I write about some personal things on here, none of which are particular to me. Most human experiences are at least somewhat common even if the way those experiences are implemented is unique. This uniqueness gives room for beautiful writing and sharing some raw and soft part of you.

It comes at a cost, of course. Exposing yourself is hard. You feel vulnerable. It can alienate others. I write about other people on here, and sometimes those people don’t like it. I consider it an obvious necessity to do this – you don’t have to name names, but if other people are part of your inner world, and you want to share it publicly, you’re going to have to reference those people even if obliquely.

A girl I’m seeing asked to read my blog. I was surprised I hadn’t sent her the link, so I did, only to immediately remember that I had written about her on here... Nothing hurtful, quite complimentary even. But nonetheless the urge to edit the post, or take it down altogether, was present.

To get your merit badge at Inkhaven you had to write a piece in each of seven categories. Id already hit all of them except for fiction. I’ve never written fiction, and when I attempted it I wrote three pieces and scrapped all of them. (I asked Max Harms for feedback on a sci-fi thing I wrote. He said it was mediocre and that he looked forward to reading the next thing I wrote because that would hopefully be better.)

So I wrote a story that was very loosely based on some people I lived with. A couple of weeks later I got a terse message from someone who recognised herself in the story. The characters were composites, and the story was clearly fiction, but it was enough to trigger a takedown request.

I would love to say that these sorts of things roll of my back, but the message immediately made my heart race. I don’t like conflict and I don’t like upsetting people. But if people can’t draw on life for their writing, there’d be no writing at all.

The temptation to self-censor is strong. But consider that if there are some things you just can’t discuss, even anonymised, then maybe you’re killing some part of yourself in service of things being more copacetic, of being a person who expresses nothing controversial. These are life goals of dead people.

I don’t want to make anyone mad. I don’t want to hurt anyone. I want to take up less space. I don’t want to break the rules. I don’t want people to laugh at me. I want to be convenient. … These are what I call the life goals of dead people, because what they all have in common is that the best possible person to achieve them is a corpse.

All of which is to say: writing about squishy personal stuff can be costly. You should do it anyway! I’ve had lots of people say how much they enjoyed my writing, who shared their struggles with heartbreak and depression because of what I wrote. (Who knew that three people at Inkhaven had done rTMS?)

Putting yourself out there is an invitation for others to connect with you more deeply, to talk to you about more sensitive topics. It will turn alienate and turn off some people, but it will bring others closer. Perhaps this is necessary – much like how your dating profile shouldn’t optimise for widespread appeal if you want to actually find a partner who is compatible with you

And to you, reading this: I write partly to express myself, and partly as an invitation to connect. If I’ve sent you a link to this blog, or even if I haven’t, reach out if something resonates! Bring it up next time we meet.

You must show your soft underbelly to the world, and others will show you theirs. There’s magic to be found there.

I suspect “grindset” is a bit out of fashion these days, but the startup world doesn’t seem to know that. For as long as I’ve been in that milieu, grinding – pushing yourself to impossible limits of effort, spending all day and night working on or thinking about your startup – has been not only valorised but presented as an essential part of running your own business.

Partly this is because startups are very unintuitive. If you’re a tiny company with a tiny team and you’re hoping to take on IBM, it seems impossible to succeed – except by applying superhuman amounts of effort.

This of course is the wrong theory of change: startups can beat large companies because those companies become sclerotic, both in the sense of their technology and of the organisations themselves.

At my last startup, we built fintech software that was miles ahead of everyone else without working ourselves to death. How could a tiny team make something better than multinational financial services companies? Because we were building something from the ground up using modern web development practices. We could do it with a tiny team because we insisted on using Boring Technology and making smart decisions that made the product easier to maintain. We also didn’t have the dead weight of a legacy platform to support.

It wasn’t because we just typed faster than the competition. If anything, programming doctrines like Extreme Programming emphasise working at a sustainable pace. Short crunches are probably an inevitable part of running a small company: planning is more art than science, you have deadlines to hit and clients to impress, so you need to pull out all the stops for a while to get things out.

Marc Andreessen talks about Mozilla needing server software that was at parity with what Microsoft was producing; while his developers were looking for a magic shortcut, he just said simply: “we don’t need a silver bullet. We need lots of lead bullets.”

