Being the self-improvement junkie that I was, I looked around for advice — and quickly noticed one trend: my friends who went to fancy boarding schools consistently scored As.

So I asked them what their secret was. The kindly answers were boring in retrospect. They were more careful than I was. They didn’t write their essays in a crazy all-nighter one day before it was due. They printed their essays and read each sentence backwards to edit. Their writing and grades were the result of hard, careful work — work I wasn’t doing.

And also — I learned — there was one hack. It was called the Magic Thesis Statement. It’s a surprisingly banal template:

By looking at X, we see Y, which most people don’t see… this is important because Z.

But once I knew it, I could see how many good essays follow this structure. To this day, much of my best writing follows this format.

So in this post, I’ll show you how my writing follows the MTS – and, through that, why it works. Hopefully, it’ll help you improve your writing, or at least provide some rope to lift yourself out of your next rut.

One of my most popular blog posts is “What are you getting paid in?” In it, I look at how Taylor Swift makes far less money than Ken Griffin, but is objectively way cooler. I use this example to explain how there are ways to get ‘paid’ that aren’t just money – for instance, status or happiness. And I conclude by explaining how it can inform your thinking about your career and what you want to get out of it.

Or, in other words:

By looking at [X = how Taylor Swift is cooler than Ken Griffin], we see [Y = that there are ways other than money to get paid], which most people don’t see… this is important because [Z = it affects how you should think about *your* career choices!]

A perfect MTS essay.

Why does it work? A professional writer might complain that the MTS is overly prescriptive.

Perhaps, but for the average writer, I think it’s often a useful way to turn a vague idea into a compelling essay. Each of the three parts of the MTS matters. Let’s look at it again:

By looking at X, we see Y, which most people don’t see… this is important because Z.

When I’m writing an essay, I often start with some combination of ‘X’ and ‘Y’. ‘Y’ is often why I’m bothering to write the essay at all. But it’s usually elucidated by ‘X’, which is some specific use-case or story I’m trying to tell. For me, the ‘X’ is often the best part of an essay – it draws the most on fun stories or life experience, so it’s what gives the essay vim.

An argument (‘Y’) on its own often sounds banal. Take the core argument of The Baked Goods Theory of Social Interaction: ‘You should go to a social event every week.’ Okey-dokey.

A story (‘X’) on its own seems irrelevant. The core story of Baked Goods Theory is that I went to a weekly choir rehearsal, and someone brought in brownies once. On its own, this is about as interesting as a bingo game at a senior home. Lovely, joyous, not essay-worthy.

But combine the two, and you get something more interesting: the claim that once a week, you should go to a social event that someone could bring brownies to. And the story is that I once went to choir and had some brownies — and through that, realized I had been in a state of crisis. The combination adds a new dimension to both the argument and the story.

But this still isn’t enough — and this is where ‘Z’ comes in. ‘Z’ is what prompts you to think about the reader, and why they’re reading an essay.

Before I lived through the experiences that would lead to Baked Goods Theory, I thought I was okay. I lived in the suburbs with my partner and worked remotely. We had friends and went to the occasional party.

But it wasn’t until I went to choir that I realized what was wrong. I had an unsustainable social life, and I didn’t realize it. And I saw how easy it was to fall into this trap.

We all know about the loneliness epidemic: that many are slowly, inexplicably ending up isolated. I saw it happening to many of my college friends, as we spread across the country for jobs and family. By writing Baked Goods Theory, I could tell them what was happening, why, and what to do about it.

‘Z’ is the most important part of an essay. I once thought that writing essays was about crystallizing a good idea. I realize now that it’s actually about making a point to a specific audience. No creative work matters without its consumers; the best essayists start from who their readers are, and what they want.

In high school and college, it’s hard to really get this, because we’re playing the academic game — doing things like ‘situating a thesis within an existing body of work’. But in most other writing, we’re actually trying to achieve something. In the workplace, you might be trying to influence your team or manager; getting a memo right matters, and so you have to appeal to whatever the company values most.

And if you’re writing on the internet like me? This writing is the most fascinating. On our browsers, we’re all primal; to keep someone’s attention, you must provoke fear, or create entertainment, or tell them how they can become a better version of themself. We’re just here for the dopamine hit. Can you deliver it?

An astute reader might notice that this essay itself is a straightforward application of the Magic Thesis Statement. By looking at how my writing applies the Magic Thesis Statement, we see that it’s an effective tool for framing essays, which most people don’t see – and this is important because it tells you how you can communicate your ideas more clearly and become more influential!

Of course, like all rules, the MTS is meant to be broken – and the very best writers throw it out the window. But the next time you find yourself struggling to put an idea into essay form, I suggest you think about the MTS. What are you trying to say? What non-obvious example or story are you using to demonstrate your point? And why should your reader care about it?

This was a real problem I to solve recently. Whenever our users create a trip, we find a beautiful photo of their destination and present it to them. To do this, we need a system that could understand anything, and respond with a hand-curated photo.

To solve this, I used LLMs for understanding, traditional software engineering in the middle, and human curation (by me) of photos by excellent (human) photographers. By walking through this, I hope to provide some inspiration for how to use LLMs in ways that feel crafted and not like slop!

The problem

I work on Stardrift, an AI travel planning app. Our central conceit is that we let people type whatever they want into a chatbox. But then we need to turn those conversations into a beautiful homepage of images for them:

Here are some silly ideas for how you could solve this. You could AI-generate an image for each conversation. But AI-generated images suck, and it’s expensive. You could Google search the destination – but that has copyright issues, and some risks:

Ultimately, what I wanted to do was hand-curate a beautiful mapping from ‘location’ to ‘destinations’. And I wanted to match a query that could be about literally anything to this.

If you break this problem down, there are effectively three problems here:

1. Take a freeform query (like ‘sfo->jfk’) and turn it into a ‘place’

2. Build a database of ‘places’ -> pictures

3. Build a software system that can take a ‘place’, look it up in a database and spit out the right picture – even if that ‘place’ isn’t in the database

To explain how I built this, I’ll run through each step-by-step.

1. What is a place?

This was the simplest part of the project, from a technical perspective – and but also the trickiest to design. (If you’re not a programmer, you may want to skip this and jump down to the next section.)

Nowadays, you can easily run a query like ‘SFO-JFK tomorrow’ through an LLM like Haiku and ask it to tell you where they’re going. Which is really cool — 5 years ago, this would have been impossible!

However, I had to be careful about what exactly I asked it for. If you think about it, a query like “SFO-JFK” should return “New York.” But “plan me a road-trip around the Isle of Skye” should return a picture of the “Isle of Skye”, not a generic picture of Scotland.

To add to the complication, very often our users mention multiple places in one chat — they might be going on a honeymoon trip through France, Germany and Belgium. We need to capture all of this.

After thinking a bit, I decided we’d build the system around this idea of a ‘place’. I decided that every query could give us a list of places, and every place would be a combination of a ‘name’ (e.g. New York), and a type of place (e.g. city/region/country).

So when the LLM is given “sfo-jfk,” it returns “the city, New York”. “Plan me a 3-day road trip around the Isle of Skye” will return “Isle of Skye, the region”. And if you ask about multiple places, we return a list. This sounds a bit dry and technical — and it is — but it was important to make the rest work.

For the programmers, here’s the base datatype:

class Place:
  type: Union["region" | "city" | "country"]
  name: str

def query_to_place(query: str) -> list[Place]:
  ...

2. Build a mapping of ‘places’ -> pictures

Now that I had a definition of a ‘place’ defined, I could start creating my database of pictures.

This part was the most fun. I ran this new function I’d written to map queries to places on a sample of real Stardrift queries, which gave me a list of places ranked by popularity.

Then I wrote a little game.

We source pictures from Unsplash, a fantastic photography bank. Unsplash has an API, so I wrote an internal tool that went through this ranked list of locations and pulled the top 5 images from Unsplash. This let me pick the best one, and it would be saved into a database:

Because of some API restrictions, I could only do about 20 places per hour. So while it only took me an afternoon to code the rest of the system, it took about 3 days to populate the database with the first 500 places, done mostly in 5-minute chunks every hour. But it was worth it – the game was very fun, and the mapping we produced is beautiful!

3. Putting it together

Now I could turn a conversation into a ‘place’, and had a database mapping ‘places’ to beautiful photos.

But our users could ask about anything, and our LLM would simply pass on whatever it said. To add insult to injury, our LLM wasn’t very precise — sometimes it would tell us the user was going to ‘NYC’, sometimes ‘New York’, sometime ‘New York City’. So how I meant to handle places that didn’t have an entry in the database?

Well, thanks to Google Maps, it’s very easy to turn any place name into latitude/longitude coordinates. So whenever we got weird input, we could turn it into a coordinate. And then we could ask: what’s the closest place that we do have in the database?

For instance, we might get a place like ‘Deadvlei.’ I have no idea where that is. But our geolocation API tells us it’s at (24.7° S, 15.2° E), so we can lookup to the closest coordinate that’s in the database, which is (24.8° S, 15.3° E). Then we can return the nearest one: in this case, Namibia:

This mostly works, but it does mean there are a few places that don’t properly show up! One of the last tools I built was a little map that showed us where we have photos, so I could see where we had gaps. It turned out we were missing a lot of Africa and South America — they are not common travel destinations. So then I manually filled them in.

We have missed places since — recently someone tried to search ‘Mongolia,’ which we didn’t have. When this happens, it sets off an alert to us, and then I manually go in and backfill it. It works well enough.

Final thoughts

There are some flaws to this system: for instance, we have a lot of cities mapped but not a lot of regions, so you might search for ‘The Gold Coast’ and get a photo of Brisbane instead. And everything is subject to my taste—my team recently made fun of me for always picking golden hour photos. (Wait, maybe that’s a feature...)

But I like it as a small but tasteful AI project. The best AI products don’t use AI for everything; they use it for what it’s good at. And that’s exactly what we did here, mixing software engineering, AI engineering and a bit of good old human curation.

This is a cross-post from the Stardrift technical blog! Shoutout to Sarah Chieng and swyx for hosting the Write & Learn meetup where I wrote this.

Don’t get me wrong. It’s very impressive. But after trying it out in a real codebase for a few weeks, I think that view is overly simplistic. Claude is now incredibly good at assembling well-designed blocks – but it still falls apart when it has to create them.

To demonstrate, I’ll run through three real examples: a Sentry debugging loop where Claude ran on its own for 90 minutes and solved the problem; an AWS migration it one-shotted in three hours; and a React refactor where it proposed a hack that would have made our codebase worse.

The same pattern explains all three. And in doing so, it also demonstrates what senior engineers actually do – and why we’ll be safe from AGI for a long time.

The Good

Running a Playwright-and-Sentry debugging loop

The most impressive thing Claude Code has done for me is debug, on its own.

I was trying to attach Sentry to our system. Sentry is a wonderful service that creates nice traces of when parts of your code run. This makes it easy to figure out why it’s running slower than you expect.

It’s usually very easy to set up, but on this day it wasn’t working. And there were no good debug logs, so the only way to figure out what was going on was to guess-and-check. I had to send a test message on our frontend, then look into the Sentry logs to see if we successfully set it up, then randomly try another approach based on the docs. It was frustrating and tedious.

So I had Claude write a little testing script with Playwright that logged into our website and sent a chat. Then I had it connect to Sentry by MCP, and look for the exact codepath I was trying to debug. Finally, I gave it the Sentry docs and told it to keep plugging away until it figured it out.

It took about an hour and a half, but Claude finally got it. This was pretty cool! The core loop of performance engineering is straightforward: make a code change, test, check tracing logs, repeat. With this tooling, Claude could do that work for us.

(The problem, if you’re interested, was that Sentry automatically sets up transactions for FastAPI endpoints but not for ones that return StreamingResponses. The solution was to write that in manually.)

Migrating from Modal to AWS ECS in an evening

I’ve used Modal happily for a year. It has the best UI for spinning up containers in the cloud on-demand. But last week we hit its limits, so I had to migrate us onto AWS.

I wanted to set up an autoscaling, containerized workflow on Amazon’s Elastic Container Service, since I knew this was the ‘right’ thing to do. I’ve set up plenty of Linux servers by hand, so I knew what to do. But I’ve never before touched Kubernetes or ECS. The pain of learning AWS’s terminology always put me off.

This time, I asked Claude to do it. I gave it Terraform and access to the aws command line tool. It one-shotted creating Dockerfiles for our code. Then it pushed them to AWS’s container registry, and set up the correct permissions using the cli, and set up the necessary AWS ECS configs in Terraform.

And it all worked on the first try! Amazing!

This is a straightforward task, but it would have taken me a day or two. I would have made a dozen mistakes and would have had to read through pages of AWS documentation. Claude crushed it, and got it all working in three hours late at night.

Conclusion

Both these use-cases are really impressive! They required a lot of detail and care. They each probably saved me a day and a half of low-value, tedious work. And Claude’s ability to track its own state and keep going was great! I can see why folks cannot shut up about Opus 4.5.

The Bad

What makes a good engineer?

I once knew a distinguished engineer named sweeks. Sweeks was legendary for his good code. People whispered about how he had single-handedly invented many of my employer’s paradigms for programming in OCaml.

Sweeks wasn’t a god. He wrote normal, bug-prone code, like you and me. He was good at coding because he was a gardener. Every time he walked into his codebase, he picked up his shears and manicured a bit of stray code. Over time, he rewrote every line of code over and over, tightening it down to only the perfectly-abstracted essentials.

Sweeks is an inspiration. So whenever I make a change in our codebase, I ask if it’s the most elegant solution. If not, I rewrite the code until it is. Putting in a hacky fix might take five minutes; repairing all the code around a change might take me thirty. But unless I’m in a rush, I always do the latter.

I bring this up because it’s a microcosm of what a senior engineer does. A senior engineer sees the non-obvious improvements and executes them quickly. A senior engineer identifies large step changes that are costly, but pay off in multiples down the road, and fights to get them through.

Opus is not a senior engineer.

Claude tries to write React, and fails

I was recently working on some gnarly React code. (In fact, it was gnarly because I vibe-coded it over Christmas and pushed it without properly cleaning it up.)

We had two components that both needed access to the same data: Component A had a ‘key’, and Component B had an ‘id’. And we had these data structures:

keyIdPairs: [(key, id)] // a list of tuples
idToData: Map<id, data>

Our problem, at its core, was that Component A needed to look up data on-demand as well. What to do?

Claude, our idiot savant, proposed a linear lookup:

// Scan the list to find the matching id
id = keyIdPairs.find(pair => pair.key == key).id
data = idToData.get(id)

But in context, this was obviously insane. key and id came from the same upstream source. So the correct solution was to derive a new map that updated whenever idToData did, keyToData: Map<key, data>.

If you give Claude the pure data problem, it comes up with the right solution. But in our actual codebase, it lost the plot. It couldn’t see the actual data problem amid all the badly-written React code. If I’d let it run wild, it would have made our frontend codebase worse.

Claude needs good legos

Nowadays, I think of Claude as a very smart child — one that loves to put together legos. Good infrastructure and abstractions are the lego blocks you give it. The bigger and better they are, the more you can do.

When I gave it Sentry, I could put it in a loop and watch it go. When I gave it Terraform and told it to go wild on the AWS CLI, it succeeded because Terraform is an excellent abstraction over cloud compute resources.

But when you don’t have good abstractions — like in our gnarly React code — Claude gets lost, and it can’t rescue itself.

Grant Slatton put it very well:

Since Claude can’t create the good abstractions. Claude’s powers are limited by how good the blocks you give it are. Have no illusions; Claude cannot reproduce Sentry and Terraform and Playwright. These are incredibly complex and well-designed pieces of code. And since Claude can’t create good abstractions on its own, there’s a limit to how much anyone can do with Claude alone. Even though everyone on X thinks you can vibe-code all software, I think the opposite is true: the value of good abstractions and well-designed infrastructure has never been higher.

If I had to boil down my criticism of Claude to one sentence, it’s this: Claude doesn’t have a soul. It doesn’t want anything. It certainly doesn’t yearn to create beautiful things. So it doesn’t produce good solutions. It doesn’t write elegant abstractions where there were none; it doesn’t manicure the code garden.

And this is all fine! It’s still a fantastic tool. But until it has a soul, we should all calm down a little. It’s nowhere near replacing all engineers. If anything, it makes us all much more important.

Every year, many of my friends go to Japan for honeymoons, vacations and fun.

I grew up in Asia, and my mother is Japanese; because of this, I went to Japan several times a year for most of my childhood.

So, unsurprisingly, many friends have asked me for advice on visiting Japan.

This is a hard question to answer. It's like asking, “I'm going to America for a week. Where should I go?”

