style is both engaging and off-putting. The editing is incredibly light and it sometimes feels more like stream of consciousness than anything else. Let’s call it storytelling. Taleb focuses on both incentives and disincentives and expresses the power of all things via negativa. The idea of the Silver Rule (vs. the Golden Rule) - and it’s role on the ideas of negative and positive rights - has stayed with me. Most importantly, as reiterated over and over and over by Taleb, identify others, leverage groups, and have for yourself some skin in whatever game that you’re playing.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

Maybe it’s OpenAI. Or Nvidia. Kinda weird that stocks went way up up and the labor force went way down for the first time ever right when ChatGPT got released, right? Probably nothing.

Speaking of OpenAI, they now own 10% of AMD.. somehow? Matt Levine describes it well. And some insider made a ton of money on calls. OpenAI had a Dev Day too, and 1,000 startups screamed. There’s an Apps SDK, an Agents SDK, and Sora. And maybe it’s not just startups screaming, but every software company that exists. Maybe that’s why they’re all attaching themselves to Sam & co. Where will search happen in 5 years? ChatGPT. Buy new house goods? ChatGPT. Pay for bespoke, handcrafted yarn sweaters? ChatGPT. Write your own app? ChatGPT. Find a flight and hotel? ChatGPT. You get the idea.

And what’s Anthropic doing? They’re capturing hearts and minds with hats. No seriously, I want one of those hats. I’d stand in line for it. Sonnet 4.5 is amazing. They’re the anti-slop company now. But it also feels like they’re positioned as the elite alternative to the behemoth. But both companies are two sides of the same coin. And the best engineers aren’t writing code anymore.

Now is all this a bubble? Maybe, but it’s all still capex from earnings. Datacenter growth keeps coming. Electricity prices are going up. And there’s plenty of liquidity. Also read the new State of AI report, it’s a lot.

What else? Israel and Hamas have a ceasefire. The White House blesses the peacemakers. Except ICE agents, who shoot pepper balls randomly. And democrats. And also let’s give Ukraine some Tomahawks.

The Church of Nigeria breaks from the Anglican Church over the new Archbishop of Canterbury. Katie Porter bombs an interview and they dig up plenty of old stuff too, and how is she actually running for Governor? China locks down the global rare earth metal market which could make things sticky very fast. A man tries to blow up the Red Mass for the Supreme Court.

Figure previews the latest robot. Tesla releases FSD v14.1. Old monkeys get a shot and get younger. Pope Leo releases his first Apostolic Exhortation, carried on from Francis, Dilexi te.

And most important of all, my daughter plays two new Taylor Swift songs.. and they’re kinda good?

On to the reading!

Timely

• Donald Trump’s Worst Idea - Tyler Cowen on skewing the market with nationalized buy signals, and how Trump might usher socialism in the side door.

• The Techno-Optimist’s Guide to Futureproofing Your Children - Will kids need careers? Colleges? Is Alpha School and AI tutors a sign of the future? Or a dead-end? I remain, skeptical, but agency is still important.

• Hamlet Is The Gen-Z Story We Need - Hamlet, it turns out, is one of the middle children of history too, with no purpose or place.

• Some Thoughts on the Sutton Interview - Dwarkesh comes back with a refined view of what “more compute” really means in the Bitter Lesson and how continual learning might work one day.

Timeless

• Machines of Loving Grace - A completely brilliant and optimistic piece from the head of Anthropic on the enormous potential of AI. Required reading.

• Language Without Mind - Stochastic parrots AI may be, but they are still startling. Will then make us forget our own inner lives and quests; our souls?

• Failure Is a Great Tutor—Don’t Fire Him - Monday night, my son caught a long throw from his friend for a game-winning flag football touchdown with 5 seconds left. I wasn’t proud of him for the play, but for the perseverance to join the team, to learn, to work so his teammates can trust him, to drop passes and get up and try again, and to practice when he’s tired.

• The Myth of the Sommelier - The Judgement of Paris, the influence of Robert Parker, and the “veneer of credibility” around expertise and wine tasting.

Books

• The Pendulum: A Granddaughter’s Search for Her Family’s Forbidden Nazi Past by Julie Lindahl - Julie Lindahl gives us a magically intimate retelling of the research she did into her family’s Nazi past. She grew up in Brazil — where her family had fled in 1960 — with strange holes and disconnections about her family’s past. There are many memoirs from the perspective of Jewish descendants, but few that I’ve seen from Nazi descendants. From her relationship with her loving but Nazi-sympathetic grandmother to the encounters with old men who still remember her cruel and fanatical SS grandfather, this book holds nothing back and is absolutely beautiful.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

The views had cratered on the Summit episodes they thought would crush — Tulsi Gabbard, Mark Cuban and Tucker Carlson. The hosts immediately suspected the modern bogeyman, shadow-banning. YouTube has a history of stamping down on right-coded content and these are clearly right wing guests, but the truth was simpler and worse. Months earlier they’d stopped bleeping profanity and started muting it instead. YouTube’s transcript system still “heard” the words, flagged the episodes, and quietly dropped them into Restricted Mode. Schools, offices, public Wi-Fi—anywhere an admin flips that filter, the videos vanish. Not removed. Not flagged. Just invisible.