But ultimately grinding catches up to you: you write unmaintainable code, make poor decisions, and your team starts to resent you. It’s a tool to be applied selectively rather than a company culture. If anything, constantly grinding and employees burning out is a sign that you’re really not getting it right as an organisation.

This is the second way startups can compete with the big guys, even though that seems impossible on its face. If you have a company ‘operating system’ that cuts organisational cruft, you can simply get more done per unit of work. Outcompete through sheer efficiency – instead of endless planning meetings, layers of approval, design focus groups, interminable email threads and all the other nonsense that large bureaucracies attract, focus on building and moving faster. Even when we started out, my cofounder could throw me a feature spec in the morning and have a working version to test by lunchtime.

I always hated the fact that SBF was lionised for sleeping on a beanbag under his desk (to say nothing of the other obviously stupid and performative stuff he did, like playing League of Legends while on investor calls). But people loved it. Surely FTX couldn’t just succeed on its merits, being one of the most popular and innovative crypto trading platforms – it could only be successful because of its wunderkind CEO who worked infinity hours.1

But “we chose boring tech and kept the org flat” doesn’t photograph well, so companies present themselves as hustle machines to signal to investors and to themselves that they are Very Committed. VCs lap it up (or worse, demand it), continuing the cycle. But it’s mostly theatre, and founders would be better served spending that energy on running their companies in a productive and sustainable way instead of doing grindset kayfabe.

I’m torn on whether the desire to move somewhere new is a push or a pull. Sometimes you’re drawn to a new place: for a new job, a community that’s a better fit, a pace of life that suits you, or just for a change. And sometimes you move because you’re escaping the old place, and all the baggage and trauma you’ve built up there.

Certainly my move to Bristol was the latter. I was escaping a shattering heartbreak and all the misery that entailed. I’m about to spend a summer in Berlin, and though I certainly feel mostly done with London, I’ve felt that way for quite a while, and the draw of Berlin is undeniable. Spending six weeks in Berkeley and the bay also stirred something in me, a sense that a close-knit community of my kind of weirdos, and being at the centre of the universe at a pivotal moment in history, might be within touching distance.

I think it’s worth being wary of the moves where you’re being pushed to leave, rather than pulled to the new place.

Of course all moves will have a bit of both. Few will move to a new place which they know nothing about. But the push moves can be hasty, and probably are ultimately less successful than ones where you’re drawn to life in the new place. Bristol had some of this shape for me. I had some amazing times there – cycling round Ashton Court in the sun, a Christian Löffler gig on 2C-B, the new friends I made and the festivals I went to. But I wasn’t in a good enough place to really able to make the most of it.

I visited a friend in Berlin a few months ago, someone recovering from a relationship breakdown and a business failing, in quick succession. It was a strange visit, but I think it cheered him up, and I loved spending time with him, enjoying Berlin’s lakes and sauna culture (I make no apologies for being an FKK enjoyer) and putting the world to rights over a beer, or several.

He couldn’t stomach London any more. Every street he walked down was imbued with some sort of memory of the old relationship, of things gone wrong. He went back for a day to collect some belongings but wasn’t going to stay a minute longer than he needed to.

His description of this was eerily familiar. For a long time I didn’t go anywhere near the spot in south London where I used to live, taking longer routes on my bike to avoid even passing through it. It took years before that visceral reaction faded. The memories didn’t have the same valence, or intensity, but they were still there, indelibly etched into the roads and buildings you once haunted.

When you’ve been in a place long enough, it can feel like every street you go down has some significance. Sometimes moving to a new place is just a good way of escaping the well-trodden streets of your old haunts, and all the memories they’re imbued with.

I went to a lightning talk on a ‘little-known’ productivity hack that will change your life. The idea? Putting monetary bounties on things to get yourself to do them.

This seems like a reasonable thing to do, especially for high-stakes things or items with a strong ‘ugh field’ that you just put off again and again.

Tying yourself to the mast

Beeminder is this idea on steroids. You set a commitment – perhaps going to the gym three times per week – and you enter a ‘1’ datapoint each time you do. If you don’t go on time, you get a monetary penalty, which starts at $5 and increases exponentially, so it quickly becomes an amount of money which you really care about.

It’s very motivating, and it’s cool to look at all your little graphs and see them ticking up.