Most people visit Japan by going to Tokyo, Osaka, and Kyoto. This is like visiting the US by going to San Francisco, New York, and DC.

You sure did go on a trip. You definitely went to America. 

But did you go to Disneyland? Did you explore the swamps of the Everglade? Did you stand in awe at the foot of the gorges of Arizona? Did you visit Sand Hill Road in Palo Alto and appreciate how this quiet street has generated trillions of dollars of economic growth – or even better, meet the people who made it happen? What about Fargo or Compton? Do you understand what it means to visit Selma? Did you visit a nice home in the suburbs?

Japan is as complicated a country as the United States: what it doesn’t have in physical expanse, it makes up for in longer history. Don’t get me wrong - I'm not saying you should go to Shimonoseki or Sekigahara on your first trip to Japan. But I think you can do better than the Empire State Building and the Painted Ladies.

Any travel guide to Japan will necessarily fall short. But here are eight pieces of advice I can give:

1. Go to an onsen

Onsen are Japanese baths; they’re split by gender. You show up, you take off all your clothes, you wash in a shower cubby (Very important! Don't skip this!), and you get into a large, steaming hot, communal bath.

Visitor always get squicked out by the nudity, which I find tragic. As a young girl, I’d go to the onsen with my mother, and it was healthy for me to see the wide range of female bodies: the old women with saggy breasts, the nubile young women, the middle-aged mothers with caesarean scars. As a child, I didn't care; as an adult, I've made my peace with it. Teenagers tend to be most self-conscious, though they're also the ones who would benefit the most!

Onsen are incredible and easily the thing I miss most about Japan when I'm abroad.

The baths themselves are pleasant, and designed to be beautiful. If you're outdoors, you can close your eyes in contemplation. If you’re in the right sentō, you might hear the neighbourhood women gossiping about local drama. The fanciest onsen are almost like theme parks, with multiple stories and several baths, each with their own collection of minerals. I often overheat very quickly, so I'll jump into the cold pool to cool down, and then peruse the other baths to decide what I'd like to sample next.

My favourite kind of onsen experience involves a hot bath outside on a snowy day. When you overheat, you can simply sit on the ledge and let the freezing air cool you down. It’s very surreal.

As a kid, I'd go to these onsen theme parks with my family; and after the onsen we would hang out in the lounge in our yukata, playing the silly games where you try to catch a goldfish. My parents would enjoy a drink from the vending machine as we sat in the massage lounge chairs and looked out at the view.

Really, Grant Slatton describes it best in Onsen Unreality:

You leave the baths and stop by one of the food stalls to get some ramen or teriyaki. Drink a crisp Asahi. Lounge around. Stroll outside. Go back in the baths. Come back out, play a carnival game. Eat another snack. Baths again.

Soon enough 6 hours have passed, but you feel like the outside world must have been frozen this whole time. You go back into the locker room, change into your normal clothes, check in your RFID wristband to pay the final bill (totally reasonable amounts, unlike Disney).

Then you walk back out onto the street. It's night. Catch the train back to Tokyo. What just happened? Why did this slightly hokey theme park make such an impact on my life? We may never know.

2. Escape the Golden Triangle (with care)

Everybody and their mother will tell you that your first trip to Japan has to be the classic ‘golden triangle’: Tokyo, Kyoto and Osaka.

Tokyo is well worth visiting. It’s such a large city that the sheer mass of tourists can’t change the basic underlying character of the city. One might even call it the Platonic ideal of urbanism; it's not for nothing that a third of all Japanese people live in the Tokyo-Yokohama area. For shopping, culture and food, it can't be beat. Like New York, you could spend a lifetime in the city and still not start to understand it.

Kyoto is also excellent, and it is not an exaggeration to say it has a beautiful old character that you can’t find elsewhere. In World War II, it was spared the worse of the American bombs; when Truman proposed dropping the nuclear bomb on Kyoto, his secretary of war nixed the plans as a crime against humanity.

He was right. But the old parts of Kyoto are also small, and now they are overrun with silly shops and mass tourist groups. If you do go, I recommend getting up at sunset to see the temples and gardens, and renting a bike to get out to the north of the city, away from Gion. And don’t go in the summer; it’s an oven.

Osaka is a lovely city with much to recommend it – if you speak Japanese. If you don’t, it is a second-rate Tokyo. I would recommend visiting a small Japanese onsen town instead, or perhaps even Morioka? Or you could go out to a suburb of Kobe or Hakodate, somewhere where you can sit in an empty kissaten and chat with the owners, and while you listen to good jazz and examine the collection of vintage antiques that they’ve picked up over the years. Kissa culture is made up of people who simply want happy lives; the 70s and 80s years represented by them is as Japanese as the lantern-lit streets of Gion.

If you like jazz, old cafes and getting off the beaten track, I recommend reading the travel writing of Craig Mod.

3. Eat cheaply

Japan truly is a wonderful place where you can eat high-quality food for almost nothing. Given the choice between a $30 sushi meal in Tokyo and a $150 sushi meal in San Francisco, I will always take the former.

It has tourist traps. The worst meal I’ve ever eaten in Japan was in Gion, and it was the most expensive, too. The menu was in English - always a bad sign.

Here is some standard advice you can find on the internet: Japanese restaurant review standards are much lower than the US; a restaurant rated 3.5 is excellent and one rated 4.0 should be considered michelin-star quality. Anything ranked 4.5 and above on Google Maps is populated entirely by tourists, and should be avoided.

My advice is to leave your hotel, wander two or three random blocks away, and walk into a restaurant that doesn’t have English on the menu, where the meals are less than 2000 yen per person.

Remember to look up, too; many restaurants are placed on the second or third floor of a building. You can use Google Translate’s camera feature to take a photo of the menu to order. Chances are, the restaurant also has one English-language copy of the menu which they’ll hastily fetch for you when you enter.

Consider visiting chain restaurants; they are all surprisingly excellent. Japan is not really a place where you need to plan your food in advance (unless you’re vegetarian); your odds of rolling well at the median restaurant are very high. 

4. Eat expensively, with care

One thing to remember about the best parts of Japan is that they aren’t made for you, especially if you can’t speak Japanese.

The small restaurants with eight seats that don’t take reservations? The restaurant owner expects the same eight to ten regulars to rotate in, and he spends the evening chatting with them. So if you and your four friends roll in and speak English all evening and treat him like a service worker – well, why would he ever want that?

Most of these places will simply turn away foreigners as a result. (It's not personal - the wrong type of Japanese people are also not invited - but they have the social sixth sense to tell when they aren’t welcome, so the issue never comes up.)

And yet, some of these places do serve food that is out of this world. So what are you to do?

One of the best ways to eat expensive and good food in Japan, is to ask your hotel concierge to help you book restaurants. (This assumes you are staying at a very fancy hotel, like the Imperial Hotel or the Four Seasons.) The hotel concierge is your social guide; they know which chefs like serving foreigners, and/or are willing to tolerate some discomfort for the extra money.

And you need a reservation: the best restaurants don’t take walk-ins; the chef goes out to buy food in the morning, and so he needs to know who’s going to show up. The earlier you can book, the better. 

Alternatively, services like omakase.in or Tableall can help; they usually fill in cancellations for restaurants, and they also serve to translate between foreign guests and Japanese chefs. This reddit thread is a good place to start if you'd like to learn more.

I always go into nice restaurants in Japan with the attitude that the chef is doing me a favour by serving me, and I recommend you do, too! I will always remember the complaint a sushi chef in Ginza once made to me, saying that the ‘other Americans with the black cards’ were so obnoxious. Don’t be that person!

And for the best time, learn Japanese and sit at the bar.

5. Read up on the War

Akihabara, or the Electric Town, is many things: nerd paradise, a surprisingly trashy and seedy neighbourhood, a cornucopia of electronic goods, an exemplar of modern Tokyo. What you can't tell, from the neon signs is that it started as a thriving black market in post-WWII Occupied Japan, when food was so scarce that half of Japan wouldn't have starved if not for emergency American imports.

If you want a book that will upend your vision of Japan, I recommend Embracing Defeat by John Dower. It is a detailed account of post-WWII Japan, an empire in occupation and collapse. He tells you the stories of the homeless people who lived in Ueno station; the brutal economy of the Akihabara station black market; the sultry glamour of the panpan ladies who pleasured American GIs in exchange for a pack of cigarettes or a pair of nylon stockings. 

Can you understand Japan without understanding the war? (Everyone knows which war it is, in Japan.) The theme of massive destruction from above runs through so much of the Japanese popular culture, even that which makes it out to us: in Miyazaki’s Nausicaa, in Attack on Titan. It was Clara Collier who pointed out to me that is no surprise, given the generational trauma from the firebombing of Tokyo in 1945.

Japan as a nation was stretched to the breaking point, waging World War II, at terrible cost to its civilians: Miyazaki also portrays this beautifully, in Grave of the Fireflies. For a manga history of how Japan ended up fighting such a destructive war in the first place, I recommend Shōwa by Shigeru Mizuki.

People who visit Japan sometimes wonder if the kindness and hospitality and order is the way things always have been, if it's something deeply genetic or ingrained culture. A knowledge of its history helped me appreciate that the Japanese attitude is hard-won, the result of prosperity and economic growth.

6. Walk, walk, and walk some more

Japan is endlessly complicated, and you will only ever scratch the surface on your first trip. People spend their entire lives in the country and never see it all. For all that I love the architecture and history of Japan, and see it where I travel, whenever I go I reflect on how I don’t have a community or friends there; and this is part of what makes travelling there always a little bit lonely.

But you can have an excellent time by wandering around and opening yourself up to conversation with the people you run into. And you can see so much of the city by picking different neighbourhoods and simply walking around: the book-lined streets of Jimbochō, the glittering flagship stores and sheltered ukiyo-e galleries of Ginza, the commercial malls and three-dimensional street grids of Shibuya and Shinjuku.

When I’m in Japan, I regularly walk 20,000+ steps a day. It’s exhausting. If possible, I recommend ramping up ahead of time, by deliberately walking around your neighbourhood or setting up a walking pad under your desk. While there’s some charm to taking the subway, you can afford to take taxis and Ubers to get around, and you should.

I always recommend to my friends that they bring a copy of Emergent Tokyo. This is not very kind advice, as the book is incredibly dense and overwhelming. But if you have the time to leaf through it as you explore Tokyo, it gives X-ray vision, explaining what is actually going on in the slightly drab and boxy buildings that you’ll see as you wander through.

7. Learn Katakana

This is by far the most high-effort tip, but learning katakana is how you 80/20 learning Japanese, if you’re an English-speaking foreigner.

What is katakana? At a high level, Japanese writing has 3 parts: hiragana, katakana and kanji. Hiragana is composed of round, simple characters; katakana is composed of sharp, simple characters; and kanji are all the complex characters, In this sentence:

> 冷たいビールを飲みたいな

Which reads, literally, ‘a cold beer, I’d like to drink!’ 冷 is kanji; たい is hiragana; ビール is katakana, and reads 'bi-ru'. Guess what that means.

In practice, kanji is often used for complex concepts, hiragana is often grammatical connective tissue (and some simple terms), and katakana is used for foreign loanwords.

A whole alphabet for foreign loanwords! It seems odd, but makes more sense once you consider that Japanese – like English – is a language that has been overrun by at least two foreign invasions. The first was the introduction of Chinese language and concepts, starting back in the 5th century; the second was the opening of Japan to the West in late 1800s, when Japanese scholars travelled to the west and brought back English and German concepts of politics, law and war.

Some of these are, as a result, non-intuitive; for instance, the Japanese word for ‘part-time job’ is arubaito, from the German ‘arbeit.’ But a helpful thing for you, you English-speaker, is that many food and tourism concepts are often written in katakana, like coffee (コーヒー, or ko-hi), hotel (ホテル, or hoteru) and ice cream (ソフトクリーム, or sofuto kuri-mu). It’s satisfying to puzzle out katakana and play the game of mapping it to English, sound by sound.

There are 52 phonetic characters, and no capital letters; so if you make some flashcards you can study them on the plane on the way over.

8. Remember the Bubble

The other big facet of Japanese trauma is 'the bubble'.

You actually see it most outside of Tokyo. A classic bubble experience is wandering around a small town and finding some odd, large bondoogle, like an abandoned restaurant with a giant hotdog on it, or a run-down seaside resort.

I once stayed in a lovely youth hostel in Tōno. At night I spent my time speaking with the owner. She reminisced about how much she had travelled in the 80s, when Japan was on the rise and everything was cheap. “Of course, that was all during the bubble. When it popped, everything changed,” she said.

I am not a macro-economist, but from what I understand, Japan wildly overextended itself in the 80s. Banks lent money that had no chance of being repaid; people made investments that had no chance of ever coming to fruition.

A German man in his seventies or eighties runs a restaurant near the Japanese seaside town where my parents spend the summer. Let’s call him Mr. Munich.

Once, my dad asked him why he was still working. Well, during the bubble he had taken out a massive loan to build a large German bierhall. He was proud of it; he showed us a picture of it. It was incredibly kitschy and cute. But when the bubble popped, business stopped, and he couldn’t pay the loan back.

In America, he would have wiped the slate clean. It would have been painful, and his creditors would have had to realize a loss, but they would have moved on. But the Japanese banks never realized those losses on their balance sheet. instead, his bank asks him, now, to pay a small fraction of the loan back; and as a result, he still works to bring in the income to pay it off, in this comical facsimile of an economy. 

You can run your economic policy this way it's a mistake. For two decades, it meant that the country's resources went into helping the old, and there were no jobs or opportunities for young Japanese people. There were a generation of broken dreams, and we see it today in how there are few new industries or technological breakthroughs coming out of Japan - or, indeed, talented young people. Perhaps it's for this reason that the government is keenly pushing tourism... and, well, I suppose that's why we're here!

If you find me helpful, please let me know! Don't take this all too prescriptively; the most important thing is to find what you want to find in Japan. It's an amazing place for serendipity to strike, if you open yourself up to it. It's a wonderful place, and I hope you have an incredible time.

A long time ago, a manager friend of mine wrote a book to collect his years of wisdom. He never published it, which is a shame because it was full of interesting insights. One that I think a lot about today was the question: “How are you paying your team?”

This friend worked in finance. You might think that people in finance, like most people, are paid in money. But it turns out that even in finance, you can’t actually always pay and motivate people with just money. 

Often, there might just not be money to go around. Even if there is, managers are often  captive to salary caps and performance bands. In any case, it’s awkward to pay one person ten times more than another, even if one person is clearly contributing ten times more than the other (many such cases exist).

With this question, my manager friend wanted to point out that you can pay people in lots of currencies. Among other things, you can pay them in quality of life, prestige, status, impact, influence, mentorship, power, autonomy, meaning, great teammates, stability and fun. And in fact most people don’t just want to be paid in money — they want to be paid some mixture of these things.

To demonstrate this point, take musicians and financiers.

A successful financier is much, much richer in dollars than a successful musician. Some googling suggests that Mitski and Grimes, both very successful alternative musicians, have net worths of about $3-5m. $5m is barely notable in the New York high society circles that most financiers run in. Even Taylor Swift, maybe one of the most successful musicians of all times, has a net worth of generously $1b; Ken Griffin, one of the most successful financiers of all time, has a net worth of $33b.

But more people want to be musicians, and I think it’s because musicians are paid in ways that financiers aren’t.

Most obviously, musicians are way cooler. They get to interact with their fans. People love their work. They naturally spend their days hanging out with other cool people – other musicians. They can work on exactly what they want to, largely when they want to – they’ve won the American Dream because they get to work on what they love and get paid! And in that way, they get paid in radical self-expression. (This is a little unfair, because I know some financiers who think that work is a means of radical self-expression. Knowing their personalities, I believe them, but it doesn’t help them get tables at fancy New York restaurants the way Taylor can.)

I don’t want to be too down on finance. People are different, and it’s a good fact about the world that different people can be paid in different ways. My math genius friends would hate interacting with most fans and musicians. They instead have stable jobs, rent beautiful apartments in New York and solve fun technical problems all day with their friends. That’s exactly how they want to get paid.

But when I worked in finance, people would sometimes shake their heads and ask why bright 20-year-olds would take the huge risk of moving to New York for unstable and uncertain careers as musicians, actors, or starving artists. I probably asked this question myself, when I was younger. Hopefully this provides some insight to the financiers.

So how do you make sure you get paid the way you want to? From what I can tell, the best way is to pick the right industry. It’s fairly straightforward to tell how an industry pays. Politics pays in power. Finance pays in money. Music and art pay in ‘coolness.’ Nonprofit work, teaching and healthcare pay in meaning and, a friend reports, sometimes a sense of superiority over others too.