They called YouTube, and Lo!— they have enough clout and connections that YouTube helped. All-In learned about the difference, went back to bleeping, the platform un-restricted the back catalog, and everyone moved on grateful for the fix.

But YouTube had all the data. Their engineers could trace exactly what triggered the restriction and when. The platform could see the problem the whole time. The creators couldn’t. That asymmetry is part of the business model.

Opacity is the architecture.

We keep treating this stuff like bureaucratic incompetence — bad dashboard UX, no Restricted Mode indicator, no timestamps where the policy tripped, not enough support staff. We’re missing the actual structure of power here. YouTube isn’t withholding information because they’re disorganized. They’re withholding it because knowing the rules in detail would let you optimize for them, work around them, test their boundaries, and, just maybe, build somewhere else.

Here’s an example from the annals of tech history. When email first became a thing it was written as a protocol. You could read an RFC, implement SMTP on your own server, and if you got the verbs right — HELO, MAIL FROM, RCPT TO — you were in. Email worked because the rules were public. Competence was the price of entry, not permission. Today you can run an email server perfectly and still be invisible. Why? Because delivery is controlled by private reputation systems that Google and Microsoft operate. You can be technically correct and practically nonexistent.

Email still exists as a protocol. But in practice, without a Gmail account, you might not have email at all.

The same drift swallowed the web. YouTube, Twitter, and Facebook all solved real problems: spam, abuse, fraud, scale. I’m not romanticizing the old internet; it was a mess. But the winning solution to those problems wasn’t the only solution.. it was the one that made the most money and maximized control.

Now our lives are lived in their house under their rules. They write the rules, enforce the rules, and decide whether you get to see the rules. When they’re feeling generous, they explain what happened after you’ve already been punished. The All-In hosts thanked YouTube for explaining what had been done to them. But when they’re not generous, you get silence.

YouTube engineers can see exactly why content gets restricted: the flags, timestamps, and policy logic. Nothing technical prevents them from exposing these verdicts to creators. Nothing stops them from making a machine-readable verdict to the wider ecosystem under an open spec.

They just don’t. Why would they? Transparency creates accountability. Accountability creates constraints. If creators could see exactly why their content got throttled, they could test the boundaries, optimize around the rules, maybe even challenge decisions that don’t make sense. If the rules were machine-readable and public, competing platforms could implement the same moderation standards and compete on execution instead of lock-in.

The vagueness protects the moat.

I keep thinking about what interoperability used to mean. Not the modern version where a platform lets you download a ZIP file of your data — thank you, thank you, very generous — but real interoperability. The ability to use (or write!) third-party clients that work better for your use case. To see the rules you’re being judged by and verify that they’re being applied correctly.

We had pieces of this. Not perfectly, not everywhere, but we had it. The web was built on open protocols because the people building it thought rules should be legible to the people being ruled. WordPress is a mess in a hundred ways, but the open and movable nature of the framework still demonstrates the values of the open early aughts. We’ve traded that ethic for scale and surveillance economics, and now we act surprised that the platforms behave like landlords instead of utilities.

Traditional antitrust asks whether prices went up, but these platforms are all ‘free.’ This capture is subtler: private APIs with private rules aren’t toll booths, they’re zoning laws you can’t appeal. We gave these platforms Section 230 immunity on the theory that they were neutral conduits. Utilities. But they operate like landlords with total discretion over their property. They get the legal protection of the first model while exercising the power of the second.

Dominance should come with obligations. If you’re going to behave like a landlord, you don’t get utility protections. And if you want utility protections, you need to publish the rules you enforce and let people verify you’re enforcing them correctly. Appeals should be APIs, not ignored suggestion boxes.

The platforms know we won’t leave. We know they won’t change because the attention economy runs on captive audiences. This is the deal: total attention for them, total opacity for us. But at least we know what we’re choosing when we stay. Before you build on someone else’s platform, ask: where’s the spec? Where’s the export? What happens if they change the rules?

The answers will sort gardeners from jailers.

The Medium is the Message, as McLuhan said. Right now, the message is: we own your data and your attention. You can’t leave.

And there were new models! Anthropic released Sonnet 4.5. It is a very good model. Rate limits? Maybe not so much. But still exciting! OpenAI, in the consistent spirit of trying to upstage other labs, released Sora 2 and we’re here for the memes. There’s Bob Ross painting, 90s Epstein Island ads, and Sam stealing Ghibli images. Also Thinking Machines finally shows something, anything, and it’s a fine fine-tuning tool. But nobody cares, we’re just waiting around for Gemini 3. Supernova is still unclaimed.

Accenture does layoffs and the S&P is clearly diverged from the labor market, I’m sure it’s nothing, but the M2 money supply keeps going up. The entire Arab world except Hamas gets on board with Trump’s Gaza peace plan and the Secretary of War grandstands at Quantico.