There are other neat features – you can’t just change your pledge to weasel; there’s a seven day cooling-off period. You can have ‘do less’ goals, but you can’t just enter no data to game them – you’ll fall off the rails if you do. It’s a very cleverly-designed system. It might also totally destroy your motivation to actually do anything, because your past self is basically torturing your future self, and this is by design.

Punishment isn’t enough

Items surrounded by an ugh field are almost certainly getting put off because they’re aversive, even though they have a high cost – if you’re not filing your taxes because you just can’t bear to do them, you’re presumably also eating the cost of late filing penalties. The punishment is already there but it isn’t enough. Adding more punishment on top just makes you feel bad.

If you aren’t going to the gym because it fucking sucks and you hate going, then punishing yourself further by having a Beeminder goal on top will just make you hate it more.

Against sledgehammering yourself in the nuts as a motivational strategy

Instead of spending the money on Beeminder fines, you could hire an accountant to do your taxes for you. Taxes fucking suck so just throw money at the problem to make them go away; it’ll be a hell of a lot cheaper than the late filing penalties.

Instead of fining yourself five bucks each time you don’t go to the gym, sign up for a gym that’s nearby and find a routine that you enjoy, or just sack off going to the gym altogether and get your exercise needs met some other way (I started doing sessions at Barry’s and found I really enjoyed them, much more than going for a run).

Sometimes negative motivation – avoiding a bad thing – is the right way to get yourself to do something. But prefer positive motivation, where you can make the task itself motivating. Stop torturing your future self. It’s better for you.

With this level of UV, people who are relatively fair-skinned burn in less than 15 minutes. Which explains the sunburn on my shoulders when I was wearing a tank top in the morning for about half an hour outside.

The UV index indicates the risk of burning for someone with lightly tanned skin, and runs from 0 to 11 (it’s open-ended; the Andes might see 20).

Intuitively we expect that the UV index is correlated with the temperature. This is a useful heuristic but for many places it won’t apply – skiers know that you can get cooked on top of the mountain, both from direct UV exposure and reflected UV from the snow (albedo). The same for coastal locations, where the sea air coming in off the coast has a cooling effect.

So for it to regularly be 10 in a place that isn’t actually all that hot isn’t that surprising, even if it feels like it ought to be. You sit outside and it’s sunny but doesn’t feel particularly hot, and before you know it you’ve been completely fried.

In some places there’s a long lag between peak UV and peak temperature. At 4pm in Britain it can feel pretty hot – the sun has had all day to warm the air – but the UV might be a third of what it was at its peak. But the mid-morning to early afternoon, when the air hasn’t had much time to heat up, feels much less threatening – which is where peak burning occurs.

There are other ways you can be exposed to UV without getting hot. Sunbeds are one obvious one; Iceland saw a sharp increase in melanoma which is attributed largely to indoor tanning. The side windows in cars and aeroplane cockpits block UVB but transmit a lot of UVA, hence the finding that pilots have twice the melanoma risk of the general population. (Laminated glass, like in car/cockpilot windshields, blocks both UVA and UVB.)

Some places expose you to more UV than just their latitude would suggest. New Zealand, for example, gets 40% more UV radiation than countries of the same latitude in the Northern hemisphere. This is due to a combination of the sun being closer during the Southern hemisphere summer, much clearer skies, and reduced UV-shielding from the ozone layer. As a result it has some of the highest rates of skin cancer in the world.

Places that are very high up are also much more dangerous – an extra 10–12% UV radiation per 1,000m elevation. Mexico City is at 2,240m so on an hot day you’ll burn much more quickly.

Worth also noting that some drugs make you more photosensitive. Doxycycline, frequently prescribed for acne, is one. Thiazide diuretics, a common type of blood pressure medication, are photosensitising too and have a documented risk of increased skin cancer.

Most people will want to use topical retinoids at some point. The evidence for them improving skin appearance and reducing wrinkles and photoaging is very robust, with research spanning several decades. They can make you slightly more photosensitive, so you should be combining it with sunscreen anyway – and being even more fastidious if you’re in a place with lots of UV.

So, citizens of sunny (but cool) places – wear sunscreen!1

Inkhaven is over. I am sad.

Surprisingly: I didn’t actually find hitting the writing that hard. Lots of people, maybe the majority of the cohort, were working right up to midnight, and didn’t have and extra posts in the bank for days when they didn’t want to write.