There’s an exchange rate between many of the currencies you can get paid in, but it’s pretty bad. You can sort of buy coolness with money through philanthropy, but it’s very expensive. You can sort of turn power into money (and money into power), but you have to be careful about it if you don’t want to go to prison for corruption. In general, I’ve found it’s usually easier to accumulate what you want through time and work than try to convert between currencies.

My manager friend was an expert in understanding this. He ran a fairly boring back-office department, but it was widely considered one of the best teams around. He was, presumably, limited in how many dollars he could pay – back-office is not sexy – but instead paid his team a lot in autonomy and great peers. If someone wanted more fun and less money, he figured out how they could spend some time on a slightly useless but super-fun technical project instead. He worked hard to find interesting people for the team, and then put in more time and effort to make sure people actually got to know each other. By thinking creatively about the problem, he was able to pay people in ways other than money.

But this isn't just for managers. I think it's a useful framework for deciding what to do next. For a long time, I thought I wanted money – so I reverse-sorted my job options by salary and picked the one at the top. But at some point I realized that making more money next year was going to cost me other things I cared about, like purpose, fun and autonomy. Money was no longer my main constraint. On reflection, I’m happier now to take something that pays fewer dollars in exchange for more of the other things I want.

If you look at it this way, some jobs that seem to pay poorly actually can pay quite well. That low-paid government job pays in stability. That precarious journalism job pays in status and influence. Working at a startup pays in fun, autonomy and potentially fast career growth, even if the value of your equity goes to zero.

Don’t get me wrong. It’s still the case that some people get paid more and some people get paid less. Some people have jobs that don’t pay very much money, status, prestige, autonomy or anything else. Many such jobs exist, and they’re bad. Some people have jobs that pay lots of all or some of these – Sam Altman's job, for instance, pays in dollars, status, coolness and fun (most of the time). It’s important to pay attention to how much you’re getting paid overall, when you sum up all the currencies.

But the next time you switch jobs, I suggest you think about this a little. If the salary seems low, is it because you’re actually getting paid in something else? If it’s high, what are you trading off for it, and are you okay with that? What do you actually want to be paid in, and are you getting it?

Thanks to Dave P for inspiration, Holden L for feedback, and Rebecca W for substantial comments and editing suggestions!

Or, more than you ever wanted to know about consumer LED light bulbs.

Table of Contents

• Consider the lumen

• How much light is enough?

• Which bulbs?

  • What is light quality?

  • Why smart bulbs?

• Placement

  • A whirlwind tour of the rationalist lighting ecosystem

    • DIY single-point sources

    • Lumenators

    • Panels?

  • Miscallaneous

• tl;dr

  • This is too complicated Just tell me what to buy.

  • I am a Serious Adult and I want a less janky setup. What should I do?

  • I live in Europe/Asia/a part of the world that is not North America. What should I do?

• Health and Safety

  • Will this cause a fire?

  • How much will this cost me in electricity?

If you’re reading this, you probably spend a lot of time indoors. Apparently, humans spend more than 90% of their time indoors – more than some whales spend underwater! (This is a fact often cited by the indoor air quality movement, but I have temporarily appropriated it for the indoor light quality movement.)

There's nothing wrong with that. The indoors is pretty great. It’s climate-controlled, computer-friendly and comfortable.

Unfortunately, it’s often not bright enough for me.

When I was in university, I spent a lot of time working in rooms like the one above. They're often known as sunrooms or solariums. Over the course of four years, I learned that my ideal room for knowledge work has floor-to-ceiling windows on one side, with a beautiful view onto some green or blue feature of nature.

My adult life is better in many ways, but in this regard has been sorely disappointing. I’ve spent a lot of time working from small, dim apartments and dreary office spaces. Needless to say, neither of these fill me with as much joy as the solariums I worked in during college.

Strangely, I’ve noticed that America in particular seems to have a dearth of light in its newer homes and spaces. (I rarely feel this way in Europe or Asia.) Maybe it's because multi-family apartments are notoriously dark because of their deep floorplates and double-loaded corridors. New York is particularly bad for this, since it’s culturally acceptable for brand-new luxury apartments to rent with no overhead lighting installed in units at all. 

But it’s not just the apartments. I have also lived in single-family homes, subject to none of these requirements, which are oddly filled with dark ‘bonus rooms’ and bedrooms with only one tiny window. Simon Sarris critiques this example single-family home beautifully here.

At some point I realized I was fitter, happier and more productive when I had more artificial lighting in my home during the day. Other people on the internet have noticed this too, especially people who live in cold and dark countries, where even in old buildings they just don't get very much light during the winter. (Look up Seasonal Affective Disorder’ for more on why.)

Ideally we would all be lucky enough to live and work in beautiful and well-designed buildings on the equatorial latitudes. But if, like me, you don’t, it turns out that you can use artificial lighting to improve your health, happiness and life. It requires some work, and there are some tricks to it. But if you want to know how I've come to think about it, read on.

(Though before we do that, I want to emphasize that this isn’t the lighting solution to solve all problems. If you want an aesthetic lighting guide, read Simon Sarris’ excellent guide to lighting a home. I recommend the extreme residential lighting approach only for studies and particularly dim living rooms.)

Consider the lumen

One very cool fact about the world is that we know how to assign numbers to brightness. The SI unit for light is the lumen, which roughly is used to measure how much light a source outputs. More relevant for us, however is the unit lux, which is lumens per square meter. In other words, lumens are a measure of emitted light, and lux are a measure of perceived brightness (i.e. what we actually care about).

I don’t have a great intuition for thinking about a single lumen or lux, but you can get a sense for it by considering the lux of common situations:

Scenario Lux Cozy night-time light 30-50 Indoor room 100 Warehouse/industrial space 500-750 Overcast day 1000 Full daylight (not direct sun) 10,000-25,000 Direct sun 32,000-100,000

(source: Wikipedia)

Surely there can’t be that much of a difference, you ask? It’s surprisingly unintuitive, but there is, because apparently we perceive light during the day logarithmically! 

This is one of the anecdotal reasons I’ve heard cited for why Season Affective Disorder happens – you actually get orders of magnitude less light during the winter than the summer. (This might also explain why British people are so grumpy all the time.)

If you want to test the light level of the room you’re in right now, you can download an app on your phone and measure it! I have no specific recommendation but this one works for iPhones. It will vary throughout your room, but you can try pointing it at the floor for a somewhat balanced reading.

I find it surprising that we are so okay with our indoor rooms being dark, relative to the outdoors. I suspect it’s a side-effect of how expensive light has been, historically. We’ve just always accepted a dim indoors as normal.

This famous 1994 paper was one of the first to collect data on how efficiently we humans have been able to make light, over time. Note the exponential increase over time! We got 5 times better at making filament lights from 1900 to 1990. And fluorescent lights are 5 times better than even the best filament lights in 1990!

Do you know what else is better than filament lights, and isn't represented here? LED lights. LEDs were a nascent technology in 1994. Since then we’ve learned to make LEDs that are 7-8 times more efficient than incandescent filament lamps.

More importantly, we’ve also learned to make high-quality LED lights. If fluorescent lights were so efficient in the 90’s, why didn't we use them more? I think we did, actually, but primarily in offices and hospitals – and what do you think of when you think of those environments? I think of horrible, sickly, overly-bright industrial settings. In the 90’s, we only had an awful light trade-off available to us: bright and awful fluorescent lights, or comfortable but dim incandescent light. (More technically, this is because fluorescent lights have terrible CRI. If you don't know what this means, I'll discuss in more detail later.)

30 years on, however, we can have both! 

I’m not quite sure why it isn’t more widespread, but there’s nothing to stop you from installing high-quality, bright LED lighting in your home or office today.

How much light is enough?

I am going to claim that you don’t actually want your room to be as bright as the outdoors. Or, more accurately, you only sort of want it to be as bright.

You can see from the chart above that ‘outdoors’ varies a lot. Overcast days, at 1000 lux, for me, are a little too dim. I personally find straight daylight (10k-100k lux) to be far too bright; when I’m outside in those settings, I reach for a hat and sunglasses.

I’ve found that I prefer to aim for about 3000 lux. You probably also pick a light level you want between 1,000 and 10,000 lux. Remember that your perception of light is logarithmic, so these are the numbers that correspond to what you would perceive as linear:

1,000 1,600 2,500 4,000 6,300 10,000

Going higher seems like nice optionality, but you start to hit some practical limitations around power and safety. This is another reason why I’ve found 2-3,000 lux to be the sweet spot.

Once you’ve picked a desired lux, you can use the calculator below to figure out roughly how many lightbulbs you’ll need. You’ll also need the dimensions of the room you’re trying to brighten. I’ve loaded it with default values for smart 800 lumen bulbs I prefer (I’ll go into the details on why later), but you can change those too if you like:

Room Lighting Calculator

Room Width (ft): Room Depth (ft): Desired Lumens/m²: Lumens per Bulb: Bulb Wattage: Bulb Price ($): Cost per kWh ($): Calculate

(If you're seeing this through your email, the calculator won't show up for you. Click here instead.)

The calculations it does for you are:

• Room width * depth to figure out how many square feet/meters we are covering

  • This isn't a completely fair calculation, because room height also matters – the lux produced by your light will depend on how much distance it has to travel to get to you. For our purposes I'm assuming your room is a standard 8-10ft high. I haven't thought too much about it, but if it's double-height I would suggest you double all the numbers above.

• Square meters * lux to figure out how many lumens we need

• Lumens / lumens per bulb to figure out how many bulbs you need

• Various other multiplications and divisions of the bulb numbers to determine total cost and power usage.

The other values I’ll explain in depth later, but roughly speaking wattage determines how you need to arrange the bulb(s) for safety and power reasons and kWh tells you how much extra electricity you’ll use per month with this setup. 

You might notice these aren’t exact. I’ve tried to round numbers where possible for simplicity.

Which bulbs?

The short answer is, Wyze White bulbs ($9 each). They have high (enough) CRI, are tunable and are affordable. If you have infinite money, buy Phillips Hue bulbs instead ($40); they’re pricey, but the Phillips smart system is by far the easiest one to set up. But if that makes sense for you, financially, you may want to look around for a better option.

Below I’ll go into why I prefer them.

What is light quality?

If you’ve made it this far, you probably (like me) are excited to learn that we’ve also learned how to assign numbers to light quality.

There are actually all sorts of numbers, and you can easily find all sorts of details about them on the internet. For our purposes, though, we only care about CRI; it’s usually one of the easiest metrics to find about a bulb. CRI stands for Color Rendering Index, and it’s important to get a bulb that is at least passable in CRI instead of terrible. 

(You can go down a rabbit hole learning how to maximize CRI and R9 and other measurements, but I think marginal effort applied there is less useful than effort applied to placement and color temperature and the other factors I’ll write about here.)

Here are two panels representing the same color swatch, photographed under different light. The top/left panel shows it under 100 CRI, and the bottom/right panel shows it under 24 CRI. Note the huge difference in color! The 24 CRI colors are completely washed out and/or pink.

24 is about the CRI of a street light; in reality you’ll almost never interact with consumer light at that level. However, it’s common for cheap LED bulbs to be 60, 70 or 80 CRI. This is usually marked on the box, or on the ‘specifications’ section of the online shopping page. Fluorescent lights are usually 50-80 CRI, though there are apparently some high-quality fluorescent lights now. (Fluorescent lights also sometimes have other color problems, which makes for beautiful night-time cyberpunk photography but are less good for lighting your office.)

The best LED bulbs nowadays are 95-98 CRI. (Incandescent light is 100 CRI.) From trial and error though, I’ve usually found 90 to be sufficient. Reddit is full of complaints about how CRI can easily be 'gamed', which is maybe why the super-high CRI bulbs don't always lead to a much better experience.

That was a lot of numbers, so here is this information again in table format:

Light type CRI Street light 24 Fluorescent light 50-80 Cheap LED 60-89 Wyze White bulbs 90+ Expensive LED 95-98 Incandescent bulbs 100

If you don’t want to buy smart bulbs for whatever reason, the best LED bulbs usually come from American brands like Cree and Feit. (Some random trivia I’ve learned about LEDs is that they’re very advanced technology, so manufacturers are usually located in Japan, America and Western Europe.)

Why smart bulbs?

This one took me a long, long time to figure out, even though it’s intuitive once you figure it out. You want smart bulbs because the ideal light experience changes a lot throughout the day.

A fact that you may already know is that white light comes in different colors, from cool to warm. The diagram above shows the different colors and how they correspond to lights you’re familiar with. Most of the light we see during the day is cool (5000K+), and most of the light we see at night is warm (2700K+).

This presents a problem: what color of lights should you get? 

Naively, you may expect 5000K to be the answer. Unfortunately, I find 5000K extremely painful and unpleasant to look at at night.

Is it better to get a warm light then? No; it ends up feeling very strange and unnatural during the day, even if my lights are very bright. 

I tried splitting it down the middle and using lights that are 3000K, but that still had the strange daytime effect, and they were also far too cold at night.

Eventually I realized that I wanted my lights to be extremely bright and 5000K during the day, and to dim (slightly) and change color temperature with the sun as it set. This is why I recommend using smart bulbs; they can do this automatically!

Note: The home automation systems can be finicky to set up. If you already hate them, you can instead set up *two* ‘dumb’ light sources attached to separate smart dimmers: one with 5000K lights and one with 2700K lights. Set the 5000K ones to be on during the day and dim around sunset, and have the 2700K lights turn on at night. If you choose this path, make sure the bulbs you buy still have high CRIs.

Placement

Let’s take a brief break, and look at this Vermeer painting together.

Obviously, this room is far too dark. It’s a great place to develop long-term eye problems. But the amount and shade of light filtering in through the window and softly illuminating the young woman is stunningly beautiful. This painting is a masterpiece for many reasons, but one of them is because of its portrayal of light.

Amateur photographers quickly learn that the best portraits are taken on overcast days, or of subjects sitting by a window. It’s because this light is diffuse and soft. 

If there’s one lesson here it's that using high-quality light matters a lot, but where it’s placed and how it’s diffused also matters a lot to your experience of it. You won’t be able to make all the parts of your study as beautiful as this Vermeer, but it’s worth keeping it in mind while you decide how you want to place your bulbs.

A whirlwind tour of the rationalist lighting ecosystem

Back to business. So you plugged in your information into the calculator above, and you’ve discovered that you need 24000 lumens. Using the Wyze bulbs I suggested means you need 30 bulbs.

“But I don’t have 30 sockets in my room to plug lights into,” you say. “What am I supposed to do?”

Never fear! Much ink has been spilled discussing lights on the rationalist blogosphere. Unfortunately for me, I only realized this after I started writing this article. If I’d known, I probably would have picked a different topic to write about.

At a high level, though, many people are trying to solve the same problem: making rented home offices nicer to work in. I’ve seen roughly 3 approaches to this:

• DIY single-point sources

• Lumenators

• Custom LED panel setups

Spoiler alert: I prefer lumenators myself, but I’ll go through the pros and cons of each of them.

DIY single-point sources

Ben Kuhn is the person who got me into this whole lighting hobby. About five years ago he wrote Your room can be as bright as the outdoors, where he details how you can create a 30,000 lumen bulb using corn lights and/or bulb splitters. For a 10x10 sqft study this produces 3000 lux, which is great!

However, when I’ve tried this, various people around me have objected, citing reasons like “it hurts my eyes” and “ouch.” Maybe you will have better luck! I also have found that I personally don’t like it because I think single points of lights diffuse light poorly. This means some areas are far too bright and others are not nearly bright enough.

That said, there are situations where an approach of this form can make sense. I once set up extreme residential lighting in a place that already had 6 overhead fixtures installed, and I was able to place 4 1,600-lumen bulbs in each of them using a 4-way splitter, for a total of 30,000-40,000 lumens. (Unfortunately, those weren’t smart bulbs, and it’s hard to find smart bulbs above 800-lumens of brightness.) It can also serve as a stopgap, or a test, if you want to see if adding another 3,000-8,000 lumens of light to your workspace will make a difference.

You need to be very careful around wattage if you do this (see notes at the end on fire safety). How careful depends on how many bulbs you have and how powerful they are. When I’ve set up lights like these in the past I’ve used:

• 1x these IKEA shades. Hang them from your ceiling so they are not in eyeshot. The shades are large, which is good because they will diffuse light and make it less painful to look at. 

• 1x this 4 in 1 splitter.

• 4x these Cree bulbs. (18w each for 2000 lumens. Multiplied by 4 it is 72w. These are not smart bulbs; I can’t find affordable 2000 lumen smart bulbs. You could also use the Wyze bulbs above, though then your light will only be 3200 lumens.)

• 1x this 100w ceiling socket cord.