Elsewhere, JK Rowling responds to Emma Watson. Nature published a new paper where skin cells were turned into eggs, fertilized, and turned into properly dividing embyros. Pope Leo blesses a block of ice. The French stop a Russian tanker. The US considers selling Tomahawks. Keir Starmer breaks unpopularity records, not sure the people like Digital IDs.

On to the reading!

Timely

• We Are The Slop - Freya India with an absolute home run. We’ve turned our lives into content. What a waste.

• The End of Thinking - Look: want to be a competent human? It involves reading and writing.

• Durbin declines award as USCCB pressure mounted - The Pillar breaks down what was going on with Dick Durban’s lifetime achievement award, the American bishops response, and Pope Leo’s as well. Shagbark had the best response to the ice blessing too: God bless Pope Leo!

• If the Average Person Did This for 6 Months, They’d Be Unrecognizable to Their Friends & Family - Everyday we make choices. You know what will make you better and what will make you worst. Which do you pick? What systems do you have to help you make the choice?

Timeless

• They Are But Dust - A gentle reminder that your children are dust, just as you are dust. Memento mori.

• The Social Responsibility of Business Is to Increase Its Profits - Milton Friedman explains why activism capitalism doesn’t work and how to properly consider the place and propriety of a company and it's executives.

• The Big World Hypothesis - A short but thoughtful paper on the idea of a world being larger than the learning agent, and what that means for algorithmic design

• How I Read - Remarks on the details of the process, reminding us all that we all have different rules and frameworks and that is OK!

Books

• Transformers, the Tech behind LLMs by 3Blue1Brown - Let’s be clear, this is NOT a book! I know, I’m devastated too. It’s hypocrisy. But it’s useful hypocrisy. This is a supremely well-executed explanation of exactly how an LLM works. The fact that it’s a whole bunch of math scares people.. well if you don’t want the math but want a great high-level understanding, this is your video. Start with the embeddings, if you want to understand how words become math, for example, how the word “King” gets modified by the context around it.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!

Psyche!

This week Trump manages to get on board with Ukranian victory. Europe warns Russia not to violate NATO airspace. Lavrov complains about NATO borders and says they mean war. Canada and Australia and England and France and Portugal all declare Palestine a state. Netanyahu delivers remarks at the UN and others walk out.

A sniper at an ICE facility goes for the agents and instead kills two immigrants and then himself. Erika Kirk delivers a remarkable statement in this political environment. Others, maybe not so much. Tim Allen follows Erika’s lead.

Nvidia is investing $100 billion in OpenAI. And if you’re not sure, a whole lot of that investment will go towards.. GPUs. Which makes a nice little circle with Oracle. Probably nothing. And of course Sam warns we must do this because otherwise we have to choose between curing cancer and free global education. Not short on ambition or confidence, that.

Apple cancels a show. Jimmy Kimmel is back. YouTube did censoring. Meta releases Vibes. Tylenol is now bad. OpenAI releases Pulse.

On to the reading!

Timely

• American Students Are Getting Dumber - The question is why? We’ve reduced the friction in our lives to such a degree.. as Matt says, “you have to do things that are difficult or you don’t get any better.”

• Abundant Intelligence - In the spirit of keeping up with what Sama says.. we will keep up with what Sama says, and then decide what he actually means by saying it.

• Rage of the Falling Elite - Rob Henderson argues that downward mobility is the lens through which to understand the political machinations of our time.

• The Bluesky-ization of the American left - Bluesky is a place lefty people hang out. The kinder, gentler Twitter.. except it’s full of rage and angst. This isn’t a fundamentally left-oriented thing, but it’s current manifestation seems to be.

Timeless

• I Can Tolerate Anything Except The Outgroup - Scott Alexander at his very, very best before people knew who Scott Alexander was. A treatise on the nature of intolerance, virtue, and performance.

• The Algorithm Will See You Now - AI is better at radiology than humans, and yet the average salary for a radiologist has skyrocketed. What is the job of a doctor and how do tools help?

• The dawn of the post-literate society - As people stop reading, how does society change? Can we still think? Short video is the lingua franca of the internet.

• Falsehoods Programmers Believe About Names - Names are simple right? You have a first name and a last name. Sometimes a middle. Oh no, no, no, they are not. And this serves as a reminder that the fixed rules of the world you think you see around you are harder to generalize than you think.

Books

• More From Less by Andrew McAfee - For the last three hundred years, growing our economy meant using more natural resources, which is bad because there’s a finite amount of those. But over the last several decades, that trend has decoupled and growing the economy now seems to need less resources - sometimes much less. Andrew McAfee describes why economic growth will not continue to be a net negative for the natural world. Buckminster Fuller called this idea ephemeralization way back in the 1930s. It was the idea you can do “more and more with less and less until eventually you can do everything with nothing.” And this idea has mostly proven right, but it’s called dematerialization by the economists.

Tweets

Some good ones, so you don’t need to scroll!

The world is amazing. Cheers!