It’s quite possible that their posts were just far better than mine, and that’s certainly the case for a lot of them. But I think a lot of people at Inkhaven were quite ADHD-brained and just started writing very late in the day. Write early!

I also didn’t find it hard to think of what to write. I came in with a few drafts and a bunch of ideas, but most of what I wrote were ideas I had at Inkhaven. Gwern gave a morsel of advice in the opening session: people say interesting things all the time which they aren’t going to do anything with. Don’t be lazy – pick up those crumbs and turn them into something.

My goal was only partly to write. There were lots of other things I wanted to get out of Inkhaven beyond publishing: making friends, finding a tribe, testing out the Bay Area for a potential move there, seeing how much being in a community was regulating for me. In some ways the writing was a head fake: yes, I have always wanted to write more, but I also wanted these other things which would be very hard to get by just showing up at SFO and trying to pull them out of thin air. The shared goal we pulling together on was what bound us together, made the whole thing an adventure.

I am mostly proud of the things I wrote. One of them was low quality (I made a point of posting eight times on one day to beat Claire’s record but the last one wasn’t great), most of them were good, some of them were excellent. I’ll put my highlights below.

I think the deadline is super motivating for some people, and so they ended up not starting writing until later in the day and going right up to the wire. Too dicey for me.

Being around a group of interesting, smart and weird people did wonders for me. I was the happiest I’ve been in a long time. I met some people I hope to keep as friends for good, but I can already feel those connections vanishing as people scatter to the wind.

The girlfriend of a resident came along a couple of days after we finished. She said the campus was a bit soulless. The odd way it’s decorated, weird rationalist-coded artwork (think Wanderer above the Sea of Fog but with a futuristic post-singularity scenescape in the background), astroturf everywhere. It’s a weird place for sure. But when she visited it was completely deserted, all the Inkhaven residents gone save for a few stragglers, stacks of empty pizza boxes and stray cans left everywhere from a party the night before. I think she would have felt differently if she’d seen it on a sunny day, full of people milling about and chatting and writing. It was terribly sad, seeing the place empty of the people who had made it a home.

I stayed in a room just across from the kitchen. My morning routine became cemented after about a week: wake up, put on my flip flops, walk the twenty paces to the kitchen, grab a filter coffee from the giant urn and talk to whoever was up. Usually it was the same little crew. And then shower and get dressed, maybe have some breakfast, and head to Glass Hall, the semi-secret coworking space with velux windows and a couple of sofas, on one of which I would always find Alec reclining, working on his piece.

By the sounds of it we were a bit less uptight than Inkhaven 1. The drinking didn’t start in earnest until the second week, but the flow increased steadily up to the end (I managed to restrain myself and avoid hangovers). Some amount of ketamine was consumed.

On the final night I brought over a girl I’d met online for a date. She enjoyed the weirdness of the place, singing karaoke with me in the ‘winner’s lounge’ and listening to two of my fave residents have a spirited debate about wealth taxes as we cuddled under a blanket by the fire. The karaoke queue app was running on my laptop so when they changed venues it locked and they needed my fingerprint. Emet came running out of building B, drenched in sweat, screaming “HENRY STANLEY WE NEED HENRY STANLEY’S FINGERPRINT”. I went up to help while he insisted to my companion that this was an AI alignment problem.

There was cuddling. There was drama. There were situationships (three, apparently; I only know the details of two so someone needs to fill me in). Someone got kicked out for posting late but was then un-kicked out but only after a literal show trial. Someone was miserable and hated the place and left overnight. Someone invited over a controversial Substacker who brought with him an entourage of coked-up lunatics who wanted to fight people. Someone got reprimanded for posting too much low-effort slop. We got crammed into a house in Bodega Bay where we slept six to a room and all nearly killed each other (but where I had the best day of the whole programme with lovely Evelyn, who took a few of us out to Jenner and then to the offsite Airbnb which had a hottub).

I don’t know why I’m telling you all this. I guess what I’m saying is: Inkhaven is a weird and interesting social experiment in which a bunch of people who have been chosen solely on the quality of their writing, without so much as a phone interview are thrown together in a big place with lots of snacks and a shared goal.