Alternatively, here is a very bright light product someone has made and is taking pre-orders for. This also suffers from the ‘poor distribution of light’ problem, but I am impressed with the founder for trying at all!

Lumenators

My preferred approach is something known as a ‘lumenator’. Basically, this is a string of lights placed up on the ceiling, with bulbs installed in them. The link above provides a guide on how to make them, though I disagree with some of their choices!

My setup has two parts:

• 48 ft, 24-socket Commercial Grade String Light Cord (medium/E26 base) (https://partylights.com/48-Medium-Base-Stringer-Black/, rated for 960W).

• 24 Wyze bulbs (8 watts each, for 192W total)

• Hooks for hanging (finicky Command hooks if you don’t want to damage your walls, screw-in ones if you are ok with spackling)

You can also use these 10-socket lights if you want fewer bulbs.

Using the right light cord is important and easy to get wrong. It must match the base of your cords (E26, for Wyze bulbs, also the most common one), and have a power rating appropriate for the bulbs (these can handle 11-15W per socket; some light strands sold on Home Depot can only take 1W since they are decorative).

Bizarrely, there seems to be only one design manufactured anywhere for this type of fixture. I checked Taobao to see if I could find a different generic Chinese version, or one with less space between the sockets, and failed. If you need more than 24 bulbs, you should get two; you can plug them into each other, up to the wattage limits specified.

The Wyze bulbs already are covered by a diffuser which does enough, and they aren’t individually that bright anyway, so I don’t think you need the lantern covers suggested in the Less Wrong post.

Note: You may need to use a knife to cut off the ‘weather-resistant sockets’ to get your bulbs in. I didn’t need to for the Wyze bulbs but I did for some cheaper Amazon bulbs I previously tried.

Panels?

Lincoln Quirk suggests that maybe the thing we really want is ‘fake windows.’ I haven’t tried it. It seems like a neat idea in theory, but it requires electrical knowledge beyond what I know about LEDs. 

There’s definitely a space here for someone to create a custom LED panel. It would be neat. Then you wouldn’t have to do all this futzing around with the bulbs and the light strands and calculators. I once thought about trying to design and sell this, and then I decided that I wasn’t insane enough to start a hardware business.

Miscellaneous

One day I went around and hung pictures on all our walls. The general guidance is to hang pictures roughly so that they are at eye height. That night my partner said, “It’s very nice that you hung up all the paintings. But why are they so strangely low?”

“There’s nothing wrong with them!” I responded. “😠!”

Note that I am 5’3” (159cm) and my partner is 5’11 (179cm). This is largely intended as a reminder that what is above eye level for you might not be above eye level for everyone; and having a bright light source at eye level is extremely unpleasant.

For this reason I recommend putting your light as high up as possible. Hang it from the ceiling, if possible. If you must affix it to the wall, put it as close to the ceiling as possible. In the past I have put it behind wherever it is I’m working, so that I can keep it out of eyeshot. You’ll need to buy, borrow or steal a ladder to do this well.

tl;dr

This is too complicated. Just tell me what to buy.

Use the calculator to figure out roughly how many bulbs you need. Then buy Wyze White bulbs, command hooks and one of these for each 10 bulbs you need. For maximum effect, you will need to figure out how to control the Wyze bulbs using the app, which will be slightly annoying. You’ll also need a ladder.

I am a Serious Adult and I want a less janky setup. What should I do?

You mean you don’t want your house to have the aesthetic of a Berkeley group house? What a fascinating preference.

Honestly, I would recommend taking the lumenator or single-source approach first anyway; it’ll cost you $100-500, which is small potatoes compared to doing a proper renovation. It'll also help you figure out where and how you want your light to be. Just accept the snarky comments from your friends for a month. They might be too busy enjoying your well-lit room to notice anyway.

I have tried to figure out what the Actual Solution is here, and mostly failed. However I know it can be done, as I once worked in an office space that had solved this lighting problem beautifully.

A solution I commonly see in Asia — where I generally find the indoors to be much better lit — is recessed overhead lights. This is nice because then you can install very bright lights but not have them in your eyes, and the light comes out of the right place in the ceiling. Presumably you need to contract with a builder to get the right LED panels and controls for them.

I live in Europe/Asia/a part of the world that is not North America. What should I do?

Enjoy your urban, walkable environment and sunny apartments!

I apologize for the America-centrism. It’s the only place I’ve tried most of these approaches, so it’s the only place I can provide details. Most of it should apply, though you’ll need to find local sources for parts. The main thing to note is that the standard power requirements may vary, because you’re wired for 220V and not 110V.

Health and Safety

Will this cause a fire/blow a fuse?

I don’t think so. (This is to the best of my belief! I am not an electrician.)

Most consumer lightbulb sellers tell you all the numbers you need to know. Importantly, the wattage tells you most of what you need to know. If you stay under the wattage ratings suggested by your

Do be careful with your extension cords. The US is unique among developed nations in allowing consumers to purchase unfused extension cords with very low current ratings, aka extremely dangerous fire hazards. Try to buy fused extension cords, or try to plug your fixtures directly into the wall.

One risk you might be worried about is blowing the fuses in your wall. But your average household circuit can conservatively handle about 15A at 125V (assuming you’re in the US). If you multiply those together to get the wattage, and we see that it can handle about 1,875W of power. Your microwave, by comparison, draws approximately 1000W of power when it is on. So if you can run a microwave on it, you’ll be fine.

If you want to go crazy and output 10k lux — or if your room is particularly large, and you only have one circuit — you will start to run into problems. But it turns out that LEDs are just so efficient, and our home circuits designed for inefficient incandescent bulbs, that you are effectively now using your home almost exactly in the way it was designed for. If you have concern, consult an electrician.

How much will this cost me in electricity?

A fun fact about North America is that electricity prices vary wildly across the continent. In the Canadian east, which imports most of its power from Quebec’s abundant hydropower, electricity can be 5-10c per kWh. In California, a state that famously doesn’t believe in capitalism, it can run to about 40c per kWh (I kid you not).

Your 100W light, if you make very odd choices and keep it on all day, will draw 100W * 24h = 2.4kWh per day. (A kWh is a kilo-Watt hour, used to measure the use of electricity over time. It represents the continuous draw of a thousand Watts over an hour). Assuming there’s about 30 days in a month, this will cost you about 72kWh, or about $8 to $36, depending on where you live. In practice, I’m guessing you would only keep it on half the day, so you can cut that number in half.

This is expensive, but consider that it trades off against the cost of therapy and being sad. Given that I think it’s well worth it, but you can make your own choices. Both of these numbers are produced by the calculator above.

Thanks to Ben E and Ross R for feedback on an early draft! Special thanks to Ben Kuhn for careful proofreading and substantive feedback.

Recently I've found myself making not-very-funny jokes about EA. 

A few weeks ago, I moved, and two of my friends came over to help me unpack books. While doing so, we found a collection of the EA classics: Doing Good Better, 80,000 Hours, an advance copy of What We Owe the Future that I'd accidentally stolen from a friend. Maybe The Precipice, too.

My friend asked, "Do you want these in the living room, or on the shelf hidden away upstairs?"

"Let's put them upstairs," I said. "I don't need the EA propaganda machine in my living room."

Around the same time, I wrote about “Big EA”. Someone asked me what it meant. I was using it as a tongue-in-cheek reference to the Berkeley/Oxford EA communities and the grantmaking institutions and non-profits around them (CEA, OpenPhil, Effective Ventures, etc.). But it was a fair question – the definition is not at all obvious.

Where did all all this cynicism come from? 2022, of course.

Remember those times, the summer of love that we’re all still slightly hungover from? In those halcyon, just-post-COVID days, Effective Altruism was everywhere. Suddenly, it seemed that everyone I knew had been EA all along. A surprising number of my friends altered their life plans to go work for EA organizations or do EA fellowships. The movement had gone mainstream: in the pages of the New Yorker, in Time magazine, all funded by ZIRP money. What We Owe The Future was playing in every bookstore. Nonprofits barely took calls from mere deca-millionaires. Ivy League students uprooted their lives to move into group houses in Berkeley, dropping out of college to start non-profits.

I, a perennial EA-adjacent, wanted to know what was up. I'd always thought EA was onto something with the malaria nets and cash transfers. But I wasn’t a longtermist, or worried about AI risk. I wondered if maybe I should be. What was I missing?

When I asked my very smart friends why they were so worried about everything, I overwhelmingly kept hearing the same answer. "I sat down and thought really hard about it, and I realized I should be,” they said.

Every time I heard this, I wanted to beat the speaker over the head with a paperclip plushie. What?! Where was the second order reasoning, here? What was the probability that everyone would come up with the same conclusion, through the same process? Shouldn't we be very suspicious of this on the meta-epistemics alone?

That phrase also implied that if I tried enough, I would also reach that conclusion! Clearly, I didn’t agree with the EA mainstream because I wasn’t trying hard enough or, worse, smart enough. As if the process of deciding how to do the most good in the world was as simple as some sort of mathematical proof!

To clarify, I don’t think my friends ever actually meant to imply that. They were just using an idiom common in the culture. (For a similar, but not exact, example of this line of thinking, consider this comment on the EA Forum. I’m not the only one who has been disconcerted by this turn of phrase.)

Nonetheless, the choice of words reflects that culture. And to me, something about this EA methodology stank. I won’t do a good job of describing what here – it’s not my main focus – but if I had to hand-wave at something, I’d say that I didn't like the assumption that thinking hard enough was enough to answer a difficult moral question, especially when all the people doing the thinking came from roughly the same elite, analytical Euro-American background. Reasoning depends on its axioms, and it felt to me that longtermist EA had limited awareness of what axioms it was depending on, because they were so commonly shared, so much in the groundwater.

I don’t want to emphasize this too much, as more mainstream critics do. I mostly think it's an issue that, for instance, few analytical Euro-Americans types have lived in a country with a sustained, high growth rate. Few have had neighbours in grinding poverty, or shared a roof with someone who supports a family on $500 a month. My complaints here are not "social justice" ones. I'm a fan of elite Euro-American analytical cultures, and EA’s focus on analysis is its relative strength, compared to other social movements. Sometimes EA just likes to take it too far.

There still was clearly something good here, and I wanted to find it. But it troubled me that when I tried to sift through and figure out what parts of EA I did like, I struggled to find anything. In 2022 EA, it felt sometimes like the only important thing was what other people thought of you. Which surely couldn’t be right.

Of course, every community has status dynamics. One thing I find it helpful to do, when debugging a community, is to ask myself, "How can an outsider come in and quickly become well-respected?" From this you find what it is that the community values, and what it is really about. In sports, it’s winning. In finance, it’s making money. In Silicon Valley, it’s developing new technology.

EA in 2022 felt so insular and circular, I couldn’t pin down what it valued. Working on AI alignment or community-building around AI alignment made you cool, I could tell – but everyone freely admitted that the field was in its early stages and that it was hard to assess who was really being successful in it. Being a successful community-builder worked too – but that was even more self-referential. Being an OpenPhil grantmaker definitely did it too, but that was because they funded all the projects that everyone else was working on. On what basis did they make their decisions? Surely not just vibes?

I wasn’t the only one who felt this way. It was systemic. Everybody thought they weren’t cool enough, there was an inner ring that they weren’t in. Even Dustin Moskovitz, the person who single-handedly funds half of EA, was told that he wasn’t a ‘core EA’!

How did a group of people dedicated to doing the most good in the world end up like this? And why were so many people — including me — drawn to it?

If you're reading this, maybe you were also there for this. You also know how it ended.

EA has become fun to kick in the last year, now that it's a wounded dog. Sometimes I wonder how people manage to find this much vitriol for it. Have you met the EAs? They’re such nice people!

Maybe people felt that EA played motte-and-bailey with malaria nets and longtermism. Maybe it’s that there are few things as harsh as the contempt we have for those who have shined brighter than us and then fallen from grace. It probably doesn't help that the implicit message of EA is an unforgiving one: "You aren't altruistic enough. You are not talented enough. You are not good enough, and you are not doing enough to save the world." 

Everybody who seriously considers the effective altruist cause has to grapple with this, and face the unending amount of suffering in the world. You can look at it and despair, but that way madness lies.

Instead, the EAs grapple with it, and try to figure out what is enough – and what they can do. And so at the core of effective altruism is a social movement focused about what you can do, today, to make the world better. It starts with the malaria nets. You, yes, you, can donate $3,000 and save a life. Right now! On the internet! All you have to do is click here – it’s not a scam, I promise. From the comfort of your living room, today, you can statistically prevent an African child from dying!

It’s okay. I haven’t given the Against Malaria Foundation my money, either. Maybe you’re an adult with a car payment and a mortgage and children to feed. In our world, your property is yours, and you don't owe it to someone you have never met. The EAs are quite alright with that.

But remember children and teenagers – university students, even – are more ideologically pure, kinder than us, and raised in a world where they have little say in what happens. To realize that there is a path to doing good that is so simple yet meaningful? That is a powerful thing.

Compare it to two other social movements that young people today are drawn to: the environmental movement and social justice causes. To me, these activist movements focus overwhelmingly on fighting the system.

Imagine the climate change movement. What do you think of? I see Greta Thunburg shouting in the United Nations about how the older generations are ruining the world for those of her age. Or people gluing themselves to subway trains, shutting down key infrastructure because they don’t know how else to amplify their voice.

The means for action that the environmental movement gives its adherents are nonsensical. I cringe when I see people separating recycling into streams which are largely dumped anyway. It breaks my heart when people decide not to have kids, because they worry about its impact on the environment.

It's imbued with this sense of hopelessness, the idea that if we plead to our parents or teachers enough then the adults will step in and fix the problem. But a powerful person doesn’t protest, or sort their recycling and hope for a gold star. A powerful person plans, prepares and acts.

The central conceit of EA – the part of it that, for some reason, drives people nuts – is that it dares to ask: what if there is no system oppressing you, just thoughtlessness? What if the problems in the world come not from villains, but from an uncaring void in the world that simply doesn't care? What if the way to fix the problems in the world is not to protest and rail against them in the hopes, but to acquire the technical, political and operational skills you need to solve them yourself, and then actually go and solve them?

The best projects that I have seen come out of EA are inspiring examples of this mindset. Consider Asterisk, one of the best intellectual magazines in print today. Or Lead Exposure Elimination Project, a nonprofit that has been called one of Founders Pledge’s “most cost-effective charities”. And, unless you’ve been living under a rock, you’ve heard of Anthropic, one of the current darlings of the tech world, a product of the EA attempt to solve AI alignment.

(As an aside, this is also why I find the apocalypse cult of AI doom so repellant, as exemplified by Pause AI or MIRI’s death with dignity strategy. It's a natural conclusion of some EA lines of thinking, but it is the most nihilistic one.)

For all of its problems, 2022 EA took this to an extreme. Old EA focused on earning-to-give and buying malaria nets. 2022 EA had learned about the limits of what money could buy, and shifted its focus. New EA was about the importance of being agentic.

In a world that has been shaped by our elders, where technological progress has stagnated for fifty years and our lives seem overly shaped by faceless bureaucracy, where most young people can’t buy a house and we’ve given up on putting humans on Mars, it's easy to lose hope that we can change anything. Many of us have; I see it often when talking to non-EA friends.

EA is not the first community to think that small groups of motivated people can do great things, but it is one of the first to apply it to the idea of doing good, and the first to send missionaries throughout the West's elite institutions to proselytize it widely. Paired with its penchant for learning deeply about cause areas, and applying basic mathematical analysis to its work, it's a potent force for positive change.

Its ideas are sometimes crazy. I can’t quite believe that shrimp welfare is important, despite some well-made arguments in its favour. But you need some level of craziness to challenge the status quo, and the world could use more of this brand of crazy. EA’s anti-authority nature can drive people up the wall – I know it has driven me up the wall. But it is the natural outgrowth of a culture that thinks the world could be so much better than it currently is, if we tried. 

So that’s EA, I guess. The bad and the good. What now?

I’ve come to believe that EA is at its best as an incubator of ideas. At its best, it gives people the funding, framework and confidence to start odd ideas that otherwise wouldn't happen. EA should still do this! 

But EA funding is a double-edged sword. It's granted to people who otherwise wouldn’t get funding, for ideas that seem crazy when proposed. This makes those organizations, and their founders and employees, very vulnerable to EA groupthink and its status dynamics, because they are completely dependent on it. And this makes those organizations more likely to be focused on cool EA ideas, and less focused on what is actually good for the world. 

Like most movements, EA is at its worst when it is overly focused on itself. In 2022, EA was so all-powerful and rich in funding that it seemed it would never need to interact with the outside world again. This was the part of Big EA that I found so dangerous: the cult that drew in talented young people and sent them to work on problems of dubious efficacy, with complete disdain for any received wisdom from the rest of the world. It seemed to have reached escape velocity, just before it came crashing back down to earth.

If I could change one thing about EA, I would change a standard part of the grant-making process. I would ask every grantee how it expects to raise from outside fundraisers eventually. (This should also help with the Girardian terror and the status dynamics. The mimetic pressure will come off when not everybody in EA is trying to raise from the same set of organizations and hiring from the same talent pool. Is it surprising to you that the EA organizations I like most, and find most effective, are ones perched on the border of EA and some other subculture?)

Of course, there’s no reason EA should change itself, and it probably won’t. But I don’t worry too much about it. More and more, I find when I'm describing someone I think will do great things, I describe them as “post-EA.” 

Maybe there was a time when EA was the best subculture around for young people who wanted to improve the world. Many of those people now are very familiar with EA, wary of it and yet inspired by it. Slowly, I see many of the best lessons from EA taken up by other movements. I see glimmers of it in progress studies, the abundance movement and even in today’s Silicon Valley, where the cool factor has shifted from enterprise SaaS to AI and hard tech. (While those movements are great, they don’t seem to have community bonds as strong as EA's. I don't know why. I worry sometimes that they won’t work as well.)

For all of its problems, EA taught us how to take the hardest problems in the world and work on them, that it is worthwhile and possible to do so. It brought people together and set them on one of the hardest and most intractable problems in the world: suffering. In doing so, it gave us the methods and motivation to attack many other hard problems: regulatory capture, political polarization, climate change, economic stagnation. Whatever else it does, it is important and good for that reason. It's why I’m glad to live in the world that EA has made.

Thanks to Eric Neyman and Ross R for extensive feedback and Holden L, Rebecca G and Aditya G for discussion. Thanks to Rachel B for discussing an early version of these ideas with me.

Tanner Greer shared several of his insights into EA in a small group discussion that I was a part of almost a year ago. I've tried not to repeat them, but they've been an influence. In particular, he pointed out to me that EA is an ideology largely marketed towards elites.

10 years ago, I started my freshman year of college. This is the advice I needed to hear, not the advice you need to hear. In fact, some of it may be actively bad for you. See Should you reverse any advice you hear?

1. You are overly obedient. You not only do what people tell you to do, but find it hard to imagine any world other than the one they present to you. Spend more time thinking about what you want, in isolation from the pressures of the world. (Keep this in mind while you read the rest of this very prescriptive document.)

2. Fortunately, you’re usually able to recognize and extricate yourself from situations that are bad for you. Unfortunately, it often takes you about twice as long as it should.

3. There will be quiet periods of your life. They will feel slow and low in potential energy. The best use of them is to work on cool projects and share them.

4. Write more, on the internet, in college and after. Not only will it increase your surface area of luck, it will also help you form a better idea of who you are.

5. Don't do things that feel bad. This is not as easy as it sounds, because you are a two-marshmallow kid.

6. Ideally, you wouldn't drink at all. Since we both know it is impossible, try to keep it to once a week.

7. When you are bored at work, don't snack. Instead, go outside for a walk in the sun.

8. New York is an extremely hedonistic place, especially for a young professional. You'll eat, drink and party too much. It will be fun, and you will be correct in feeling that there is more to life than this.

9. Despite what you think, money is not the constraining factor on your ability to do things. If I had to guess what it is, it's probably your social environment (your friends and mentors) and your creativity.

10. Do a quantitative trading internship. Despite the fact that the trading interns seem obnoxious and scary, they are not smarter or better than you.

11. You currently have high intelligence and low agency. Be aware that you are naturally drawn to people like you, but where possible, try to find and spend time with intelligent people who have high agency.

12. Join an a capella or dance group in college.

13. You are right to be suspicious of Big EA.

14. Despite what the Europeans will tell you, finance is a fine discipline, and you should become an expert in it. It is the system of money, and it is interesting and well worth knowing for that reason.

15. Everybody in finance has a number. For almost everybody that number goes up over time. What makes you think it'll be different for you?

16. You dream of living in a sunny home where you can go downstairs and get a coffee at a cute shop around the corner. You are correct that this will make you happy.

17. You like Brooklyn brownstone apartments more than fancy luxury apartments because they are in point access blocks and have windows on two sides.

18. Live in the same neighbourhood as your friends.

19. Few experiences are as joyful as cycling around a beautiful city on a sunny day.

20. At some point, you will face a choice between work and the love of your life. Pick the boy.

21. At some point, the love of your life will face a choice between work and you. If he doesn't pick you, dump him.

22. Don't fall in love with something that can't love you back. Companies do not spend a week crying when you break up with them. This is also true of many boys.

23. People will occasionally make unsolicited, snide and rude comments to you. Ignore them; this is usually a sign that they are deeply insecure. If they are in a position of authority over you, get away from them.

24. You are more inclined to exit than to voice. It is better to do one than neither, but know that leaving is always expensive.

25. You are not direct enough when criticizing people. 

26. You get angry when the world is unjust. At some point, you will be taught that it is never okay to show it. That's the wrong lesson, though a common one for young women. The right one is that you shouldn't spend time around people who don't understand and appreciate why you are angry.

27. You also become angry and sad because you let people push past your boundaries. Stop letting them. Often, just saying 'no' is enough.

28. You are high in neuroticism, a trait that correlates with worse social relationships and an unhappier life. Sorry; there's not much you can do about it other than be aware of it. Its saving grace is that it means you are attuned to what is wrong in the world and driven to fix it.

29. At lot of your work will involve taking some flow in the world and optimizing some property of them: latency, throughput or bandwidth. To do this, you must figure out what limits them and then remove that bottleneck. This sounds simple, but you will not really understand the depth of this discipline until you work with a master of it.

30. Start tracking hours spent on deep work. It is an obvious and very visible metric of how much work you have done.

31. If you enjoy interning somewhere, but the person you like most announces during your internship that they're leaving the firm, think hard about what message you should take away from that.

32. The most important thing to consider when picking a job is who your manager and team will be. The company is only important insofar as it interferes with or supports your ability to work with your team. Conveniently, the most prestigious and well-known companies are the best at getting out of your way.

33. A $25 Airtag will save you 10 minutes a day of looking for your wallet.

34. Sometimes, you will worry that things feel too pleasant and stable and that you are in a rut. Don't worry; something will jolt you out of it eventually. Enjoy the good parts while they last.

35. Status exists in all domains. Status in areas you don't care about is useless and mind-corroding. The best way to accumulate it in areas you care about is by doing cool things.

36. Nerd cultures have status dynamics, too. In fact, they are often stronger there because there are few defenses against it.

37. Your inbound opportunities are worse than the ones you carefully select and nurture over time. This is true both if you are an engineer looking for a job and if you are a 19-year-old girl looking for someone to date.

38. Be suspicious of people who fawn over you with affection for no reason. They usually have a system for finding people like you, and they will grind you up in the cogs of their machine. This is true both if you are an engineer looking for a job and if you are a 19-year-old girl looking for someone to date.

39. It is not bad to be in a transactional relationship as long as you get something out of it.

40. Help people. Some of them may not remember, but you will feel good about doing it anyway. Some of them will, and it will create a strong bond between you.

41. When you are in college, there are so many potential friends around that it is correct to filter out people who are hard to contact. Once you are older, this becomes much less true.

42. The best reason to move to New York is because all your friends are moving there too.

43. The best way to make friends is by working on a large and intense project together. This could be a conference, a camp or a startup.

44. Try to visit a nice, suburban house of an American friend during college. You will not understand the country until you do this.

45. Learn to drive, as soon as possible.

46. Being on an F-1 visa does not constrain your ability to work on a startup.

47. In 2020 and 2021, mass hysteria will sweep the world. (There will be a virus, too, but since you are a healthy 25-year-old and don't know any over 50, it will not notably affect your life.) Realize that nobody at any large company will get any work done during this time. Don't throw effort into work during this time, or you will burn out.

48. Stick it out to get your 2020 bonus, then seriously consider leaving your job and moving into crypto. It's a degenerate and trash subculture, but it's one of the only high-agency environments you have access to in 2020 and 2021. If you don't do that, at least switch teams.

49. For the love of God, buy Solana at $3.

50. You will be deeply unhappy in any job that requires you to sit in a windowless box from 9-5 all day. Fortunately, very few good jobs require this.

51. You should aim to produce things at about a 1:10 ratio to how much you consume them. This is true of food, videos, internet comments and parties. Realistically, you might achieve 1:100 or 1:1000.

52. Get into the habit of making mid-term or long-term stock bets of about .5-1% of your net worth. This will never make you rich, but it will inform you about how much you know relative to the rest of the world.

53. 100% VOO, after an emergency fund, is a fine allocation of savings for a young professional with no family obligations. Interactive Brokers will let you borrow against it on margin if you need temporary liquidity on short notice.

54. Most advice is given by and directed at people who have less education, money, technical ability, and social capital than you. You should consider it accordingly. Do note that I have more of all of those things than you.

55. Don't trust people when they say that they are experts and they know what they are doing. Ask the questions you need to ask to understand at least the high level of what is going on.

56. People who lie to you about small and unimportant things will lie to you about big and important things.

57. Most managers lie.

58. Mathcamp is a strangely good selector for people you like. You should not spend 100% of your time with Mathcampers, but 10-30% is a good ratio.

59. Avoid makeup. Wear imported Korean sunscreen every day.

60. Take a week off every year and go to Burning Man. It will change your conception of what is possible in the physical world.

61. You will understand a person when you understand what it is that they want most in life.

62. Get a coach to teach you proper form, then lift weights 2-3 times a week. That is all you need to do to stay healthy.

63. Try very hard not to gain weight. Once it goes on, it is hard to take off.

64. Grow a garden. It will teach you about patience and consistency. It will also teach you about the value of automation, and a useful amount about plumbing.

65. Programming is not the only technical skill worth knowing.

66. Spend lots of time posting thoughts in good group chats. A tight-knit Discord server is great for this, but it is hard to kickstart.

67. People will tell you that it is good and right to be Scotty. They are wrong; you should be like Kirk.

68. Do some wet lab biology work in college. Also, take a campus job that requires you to do sales.

69. Take only one hard class a semester. Really. You'll learn more, have a better time and get really good grades.

70. Professors are extremely variable in quality.

71. Start trying to understand the world through Fermi calculations. Memorize important numbers. This is an extremely useful skill.

72. Try to make ambitious and smart friends who do cool things. The best way to do so is to be smart and ambitious yourself, and do cool things.

73. It is important to do some on-call/production support, but you are very good at it and will be asked to do it more than others. Say no. It will take away from your ability to develop more valuable skills.

74. Every month, write an email to your friends around the world. This is the best way to stay in touch with people who you may not see again for years.

75. Bookmark and catalogue good essays on the internet.

76. Organize your cables by putting them into labeled ziploc bags. Put the ziploc bags in a box and label it 'cables'. Now you will have a cleaner desk cabinet and infinite cables.

77. The best work comes out of great teams.

78. You will never live in Europe. That's OK; it is better to visit than to live there.

79. Asia is not a good place for software engineers.

80. Run essays through Grammarly before you publish them.

81. Good work requires much more than just grinding. You need to be creative, well-connected and communicative, too.

82. You will learn the key parts of what can be generically taught about software engineering in a year and a half. There is much more to learn, but you will need to teach it to yourself or seek out specialized teachers.

83. People show you much of their character in the first few days you know them.

84. There is a part of you that will always rebel against the world you are a part of. Keep it alive; it's usually right.

85. When you are young, it feels like you will cycle through hobbies and ways of life quickly. Don't stress about it. At some point, you will slow down and develop a routine.

86. You are more like a plant than you realize. Drink lots of water and spend time in the sun.

87. If you find yourself spending a lot of time in the bathroom on your phone, it is because you work in a large, open-floor office all the time, and it's the only place you can get away from the chaos. Consider changing your lifestyle so that you don't need to do this.

88. You work better and feel happier immediately after you drink coffee. You will crash a few hours later. A pill of tyrosine can do wonders in the mid-afternoon.

89. Avoid carbs and eat lots of protein.

90. You only need three pots and pans in your kitchen: a wok, a cast-iron pan, and a Dutch oven.

91. You cannot work and travel at the same time. Don't try.

92. Hang out with your friends at least three or four times a week. Best if it's every day.

93. You don't need to be in your employer's office to do good work. You will feel great if you spend your days working in a room with your friends, even if they're working on something completely different.

94. Don't work with people who you don't technically respect. If your employer keeps insisting that you do, quit.

95. The best way for you to be happy with your work is to control who you work with. The way you do this is by deciding who your team hires and fires.

96. Consider deleting as often as you consider adding. This holds true for code, furniture and obligations.

97. The trick to surviving a big institution is to find your group within it. Then ignore everything else.

98. American culture, despite many words to the contrary, is built around self-segregation. You will be happier once you stop fighting this.

99. In college, try to make a cool app that all your friends want to use.

100. I would trade half my current net worth for a world in which I had a stronger community of friends and had worked more on my own projects instead of someone else's. Consider that when you make your choices, on the margin.

101. The next decade of your life will be a rollercoaster, but it will also amazing. You will make friends who will become family, find the love of your life and get really good at your craft. I wouldn't swap places with you. But I've had a great ride and I hope the next decade of my life is as good.

Thanks to Aditya G, Rebecca W, Holden L, Betsy P and Rebecca B for discussion and feedback. Thanks to Trevor C for the conversation that inspired this post!

If you want to know more about any specific point, send me an email!

I’ve recently come up with an idea that I like to call the Baked Goods Theory of social interaction. It goes like this:

If you don’t have at least one event a week where you can give away a tray of baked goods, then your social life is unsustainable.

What sort of event qualifies? You need to use some common sense for this, but here’s some guidelines that I think apply:

• it should involve between 8-30 people.

• it can’t be a commercial activity (it’s hard to give away brownies at restaurants or out at the club!)

• it must include people you see on a semi-regular basis. You should have some reason to like them. (Most people would not give away baked goods to strangers.)

• it has to be in person (virtual cookies are great[1] but don’t count.)

[1] Despite the EU’s claims otherwise.

Other than those requirements, though, the event proscriptions are fairly loose. Loose, weekly community gatherings qualify. If you go to four different events every Sunday, each of which repeats monthly, you’re probably good. Even the action of simply going into a mid-sized office once a week counts.

There's an important additional clause here that's subtle. The group should be a place where every person can contribute, and where those contributions change the nature of what the group of doing. Lectures, or large gatherings like parties, don't quite work, even though you could bring a tray of cookies to those events. This is why there's an upper cap of about 30 people. I think the baked goods analogy breaks down a bit here, but a friend of mine points out that maybe you don't want to hand out baked goods to a bunch of strangers anyway.

Why baked goods?

Many baking recipes call for you to make far too much food for a person to consume in one sitting. A cookie recipe will often make 24 cookies, and a brownie recipe will make thousands of calories of brownies. Most of the time, it’s too much for a family to eat over a week. So despite the fact that baking itself is a solitary act, it’s surprisingly social. You have to know people to offload give your baked goods to.

(You don’t actually have to bake a tray of cookies. You just have to know that you could.)

Why does this matter?

The key word in this theory is sustainable, a trendy green word that I’ve misappropriated for my own uses. ‘Sustainable’ means that you can keep doing something for the foreseeable future. Here’s an example familiar to most: if you’re broke and spending down your savings, your life isn’t sustainable — you’re going to hit zero at some point, or your credit card will cut you off (if you choose to go negative).

The notions of saving and income exist in social networks too. When you spend lots of time socializing with people, you’re investing in your social future (and reaping its rewards, too!). You’ll naturally meet people, and you’ll spend time with them and their friends, making more friends in turn.

For most people, high school and college are natural places to accumulate social capital. Later in life, though, it’s easy to start spending down social capital without ever re-accumulating it. If you started off with plenty of friends, you’ll probably be fine — for a while.

But modern social lives have a natural and surprising attrition. You’ll get in a fight with your friends, or they’ll start dating someone you hate, or they’ll move across the country. Once you get older, they’ll start literally dying.

And, if you’re not careful, it will be hard to make new friends. Many of my best friendships started as weak ties with people I saw regularly but infrequently. Over time, we started spending time with each other, and those regular gatherings helped provide regular touchpoints for our relationship until it was strong enough that we just hung out independently anyway. But you need to have people in all stages of the pipeline — it takes five years to make a friend you’ve known for five years.

This is where your baked goods social group helps: it’s a natural source of weak ties. And yet it’s worth saying because it’s far too easy, especially today, to become a social shut-in. My modal archetype of this is a white-collar remote worker who lives with their partner, in a suburb they moved to in order to be able to afford a home.

I’ve noticed that I like to phrase this theory in the negative (if you don’t have this then things are bad) instead of the positive (if you have this, things are good). I think it’s because it expresses a state of emergency — if you aren’t seeing a group of weak ties regularly, things are bad!

I myself was in this bad state when I moved to San Francisco recently. I saw plenty of people, but it was only because I put a lot of effort into going to social events and arranging social dates. Those efforts helped kickstart my socializing, but it would have taken far too much energy for me to maintain that all the time. It wasn’t sustainable.

Then I joined a local community choir, where one of the tenors occasionally brought baked goods to rehearsal. Realization struck. Rehearsal was a good thing for me, and a source of easy weak ties that I’d been missing since our move.

Our choir is on hiatus, and when we resume we’ll be on a once-a-fortnight schedule. But now I know that I should try to fill the rest of my weeks with at least one sort of regular gathering.

A lack of sustainability, itself, isn’t always bad. Sometimes life changes happen. Maybe you need to move across the country to take care of family, or you’re taking a year off to travel the world, or you’re putting your head down for a few months to work. But I think it’s well worth it to pay attention to the alarm bells if your life keeps constantly being like this.

Some notes on community

If you feel like it’s hard to find such a community, you’re not alone. Tanner Greer has written about the fall of the community-based organizations and the impact it’s had on our ability to shape the world.

As an example, he discusses how during the Civil War, women’s communities organized across America to create nursing organizations, which fed and sheltered soldiers. During modern crises, it’s unlikely anything like that would happen — we’d probably all stay at home, or wait for the government to tell us what to do, or spend all day shouting on Twitter about what we think should happen. (Isn’t that what happened during COVID?)

I suspect there is also something good and heartening about being part of a community that’s gathered to work on a common cause. Nowadays, those causes are often singing songs together or playing football. But there’s no reason it can’t be to plan urban design together, or to organize voter drives, or to build and maintain a local makerspace.

Have you ever seen Exchange Place in Jersey City? Exactly opposite of New York’s Financial District, across the East River, it’s an almost comical sight. It looks like an extension of Manhattan, full of gleaming skyscrapers slammed right up against the water. It’s a busy urban hub, home to Goldman Sachs’ secondary office, a walkable downtown neighbourhood, and a cluster of residential towers squeezed right up against the ferry terminal.

For a certain type of commuter, it’s a dream. Someone living in Exchange Place, the area of Jersey City situated right against the river, has three direct paths into FiDi. A PATH train can take you to World Trade Center in four minutes. If you want to commute without going indoors, a direct ferry runs once every seven minutes from Paulus Hook to Battery Park City. The journey takes only seven minutes, making your worst case travel time about fifteen minutes.

It’s one of the most affordable places that a commuter into lower Manhattan can live. The neighbourhoods around the Financial District – Soho, Tribeca and the West Village – are some of the most expensive in the country, and the specific geography of Manhattan makes accessing the Financial District from cheaper neighbourhoods inconvenient. (On top of that, a neat interstate arbitrage means that Jersey City residents can make New York City salaries and take 3% more home than their colleagues after taxes.)

The PATH’s inconvenient late-night service means it's not perfect for everyone. Young yuppies happily trade off those delicious tax advantages for a decent nightlife. But it’s a wonderful, more-affordable suburb for families with young children or those who want to spend quiet evenings at home. And, importantly, its existence makes housing in other parts of New York cheaper for those who would rather live closer to parties.

If you squint, San Francisco looks like Manhattan: its financial district is right against the water and inaccessible by land. Right across the Bay is a vast expanse of the mainland, connected by train. So when I look across from Embarcadero to the other end of the Bay Bridge, I somewhat expect to see a Jersey City-like built-up area.

But there’s nothing there. And when I take the BART and stop in West Oakland, a little part of me looks out and cries out in despair for what could be. Indeed, why isn’t it a vibrant, metropolitan commuter suburb? And it turns out that that it used to be, but it was killed by the highways. The real mystery, to me, is why it hasn't come back yet.

Why West Oakland?

West Oakland has the best BART connectivity in the Bay Area. It is the only stop in Oakland does not require you to run through downtown Oakland to get to San Francisco. Every transbay BART train runs through it.

Compare it to MacArthur, an Oakland area earmarked for development. In the middle of the day on a weekday, there are 8 minutes between trains, and it takes 16 minutes to get to MacArthur, meaning that in the worst case, a MacArthur-to-Embarcadero trip takes 24 minutes.

From West Oakland, trains run every 3-4 minutes to Embarcadero and the trip takes 6-7 minutes, making the worst case travel time 11 minutes. This means a West Oakland to Embarcadero takes 15-to-20-minutes, compared with 30-to-40-minutes from MacArthur, once you factor in time to get in and out of the station on both ends.

(MacArthur is actually not that far from the entrance to the Transbay tunnel, but trains going into San Francisco have to divert and run through downtown Oakland, adding ten minutes to the overall commute. In fact, in 2001, BART proposed building a bypass from the Transbay tunnel to MacArthur, which would improve throughput and commute times throughout northern East Bay. Combined with a second transbay tunnel, it would effectively double BART’s capacity. Obviously, it was never built.)

Every East Bay BART commuter passes through West Oakland, every day. All you have to do is look out the window. Immediately outside the station is a block of four-story affordable apartments. But beyond that are empty lots, abandoned stores and blocks upon blocks of single-family houses. There's little commercial activity, and, separately, crime rates are high. The neighborhood's prices reflect it: lots of land right next to the BART station sell for $200k.

It wasn’t always like this. The simple fact of geography that I noticed has made West Oakland a highly desirable area for almost a century. In the 1800s and early 1900s, American rail companies built cross-continental rail. Those lines terminated in West Oakland. This made the area a massive rail hub, and working-class immigrants from Ireland, Greece, and Italy moved to work for the train companies. West Oakland in particular was a company town for Southern Pacific, the dominant railroad company in California.

During World War II, it became a center for shipbuilding efforts on the Pacific Front. Black Americans from the South were drawn to the economic opportunity and moved en masse into the neighborhood. This made it a vibrant place — West Oakland was eventually known as the ‘Blues capital of the West.’ Black-owned businesses filled the streets, and a young shoe-shiner remarked that when he shined shoes, he shined the shoes of Black preachers and lawyers. Seventh Street was its main street, a bustling thoroughfare of butcher’s shops, groceries, and nightclubs.

This documentary provides a lively depiction of life at the time, as told by former residents. Ironically, it's funded by the California Department of Transportation and the Federal Highway Administration. It features interviews with porters, with women from the South who moved up to work on ships, with people who used to walk by the neighbourhood stores and people who used to go to the jazz clubs in the area.

What happened? If you’re American, it’s probably obvious. But it often shocks foreigners — like me — how much of America’s current urban landscape was created and not emergent.

A naive observer might think that the low-rise strip malls and dark underpasses that fill America result from humanity trying and failing to fill a mass of land too large for it. In some parts of the country, perhaps. But all too often, those strangely abandoned urban areas were razed and destroyed to make space for automobiles and highways.

That’s what happened to West Oakland. In the 1950s, its prime location made it an attractive target for the Federal Highways Administration. With the mandate of creating easy access from desirable suburbs into cities, it was granted huge powers of eminent domain to take land, tear down the houses on them, and put up suburbs.

In line with policies at the time, those highways tore ‘white roads through black bedrooms.’ West Oakland was a redlined neighbourhood. It was in a geographically ideal area. It had working-class population, with a largely non-white population, that couldn't advocate for itself. It was the perfect target for 'urban renewal'.

And so the FHA put up the Cypress Freeway right through Seventh street, introducing access to the Bay Bridge for suburbs in the north, and tearing the neighbourhood into two.

The neighbourhood limped on until the introduction of a BART station in the 1960s. Former residents describe this as its death knell. I’m not sure it needed to be — Asian cities provide prime examples of how train line access, even those with overground access, can make a neighborhood more and not less desirable.

But some of the choices around BART in the area were baffling. West Oakland station was built as a Park and Ride station, even though almost all neighborhoods around it are also served by BART. As a sign of how little the planners cared about the area they were tearing through, the station was called "Oakland West" for years, instead of "West Oakland".

Planners also put the line overground. This was not a West Oakland-specific problem – the Berkeley BART line also originally ran aboveground, until the neighbourhood gathered together and raised the money to put it underground. It took nontrivial amounts of political organization, too – BART originally tried to claim that it would cost $10m per mile to put the line underground, but Berkeley residents hired their own engineers and argued that it could be done for $5m per mile.

But West Oakland, filled with working-class families, could not put up the political opposition or raise the funds needed to do so. The resulting noise made it impossible to hear music in the neighborhood’s jazz clubs, and residents slowly drifted away.

What about today?

The sixties were a long time ago. Things have changed. For one, California is suffering from a critical housing shortage. American tastes in urban design are trending more towards transit-oriented development, because of concerns about climate change and a wider awareness of how the rest of the world lives.

West Oakland itself has changed, too. In the 1980s, the Lomo Prieta earthquake destroyed the Cypress Freeway. Thanks to local opposition, it was redirected around the neighbourhood instead of through it when it was rebuilt.

If the freeway killed the neighbourhood, why didn’t moving it away bring West Oakland back? Yes, it’s hard to undo the damage from decades downzoning and destruction. But it’s been forty years since the Cypress Freeway fell. In light of the Bay’s housing crisis today, why isn’t West Oakland more built up?

I’m not alone in thinking there’s a missed opportunity here. Mandela Station is a beautiful, utopian vision of a beautiful, walkable neighbourhood that would replace the current BART parking lot. It hasn’t been built because, as far as I can tell, the developers haven’t been able to raise funding – though maybe that's changed recently.

There are other attempts to build in the area. Indeed, the block immediately in front of West Oakland is home to three-storey affordable housing apartments. Developers have been trying to build market-rate housing there for years, like the Golden West project that would add 222 market-rate units and 16 low-income units.

Unlike in most San Francisco suburbs, opposition to building in the neighbourhood isn’t coming from wealthy homeowners. 70-80% of West Oakland residents rent.

Instead, it’s being opposed by a group of labour unions. Developers won't commit to using union labour on the grounds that it would make development unaffordable. They're probably right, but Oakland is a union town, so new buildings will be held hostage until morale improves.

Occasionally, some people say building will gentrify the neighbourhood and destroy its Black neighbourhood character. But as Darrel Owens cogently points out, gentrification doesn't look like new towers. The face of gentrification is East Bay neighbourhoods that look like traditional suburbs and don’t build. Those are the neighbourhoods that end up pushing out their Black and Latinx population:

Let's go north to the MacArthur BART station and see if there’s a pattern.

- West of MacArthur BART: Longfellow District. 2,848 Black residents in 2010. 45 homes built in 10 years or a 1.7% increase in housing.

- East of MacArthur BART: west Temescal, including the MacArthur BART Towers. 975 Black residents in 2010. 580 homes built in 10 years or a 27% increase in housing.

54 Black residents were displaced from the Temescal tract including the MacArthur BART highrise and Longfellow.

1,019.

Yes, read it again: 1,019 Black residents had gone. 35.8% of the Black population of Longfellow disappeared versus 5.5% for the MacArthur tower side. If we looked at that as a weekly average, that means for every 4 days in the last 10 years, 1 Black person disappeared from Longfellow—just one Oakland neighborhood alone.

This is the largest decline of Black residents of any residential neighborhood in the entire East Bay and reveals that Northwest Oakland is ground zero for Black displacement and gentrification in the Bay Area.

So, unfortunately, it looks like a funhouse mirror version of classic NIMBY-ism. There are no entrenched homeowners to fight construction, but a combination of unions plus slow planning bureaucracy and affordable housing requirements bleed all attempts to build to death. And meanwhile neighbourhood residents are denied cheaper rents, commercial opportunities, and all the other benefits of urban development.

West Oakland could be so much more; it could be the Bay's Exchange Place. The tragedy of West Oakland is not merely that it was devastated in the 1950s. It is that modern NIMBY-ism won't let it build itself back up to the vibrant neighbourhood it used to be.

One day I will own a home and tile my bathroom in aperiodic monotiles.

My friend Nolen is making a video game ~once a week. His games are fun, and his blog posts are awesome – not just as reflections on working, but on transitioning to self-directed work after years in finance. I recommend them!

While I'm plugging things, I wrote a GPT-based app that answers questions in using content from Marginal Revolution. Give it a spin and let me know if it breaks! It gives particularly good travel advice for exotic locations.

This person made an LLM fine-tuned on messages with his friends in a group chat. This is awesome, and I’d like to try it one day.

I find Nadia Asparouhova’s posts generally very good. Her most recent article is on talent distribution, in which she claims there are three distributions: normal, Pareto, and bimodal.

Most conventional jobs, like manufacturing, are normal — companies want to standardize quality as much as possible. Pareto companies are ones where prestige and conscientiousness are good signals of success. They’re the ones elite school students go to — consulting, big tech, investment banking. And bimodal companies are centered around extremely weird, ‘creative’ people. They end up with antisocial, creative geniuses who form supporting pods around that genius. Apple under Steve Jobs and fashion design houses are examples.

The most interesting insight I got out of this was that Big Tech is now Pareto-distributed but was originally more bimodal. Many of the things I think of as 'hacker-y' are a result of that bimodal influence, and I think this helps explain why Big Tech pays lip service to hacker culture without actually integrating it into its culture. (Specifically, this is because technical innovation requires creative people, operational excellence in sales/distribution requires Pareto people, and most tech companies nowadays are not technically innovative.)

I appreciated this piece about how Bryne Hobart made his career after dropping out of college. Given how influential he is nowadays, it’s pretty interesting that he didn’t go the conventional high-prestige route (by contrast to Matt Levine, who defines prestige!)

In more experimental-school writing, I liked this post on a family’s unlikely homeschooling journey. They did remote school the right way:

- Each child could work at their own pace, largely through playing educational games and apps that adapted to where they were. There was no particular endpoint that the kids needed to get to at the end of the semester.
- Group video calls were limited in size to no more than 6 kids (and often smaller), so kids got lots of personal interaction with their tutors and each other. Even as an adult, I find video calls larger than 6 people overwhelming.
- Regular movement breaks, where the kids had jumping jack competitions, did Cosmic Kids yoga videos, held dance parties, and ran around the house for scavenger hunts.
- Took advantage of existing materials: the program did not reinvent the wheel but instead made use of excellent, existing online videos and educational apps.

This is what remote school should have been! The tragedy of Zoom School is that we took all the worst parts of school and combined them with the worst parts of video calls.

Here is another piece on Living near your friends. This is very useful because it offers a straightforward list of tips:

• Tip #1: Host events regularly.

• Tip #2: Enable short-term stays - specifically, offer your friends a one-month sublet.

• Tip #3: Help your friends get leases

• Tip #4: Do roommate matchmaking

I appreciated this Crimson article on tensions at Harvard between the descendants of enslaved people and the children of well-educated African immigrants.

The complexities of immigration are often far more nuanced than people notice, and I find the history underlying waves of immigration to the US fascinating. At the elite college where I went to school, I noticed a similar pattern with Chinese Americans – most of those classmates were second-generation immigrants from well-educated Chinese or Taiwanese families, and almost none were third much less fourth, fifth, or sixth generation Chinese immigrants. In educational outcomes, it is useful to distinguish the effects of class from race.

I have half-joked that I will write the history of medium-frequency quantitative trading when I am retired and post-scarcity. I find the relationship between SIG and its spin-offs (and its spin-offs’ spin-offs) fascinating. So I was happy to find this 2007 article, which is an excellent profile of Jeff Yass. One of the crazier stories includes:

Once, Yass offered his golf caddie a ridiculous amount of money (company lore has it at one million bucks) if said caddie could make a hole-in-one. Yass clearly decided, a long time ago, that he was going to have fun getting rich — and then more fun spending his dough.

An interesting piece of trivia about Yass is that he didn’t come from the reputationally concerned New York elite. Prop trading, like investment banking, is now the thing that upper-middle-class Ivy League math grads do for a guaranteed ticket into generational wealth. But Yass and his friends were crass and not naturally ‘finance’ people. Were they part of that bimodal talent distribution? [edit: A friend points out that Yass was Jewish during the era of quota discrimination – the article highlights that he went to SUNY Binghamton, but he probably would have gone to a very different school today.]

I’ve been trying to take this piece on rest in motion to heart. A strange thing I’ve realized is that my rest is better when my life is busy. I found it difficult to feel good about my downtime when my life was idle. But now I’m busy and enjoy the non-work times much more.

Advertisements and media often push the narrative that the purpose of all our toil is to win a chance at relaxation. We’re supposed to work hard at boring jobs in order to earn our vacations. We’re supposed to work hard for decades so that we can retire. (We’re supposed to conceive of heaven as a place where nobody does anything except lounge on clouds.) I call bullshit. For almost everybody, inaction is boring. That’s why we pick up books, go exploring, and take up hobbies. The ground state is an active state, not a passive one.

And yet, most people have this model of the world where whenever they’re not resting, they’re taking damage. When the homework isn’t done, they’re taking damage. When they’re reading a textbook, they’re taking damage. When they go to sleep with work unfinished, they’re taking damage. When they’re at a large social event, they’re taking damage. Some part of them yearns to be in the rest state, where they don’t need to do all these things, and insofar as they aren’t, they’re suffering a little.

(note: I wrote this a long time ago. Nowadays we would call this RAG. The website fell over because Pinecone took down my database, and I no longer maintain it. Caveat emptor!)

I recently wrote MRBot, a website that answers questions using information from the blog Marginal Revolution. Writing GPT bots is in vogue nowadays, and I won’t pretend that this is breaking any new ground. But it’s a fun project, and I recommend it.

In my case, I like to check Marginal Revolution for travel advice, as I find that it often has good angles other sites don’t capture. However, using traditional search can be a bit annoying — for instance, if I want advice on travelling to Kenya, I might need to do multiple searches for Kenyan food, and advice on Nairobi, and general advice on travelling to eastern Africa. Then I’ll need to use my brain to filter and synthesize my results.

The benefit of embeddings-based search, which is what MRBot uses, is that it can effectively do all these searches at once. Because it then uses GPT to summarize the results it gets, it can also provide useful context to tie results together, saving my brain a bunch of cognitive effort. It provides a different — and, hopefully, better — search experience.

Here’s an example query and answer:

Q: Where should I eat in Mexico?

A: To eat well in Mexico, follow these tips:

1. Look for time-specific food, as it’s usually fresh and for locals. For example, in San Miguel, there is barbacoa from 8-10 a.m., carnitas from 11-4, and chorizo after 8 p.m.

2. Often the best meals are served in places with no names, like roadside stands with plastic tables and chairs.

3. Roadside restaurants on the edges of towns or between towns serve some of the best food in Mexico.

4. Focus on breakfast and lunch for your best meals, as dinner is often an afterthought and many good places close by 8 p.m.

Specific recommendations include:

• Tres Marias, a food village off the highway on the way to Cuernavaca, known for green chilaquiles and chorizo tacos.

• Coyoacan, a suburb of Mexico City, where you can find excellent and sanitary food stalls (“comedores”) near the Frida Kahlo museum.

• Pujol in Mexico City, which is a fancy place that does wonderful things with vegetables and offers the Menu de la Tierra.

Remember that the best food experiences in Mexico often come from exploring local markets, street food vendors, and small, family-run establishments.

It’s pretty neat!

The authoritative guide on how to do this is Question answering using embeddings, an ipython notebook provided by OpenAI. I won’t repeat what’s there; instead, you should think of this blog post as a companion guide to that cookbook.

(An embedding, in this context, is a way to transform a passage of words into a numerical representation that a computer can work with. It usually looks like a vector (or list) of 300-1500 floating-point numbers. Embeddings are useful in machine learning because they represent semantic differences as numerical differences, which means you can use them to cluster text or to ‘measure’ how similar/dissimilar different passages of text are.)

I’m going to assume some familiarity with systems design and AI, and I won’t go through all the details of how it works. (To be honest, I don’t grok transformers yet and I don’t know how embedding models are trained — but it turns out the libraries provide excellent enough abstraction libraries that I don’t need to!)

At a high level, here’s what the project involved:

• Step one & two: do data pre-processing (done once)

  • First you need to actually scrape data and put it in some retrievable place.

  • Then, you need to decide how to generate embeddings for the data, and store those embeddings somewhere.

• Step three: retrieval & summarization (done once per query)

  • Whenever you get a query, generate an embedding for that.

  • Then, do similarity search between your query embedding and all the data embeddings, and find the data that’s closest to your query.

  • You could return this on its own, in a Google-like way. However, most people who do this usually add some summary/synthesis by sending this data to GPT.

• (Optional) Step four: build a user-friendly UI

  • This actually involves no ML knowledge at all, but I dumped some notes on this here anyway.

Note that if you're not doing this as a learning exercise, developing question-answerers manually is slightly overkill nowadays. There are libraries and services out there that can do this for you. LlamaIndex is an example.

Step one: Gather data

The first thing I needed to do was get my data. This required no ML knowledge at all; it’s all traditional scraping and parsing. Fortunately, Marginal Revolution is pretty easy to scrape. I’m not the first one to try.

I wrote a script in python using BeautifulSoup and dumped each article into a sqlite database. I saved the raw HTML of each post. I wouldn’t recommend it. It’s worthwhile to save URLs and bold/italic information, as GPT can use this in its response, but I could have done that by saving Markdown instead.

Step two: Generate embeddings

Now that I had my data, I needed to generate my embeddings.

If I were to go back and re-do this project, this is the step I would spend the most time revisiting. The quality of the embeddings is the most important part of this project.

Unfortunately, it’s not something that can really be systematized. One thing I’ve been surprised about is how much ML research is an art, not a science. The end-quality of your product or model seems to depend a lot on your hyperparameters and/or design choices. In this case, I found the two key variables were:

• Which embeddings model I chose to use

• How I chunked my text

Which embeddings model?

You should collect some basic statistics on your model before deciding which embedding model to use. I say this from bitter experience, because I didn’t. For your benefit, here are the relevant ones anyway:

• I scraped 34,305 articles.

• The articles were an average of 199 words each. The shortest article had 2 words, and the longest had 4958. (‘Words’ here is counted by splitting html by spaces.)

• In total there were around 6.8 million words, which we can ballpark as about 9 million tokens (in English, a word usually requires 4/3 tokens).

The OpenAI cookbook suggests using their text-embedding-ada-002 model. At $0.0004 / 1k tokens, this is fairly cheap, but not free. I needed to embed 9 million tokens, so embedding this using OpenAI would have cost me 4 dollars. You might be happy to pay $4, but sad if your data sample is orders of magnitude larger than mine.

(You also need to pay to use this embeddings model on every query you get, but the cost of calling GPT on every query dominates this, so it’s not really worth worrying about.)

I didn’t do this math before I started working, and I was feeling cheap, so I used an open-source model instead. HuggingFace’s sentence_transformers library makes it very easy to download and use these models. There isn’t a quality dropoff – this article claims that all-MiniLM-L6-v2 is better than ada, and has a more detailed discussion on how embeddings models compare. Based on it, I used all-MiniLM-L6-v2.

This saved me money, but caused me pain. One downside of using an open-source model is that you’ll need to make sure you can generate embeddings at scale in production. This can also make it a pain to deploy (see my notes later). You don’t need to worry about any of this if you use OpenAI’s api.

You might also get great results from customizing embeddings — this OpenAI cookbook demonstrates how you would do that. I didn’t try it.

Chunking

Like GPT, most of these libraries have a maximum context window. text-embedding-ada-002’s sequence length is 8192 tokens, which is long enough for almost all MR posts. all-MiniLM-L6-v2's is 512 words, which is long enough for the average post.

Even if you’re not limited by your embedding model, you’re ultimately limited by GPT. GPT-3.5 has a context window of 4k tokens, and regular GPT-4 has a context window of 8k tokens.

This means you need to break your data up into pieces. How? It depends! It’s up to you whether it makes sense to feed it 10 results of 300 tokens each, or 2 results of 1500 tokens each.

I wanted to get through this and did the simplest thing possible: I split each article into paragraphs, and considered each paragraph a chunk. This works well for some Marginal Revolution articles, like this one, where each paragraph basically covers a different topic.

However, it works less well on others, like this one:

Title: Russia and China

Authored by Philip Snow, the subtitle is Four Centuries of Conflict and Concord. This book is excellent and definitive and serves up plenty of economic history, here is one bit from the opening section:

The trade nonetheless went ahead with surprising placidity. Now and again there were small incidents in the form of cattle-rustling or border raids. In 1742 some Russians were reported to have crossed the frontier in search of fuel, and in 1744 two drunken Russians killed two Chinese traders in a squabble over vodka.

I am looking forward to reading the rest, you can buy it here.

The fact that Cowen recommends the book is in a different paragraph to that which actually describes the book. So if our embeddings-based search only returns one of these paragraphs and not another, it'll be confusing to GPT and the reader.

For instance, the paragraph "Authored by Philip Snow, the subtitle is Four Centuries of Conflict and Concord. This book is excellent and definitive and serves up plenty of economic history, here is one bit from the opening section:" will only come up if you search for Philip Snow and not if you search for Russia, or China, or great power conflict.

I decided not to worry about this. Once I'd embedded enough paragraphs, I gave my search a try. However, my search results were awful, as paragraphs alone weren't enough context. I still didn't want to embed entire articles, so I added the title to the embedding as well. That fixed the problem, as it turns out that titles can do a lot to contextualize individual paragraphs of posts.

For instance, in our example above, now we're capturing the paragraph "Russia and China: Authored by Philip Snow, the subtitle is Four Centuries of Conflict and Concord. This book is excellent and definitive and serves up plenty of economic history, here is one bit from the opening section:". This is now much more likely to come up if we search for "great power conflict" or "Chinese foreign policy"!

In my opinion, this is the most fun part of the project to tweak. If I re-did this, I would try splitting up MR posts into those which are high-context and those which are just lists of things, and treating them differently! And if I were working on this at a larger scale, I would write some tooling to make it easy to try different embedding models on different datasets.

Step 2.5: Storage

This is a half-step because you need to figure it out while you’re generating embeddings. To start off with, I actually stored my embeddings in SQLite while I was generating them (as blobs). I'd retrieve them and store them in memory and iterate all of them while calculating similarity.

This got to be too slow in production. I've been told there are various hacks that will let you optimize this. I decided to throw money at the problem instead, and switched over to Pinecone.

Step 3: Retrieval

This is actually the easy part! The OpenAI cookbook gives you all the code you need. My modified version of it (swapping out the OpenAI call for a sentence_transformer call) basically looked like this:

  import pandas as pd
  from sentence_transformers import SentenceTransformer, LoggingHandler
  from scipy import spatial  # for calculating vector similarities for search 

  model = SentenceTransformer('all-MiniLM-L6-v2')

  def strings_ranked_by_relatedness(
      query: str,
      embeddings: list[tuple[Article, np.ndarray]],
      relatedness_fn=lambda x, y: 1 - spatial.distance.cosine(x, y),
      top_n: int = 100
  ) -> tuple[list[str], list[float]]:
      “””Returns a list of strings and relatednesses, sorted from most related to least.”””
      query_embedding = model.encode(query)
      articles_and_relatednesses = [
          (article, relatedness_fn(query_embedding, e))
          for (article, e) in embeddings
      ]
      articles_and_relatednesses.sort(key=lambda x: x[1], reverse=True)
      articles, relatednesses = zip(*articles_and_relatednesses)
      return articles[:top_n], relatednesses[:top_n]

If your process is anything like mine, your most related articles will probably be a bit of a mess. Here’s a shortened version of what mine looks like, for the query about Mexican food:

Use the below articles from Marginal Revolution to answer the subsequent question.
Do your best to synthesize the results, as the articles may not directly answer the question or might be irrelevant.
Do not add information that is not in the articles.
Give specific examples. If the user’s query is not a question, say ‘I don’t know how to answer that.’
How to eat well anywhere in Mexico: <p>5. Roadside restaurants, on the edges of towns or between towns, serve some of the best food in Mexico or anyhere else for that matter. Some of these restaurants even have names, though you can overlook that in the interests of eating well. </p>
How to eat well anywhere in Mexico: <p>1. Look for time-specific food. In San Miguel for instance, there is barbacoa [barbecue] from 8-10 a.m., carnitas from about 11-4, and wonderful chorizo after 8 p.m. In Mexico, if the food is available only part of the day, it’s almost always good. It’s for locals and there is no storage in these places so it’s also extremely fresh.</p>
How to eat well anywhere in Mexico: <p>You’ll sometimes hear fallacious claims that San Miguel Allende or Guanajuato or other parts of Mexico don’t have superb food. What is true, in many Mexican cities, is that almost every place near the main square is only so-so. Here’s what to do:</p>
[...]
Where to eat right next to Phoenix airport?: <p>Chili or Mexican food would be ideal. Thanks in advance for the help…</p>
Some food notes from Mexico City: <p><a href=“http://pujol.com.mx/“>Pujol</a> does wonderful things with vegetables and is perhaps the best f la Tierra.</p>
Some food notes from Mexico City: <p>They have done away with the food stalls at the Zócalo. In Mexico City calorie-counting menus are common and gelato is being replaced by frozen yogurt (!).
Question: Where should I eat in Mexico?

This is, I think, why most versions of embeddings search then call GPT to summarize the results. It's a pain for a human to go through and extract the useful information. However, if the goal is just to get enough usable information for GPT to synthesize it, you can then feed this in and get something human-readable out the other side.

You'll want to keep the temperature low so that GPT doesn't hallucinate information. I found that GPT-4 was notably better at synthesizing information than GPT-3 was.

This does get expensive! I’m just paying for MRBot queries right now, because this is fairly niche. (Happily, OpenAI allows me to set a monthly cap on what I’ll pay them, which I’m using to limit my costs.)

If you were doing this commercially, you’d probably want to charge customers, or have them charge it to their own OpenAI account. If you were doing this in-house, the cost of each query is probably negligible compared to how much time the people working with you will save, but if you care you can implement token tracking and quotas.

Step 4: UI

Hilariously, this step took me longer than the last two (or three) put together. However, I wanted to make a nice website that my friends and I could access from our phones, and that required some traditional software development.

MRBot is a simple React frontend (deployed using Cloudflare Pages) that calls out to a backend Flask service. I spent an absurd amount of time trying to figure out whether the best practices around deploying a Flask app have changed at all in the last 10 years. They haven’t.[1]

[1] With one caveat — I spent some time poking around to see if I could make a nice Docker setup for the application. I could, but only by making Docker images that were 6GB+ in size (!). It turns out that the CUDA libraries used by sentence_transformers are really big. If you instead use OpenAI’s API to use your embeddings, I suspect you don’t need to include these huge libraries in your requirements.txt, which makes deploying with Docker reasonable.

Peculiarities of OpenAI API calls

OpenAI responses are really slow. If you use gunicorn, you’ll want to make sure the workers are using gevent, or you’ll block your workers while waiting on the api call to return. asyncio will do a reasonable thing here too, but I’m not familiar with it.

The only really interesting thing I’ll note here is that it was a little bit of effort to stream the OpenAI response to the client. However, it’s worth doing this because it’s a notable improvement to the user’s experience. I implemented it using a hacky version of server-side events.

A plug for Figma

One thing I’ve learned recently while developing projects is that Figma is awesome for prototyping websites. I used to design websites by tweaking raw code and looking at it in Chrome.

In practice, though, you really don’t want to do this more than necessary — it’s annoying to remember the syntax for the changes you want to do, and time-consuming to wait for it all to reload. Nowadays, I prototype websites by using auto layout in Figma and basic component libraries. Once I have a layout I like, I then turn that into HTML, and usually end up with something decent. It’s notably improved my frontend development experience.

A final point about ethics. I added a large disclaimer to the MRBot website clarifying that nothing here should be taken as the view of the authors. As discussed above, I made a lot of creative choices when writing this bot, all of which change its output in ways that don't reflect the authors' intent. All faults are mine, etc. And I don't think it would be right, or legal, to charge for this.

The legal issues here remind me of those around fan remixes or fanfiction though, where technology changed how people could appreciate work faster than copyright law could keep up. If you write a bot that summarizes Wikipedia entries, can you sell that? What if it draws on information from multiple sources, some of which are freely-available and some of which are not? I'd like us to err on the side of experimentation, but others might not.

But other than that, that's it! I hope you learned something from this. Do let me know if you end up doing something similar!

• you are in your twenties or early thirties,

• you are a person who has periods and will go into menopause,

• you have a good job with either fertility benefits or enough pay for you to cover costs out-of-pocket and

• you can spend a few months in a city with a good fertility clinic,

I recommend you get your eggs frozen. Yes, even if you're in your early twenties. It's definitely an expensive process that is uncomfortable. But it's also cool technology. It will provide a lot of peace of mind if you know you want to have kids eventually, but not now. Even if it doesn't work out for you, you'll learn a decent amount about your natural fertility – I couldn't complete my egg-freezing cycle, but the information I learned from it was very valuable in planning my life and career.

As a society, we haven't figured out how to make female fertility less fickle. In an ideal world we would be able to create eggs from skin cells and gestate embryos in artificial wombs, bringing the human costs of pregnancy to zero. I hope that's the world my children and grandchildren live in, but it's not the one we live in today.

Today, if you don't have eggs frozen and you get ovarian cancer, or want to take testosterone, you lose the ability to have a natural child. Even absent catastrophic tail risks, it's nice to have a few eggs frozen – a friend of mine plans to have her first few children naturally, but wants to have the option to have more children after menopause.

The great thing is that we have the technology to help with this! It's not ideal, but it's also not terrible, and I think the trade-off is worthwhile for most women in developed countries who can pay the financial costs.

The rest of this post goes through how egg freezing works, what the process actually looks like, and how you can arrange it.

How does it work?

Let's start by recapping some facts about your period that you may or may not have learned in your 7th grade 'sex-ed' classes. (Yes, the ones where they separated the ovary-owners and the sperm-carriers into different rooms.)

• When you're born with ovaries, you are also born with a prdetermined count of all the eggs you'll ever have.

• This count is represented as your count of 'primordial follicles': follicles (like hair follicles) that will grow eggs when needed.

• During a menstrual cycle:

  • Your body selects several follicles and has them grow eggs. This number might vary from 8-40.

  • One of those eggs is selected as the 'best' egg. When you ovulate, that egg is released into your uterus. The other egg candidates get trashed.

  • That one egg can become fertilized or not. If it isn't, it's expelled with your uterine lining: this is what causes monthly bleeding.

Egg freezing hijacks this process. Instead of disposing of those extra eggs, you take hormones that encourage them to grow. Then, once they've all grown, and just before you would normally ovulate, you take medication that encourages them to be released. Immediately after that, your surgeon goes and retrieves the eggs.

In more detail:

1. The first step in any egg-freezing process is for you to get basic fertility testing. This usually looks like an ultrasound and a blood test.

   • The ultrasound will test for your antral follicle count (AFC), or how many 'mature' follicles a nurse can see on an ultrasound. This usually provides a good guide to how many eggs your doctor will be able to retrieve. Here's a guide to what different counts indicate. A pleasant side-effect of this ultrasound is that if you have endometriosis or any other issues, you'll be able to discover that too.

   • The blood test will tell you your anti-Mullerian hormone levels. This provides a guide to how many primordial follicles you have left. Here's a comprehensive guide to what different levels mean..

   • If these numbers are too low, this can indicate that you'll have problems stimulating enough eggs for egg retrieval to be worthwhile.

   • You can do this anywhere and at any time. If you're thinking of doing egg freezing, I'd recommend doing this as soon as possible. It'll give you a good indication of how many cycles you'll need to do.

2. Then, you go on a course of drugs that your doctor recommends to stimulate your follicles. This will last around 2-3 weeks.

   • You'll need to be near your clinic for this, as this stage usually requires you to go in for an ultrasound two or three times a week. Nurses will monitor to make sure the drugs are working as expected and to make sure all the follicles are growing at the same speed. If they aren't, this is a likely place where your cycle will be 'cancelled.'

   • These drugs (known as FSH hormone) will cause all your follicles to develop eggs. This will cause your ovaries to feel heavy, and you might feel bloated. Doctors usually recommend that you avoid exercise or jumping, as you can cause a rare condition called ovarian torsion, where your ovaries twist and cut off blood flow to themselves.

   • These drugs are usually subcutaneous injectable drugs. Injections sound scary, but these aren't! They're administered with devices that look like pens, and the needles are incredibly thin. You stab yourself near the stomach, press a button, hold it there for a few seconds, then take it out. It's as painful as a pinch. Your nurses should demonstrate.

   • At the beginning, you'll likely take one injection a day. Towards the end, you'll need to take injections in the morning and evening.

3. Finally, just before your body would naturally ovulate, you take a 'trigger' shot that prepares all the eggs to be released. Shortly after that (within 24 hours), your doctor retrieves all the eggs.

   • The mechanics of the surgery are pretty cool. Your doctor will stab a large needle into your ovaries through your vaginal wall. Using an ultrasound, she'll identify each egg and 'suck' them into the needle.

   • You're under general anesthesia for this. People describe it as pretty pleasant. However, you'll be very loopy and vulnerable afterward, and the clinic I used required you to have someone to pick you up. Recovery can be very painful and last several days, though some women don't have any pain.

As you can see, an egg-freezing cycle lines up nicely with a menstrual cycle, which is why it’s called a ‘cycle.’

You need 10-12 high-quality eggs to get an expected child. The number of eggs retrieved in one cycle varies a lot from person to person — from 8 to 40. (The early fertility testing can give you a guide to how many you’ll expect). Younger women tend to have more eggs, and those eggs tend to be of higher-quality. I found that doctors tend to be quite excited (and a little confused) if you do this while you’re in your early-to-mid twenties, as it’s pretty rare.

Depending on how many eggs you’re able to retrieve and how many children you want, you may need to do anywhere from 1-3 cycles. Most companies with fertility benefits will cover multiple cycles with a pre-defined cap.

Note that if you want to have the option for surrogacy, you need to do tests before egg retrieval! These are FDA-mandated in the United States. They're fairly simple tests, but your clinic may not pre-emptively tell you about them.

This LessWrong post has more details on how to increase the number of eggs you can harvest, and some guidance on where the best clinics are. (It claims to be about how to have polygenically screened children, but the best way to do that is to retrieve lots of eggs in an egg freezing/IVF cycle, so it's packed with useful information for us too!)

Why do I need 10 eggs to get an expected child?

The pipeline from egg retrieval to pregnancy is pretty abysmal. If you do the math, one cycle produces, in expectation, one kid – even if you retrieve 10 or 15 eggs!

The rough breakdown looks something like this (from here):

• From 10 – 12 harvested eggs, 7 – 9 will be suitable for vitrification and storage.

• Approximately 70-90% of the eggs will survive warming in the future.

• Approximately 50-80% of surviving eggs will fertilize.

• Approximately 50-90% of fertilized eggs will develop into embryos.

• A single embryo will have a 20-35% chance of developing into a pregnancy.

The final pregnancy number sounds a bit low to me, but the estimated loss in the pipeline seems about right.

You can opt to freeze embryos instead of eggs. I've heard varying opinions on whether this increases your chance of success. It certainly reduces volatility, as you can realize the risk of having an embryo upfront and decide whether to do another egg-freezing cycle while you're still fertile. Obviously, it comes with other trade-offs.

How do I actually do this?

The price I have was told for each cycle is about $10-12k, though I’ve heard claims that it’s closer to $20k now. The cost is high, but many tech companies have fertility benefits that cover it completely for tech workers and their partners. (Google, notably, also provides surrogacy benefits). If your job doesn’t cover it, the price does put it out of reach for most people, but I think it’s worthwhile if you or your partner can afford it. I don’t know if it’s deductible or not.

If you can arrange for it to happen – especially if you are travelling to get your eggs frozen – it’s better to do it with some friends. That was what I did. It makes the process much more fun. You can have ‘first shot’ parties and compare experiences.

The quality of your clinic matters a lot. I used Spring Fertility in San Francisco, which was decent. They are also in New York. The experience is concierge-like, and many clinics will assign you a ‘case manager,' though the quality of this varies a lot (mine was sometimes hard to reach). A friend of mine had an excellent experience with Dr. Klatsky, who is in New York.

The clinics had fairly good availability when I tried talking with them. They were pretty happy to arrange for freezing/retrieval to happen during times that worked for me.

If you do end up freezing your eggs, let me know how it goes!

Their broader concern is that it was long customary in the United States for people with room to spare in their home to occasionally supplement their income by taking in boarders. So a young person new in town might rent a bedroom from an empty nest couple. Or a family in need of extra resources might make the kids double-up in a room to free up space for a boarder. One countermeasure they advise is to make it illegal to advertise your willingness to accept a boarder.

It's a shame. Freedom of movement is important for economic development, and efficient markets in shelter are good for the world. It's also anti-American – if you own land, you should be able to do whatever you want with it!

I generally believe in supporting creators I enjoy following, especially independent ones. So, a long time ago, I started supporting Tanner Greer, an erstwhile China scholar and American foreign policy critic (there are far fewer people in the intersection of this space than I'd like).

Once a month, he’ll jump on a Zoom call with higher-tier Patreon supporters for a monthly discussion. March's topic was Effective Altruism, the influence of rationalism on it, and its similarities and differences to other intellectual social movements. (I particularly enjoy Greer's perspective on social organizations because he is ex-Mormon. The Mormons have one of the world’s most successful, structured social movements, and there is a lot for the rest of us to learn from them.)

In the discussion, he brought up an old article of his: Questing for Transcendence. It claims that few experiences can consume a human as much as that of a higher purpose. A Mormon missionary will live without privacy for two years in a foreign land and enjoy every second; an EA will do mundane office admin with a smile because they know they are saving the world. But if you ever fall out of that mindset:

Many [returning missionaries] report a sense of loss and aimlessness upon returning to "the real world" [...] For the past two years there was a purpose behind everything they did, a purpose whose scope far transcended their individual concerns. They had given everything–“[heart, might, mind and strength“—to this work, and now they are expected to go back to racking up rewards points on their credit card? How could they?

I don't believe in transcendence anymore. I've been caught up in smaller versions of it before, and it is glorious. It's also very dangerous, and I think one is better off finding purpose in smaller joys. But it's worthwhile to think about what it is we're giving up.

If you’re interested in New York politics, I recommend you keep an eye on this Foundations of New York course. It meets once a week and is run by someone who bills himself as a New York local politics expert. For what it's worth, he has a cool substack. The pitch is:

- The anti-politics meme: the unconscious programming most people have that steers them away from politics.
- The anti-concreteness meme: the rhetorical habits and cultural defaults that prevent people from understanding government the same way they expect to understand a physical trade, science, or math.
- No easy way to learn: politics is as complicated as any STEM field, but you will not readily find easy introductions to it anywhere. Most people only have the option of “figure it out yourself,” with no obvious set of YouTube videos or books to assist them.
- A loss of social technology: New Yorkers (and Americans) were not always so helpless and disengaged from their civic order. People accomplished extraordinary things in the civic realm, and you can become one of these people.

If I lived in New York, I would sign up for it. (I would also sign up for an equivalent Bay Area course if it were available). Applications have closed, unfortunately.

I find that I collect stories of financial crime nowadays. This one was forwarded to me by a friend, featuring a band of four young Americans who defrauded the Bank of England in 1872. Many components will be familiar if you've read Lying for Money.

The fraudsters started with consignment fraud in the US — i.e., borrowing against goods they knew would never arrive. Then, they sought their sights overseas and started forging cheques that would take weeks to clear. Many banks knew this and would not allow customers to borrow against them — but the renowned Bank of England, trusting in the reputation of its clientele, would. Which made it a classic target for fraud.

Dan Wang is an analyst who lives in China (edit: now at Yale Law School, thanks to Lifan Z for the correction!), best known for writing an annual letter that is always interesting, thoughtful and very long. This year’s is no exception. The best part describes his self-imposed exile to Yunnan during Shanghai’s lockdown, and is an ode to the magic of travel:

When I miss the food of Yunnan, it is the dishes from Xishuangbanna that make me most dreamy. The city’s lifestyle is nocturnal since the people are dependent on rubber production: rubber trees are best tapped at night when temperatures are cool. Therefore the streets are fairly empty in the midday sun, coming alive in the evening. That is when people crack open beers and enjoy grilled meats before they enter the forests.

The rest of the piece is a thoughtful reflection on China's COVID madness and the limitations of state power. I particularly enjoyed this passage:

The Chinese state is usually levelheaded; but every so often it succumbs to a manic episode, in which it grips the population, not relenting until it has shaken them out of their pots for backyard steel furnaces, out of their schools for class struggle, or out of their minds for dynamic zero clearing. It then comes to its senses and sets down a battered people, as the rest of the world looks on aghast. The state is then sane and sober once more, though the people feel the occasional nervous tremor.

I didn't completely agree with this piece on what a Carribean think tank should focus on, but I thought it was insightful (and I learned a lot!). I didn't realize that four Nobel laureates came from the Carribean, or that:

Singapore’s Lee Kuan Yew looked to Jamaica as an early model good governance for small states.

I agree with the author that the Caribbean has, somehow, shot itself in the foot. As the author points out, the Caribbean is blessed with the natural resource of beautiful land, but fails to convert the dollars from the resulting tourism into services for its citizens. It is expensive, difficult to visit, and not a very nice place to live.

Here's my pet political take. Several countries in the Caribbean are blessed with strong legal systems (because of British rule) and proximity to the US. I suspect a charter city with generous immigration policy, a functioning port, regular flights to Miami and good internet would bring in a huge amount of talent and businesses that want access to the US market but can't live there because of restrictive immigration policy. (In effect, this is what Hong Kong and Singapore did – in fact, such a city that appealed to Asian immigrants would have a huge impact.) Why doesn't this city exist? Shouldn't it?

I enjoyed this conversation between Patrick Collison and Reid Hoffman. It's largely about Stripe is handling 'this macro environment', but I include it because I found this quote hilarious:

I, as a teenager, was really into programming languages and Lisp and all this kind of crazy stuff, but at least sort of aspired to being some kind of technologist. John was always super interested in business.It was early in Stripe’s history, and I pitched John on, “Well we should just make Stripe free because we’d grow much faster.” And John’s like, “Interesting. Can you explain the business model embodied in that, to me?” And, I didn’t really have one.

You may or may not have seen this Less Wrong post on the childhoods of exceptional people. You may or may not agree with the methodology, but at the very least it highlights how being raised amongst adults and tutored one-on-one might not actually be the worst thing for a child.

(On a related note, one of the coolest things I've heard about this month is Fusion Academy, which basically operates as school through one-on-one tutoring with a common space attached for kids to hang out in when they're not in class. As a result, you get a lot of freedom-you can take week-long breaks as needed, and you can study and move forward at your own pace. It's very expensive – on par with New York private schooling – but I think it would be cool if this educational model was accessible to all children who wanted it. Maybe AI can get us there?)

This piece is written by a 22-year-old Oxford graduate who apparently decided to move to Taliban-controlled Kabul as a foreign correspondent after graduating. I have done some crazy things, but even I am nowhere near that crazy. It is an interesting portrait. There are a few zingers, like:

“Nobody is going to hurt us, dude,” he said, curling his lips into a smile. “We’re not in San Francisco.”

But for the most part, he portrays life in Kabul as fairly mundane. Former suicide bombers put down their vests and now approve forms in office; Michelle Obama's Becoming features in book stores. People still worry about suicide bombs, but the bogeyman now is the Islamic State.

If you are a fan of home espresso machines, you will enjoy this piece on the history of home coffee machines. Spoiler: Decent Coffee is the pinnacle of modern home coffee technology.  

I've started to hear the phrase the 'professional managerial class' (or PMC) a lot. If you haven't heard this, think of the sort of job a McKinsey consultant grows up to hold. I enjoyed Marc Andreesen's take on the topic (in a podcast with Dwarkesh Patel), which was that the PMC is what you get in capitalist development after the bourgeoisie. The bourgeois capitalist owns and controls his business; in late-stage capitalism companies are owned by dispersed shareholders but managed by the professional managerial class.

The reason he's thought about this is because it's relevant to how startups scale:

When the companies we fund get to scale, they tend to get pulled into the managerial orbits, they tend to get pulled into the managerial matrix, which by the way, is when they stop being able to build new things, which is what causes the smart and aggressive people at those companies to leave and then come back to us and raise money and start a new bourgeois capitalist company.

My friend Sheon wrote a fun piece for Wired, making the Case for Software Criticism. I went in expecting to argue back that software is fundamentally boring and there's not much to say about it, but he was way ahead of me:

A software critic may begin with some requisite cultural history on the labor of writing, but then also provide a bit of technical (and even geeky) history-cum-explainer on how the operational transformation (OT) technology of Google Docs paved the way for real-time collaboration tools in other fields, such as Figma for design or Colab for programming. And how the research in conflict-free replicated data type (CRDT) could make this mode of working a default mode of collaboration in the future. And what that means culturally and sociologically.

If you're a certain type of nerd (and I am), that is pretty interesting. In the end, I left thinking that it's in fact a battle cry for more people to do a certain type of analysis that combines art history with engineering knowledge. If you're capable of placing a Renault in its social and historical context, maybe you should instead be placing modern consumer software in its social and historical context. (In fact, two writers I enjoy reading most – Matt Levine and Bryne Hobart – arguably appeal to me because they deeply analyze the social and historical context of major businesses). And maybe our schools and humanities system should value this the way they value art history.