I don’t know if it’s going to create the next Scott Alexander. But I can sure imagine it causing a lot more thoughtful material being put out into the world.

I wanted it to last forever. It wasn’t like any place I’ve ever been. I regret not talking to people more, not working harder on my posts, not soaking up every minute of it. I know that it couldn’t have lasted forever – am I going to just burn my savings indefinitely to write 500 words a day which will mostly be read by nobody? And I know that its shortness was part of what made it so sweet.

But still. I miss you guys.

My best posts:

Signups for Inkhaven batch 3 are open!

We’d been housemates for maybe six months. You’d come over from New Zealand for a job and found our Peckham group house online, a little community of friendly weirdos. It was a toss-up between two people in the end, and I picked you – you would be working from the office full-time, and if you weren’t a good fit then at least we wouldn’t see much of you. That turned out to be a fateful choice.

You’d come home from a rooftop wine bar in the old factory building on Rye Lane, open for the first time after being closed for the winter. You wanted to show me photos you’d taken of the sunset. You sat next to me on the sofa, sliding a little too close, your thigh touching mine. I could smell the alcohol on your breath.

In that moment some part of my brain rewired itself, and no matter how bad an idea it was I manufactured our next encounter, a house movie night or dinner or some such. We drank and smoked weed. And when the others went to bed, I draped my arm around you, and you leant on my shoulder and we lay down and held each other and after a while you softly kissed my neck and I looked at you and kissed you back.

“I don’t know if we should,” I said, attempting restraint. “That’s what makes it so exciting,” you replied.

⁂

I still don’t understand what happened to me in those following few weeks. The limerence was brutal, totally consuming, until it devolved into something almost psychotic. My mental health fell off a cliff – who develops an anxiety disorder at 32? – to the point that I had to take time off work. You went from hot to cold and back again, assuring me each time we had sex that it would be the last, until I made it conditional on you not saying that again.

A housemate objected vocally to our seeing each other, saying I was going to destroy the houseshare, the home that she depended on – but admitting, months later, that the real reason was her own jealousy and insecurity. You found out and got upset. She must think I’m such a slut, you said, sitting on the end of my bed, head in your hands. You decided to break things off and for some stupid reason I went along with it – a level of passiveness of which I am still ashamed.

But when I realised that would mean the end of something so significant, that ‘maybe in another world’ actually just meant ‘never’, my favourite Rumi quote came to mind – run from what’s comfortable, destroy your reputation, be notorious – and pretty soon I was knocking on your door again.

I knew you’d be returning to New Zealand in a few months, though part of me hoped you would stay. I prepared to move out – it didn’t seem wise to be seriously dating a housemate, and I wanted to make the most of our time together, give things a shot. Quietly I wondered how practical remote work might be with a twelve-hour time difference.

And then, quite suddenly, you announced you had quit your (prestigious, high-status, yet dreadful) job and were going to travel round Europe. Your November departure became August, then late July, then early July. I couldn’t understand why you wanted to cut things short, when most nights (and mornings) we’d be sharing a bed or the sofa.

I suggested I could join you for some of your trip, but you wanted your independence, and I knew better than to press.

⁂

It was just us in the house on that final summer’s day. I woke up at 5am, watched you sleeping next to me. I made you breakfast. You asked for a photo of you wearing your huge hiking rucksack by the open front door. I gave you a note I’d written the day before, my handwriting shaky from anxiety; told you not to read it until you’d crossed the border. Then I said goodbye.

I went to my room and looked at the photos I’d just taken. In one of them you’re laughing, mouth open, nose wrinkled up in that trademark squint. It’s how I picture you, even now.

I panicked. Maybe I could stop you, or delay you a day or two. I put the phone down, threw the door open and ran after you, but you were already gone.

⁂

I can point out on a map, from memory:

The town where you grew up, the house you spent your adolescence in, the high school you went to, the hill you had to climb at the weekend to get cell reception to see if any boys had texted you, the forest where you went bush bashing and got lost and nearly died. The firebreaks you rode your dirt bike up and down as a teenager. The river you were tubing down when you were twelve and your friend broke his arm and it hung limply at a sickening angle. That river ran past the campsite your parents owned, now sold and closed; in the summers you helped tourists pan for gold there.

Why am I like this.

For a long time there wasn’t a day when I didn’t think about you, about the time we spent together. About the first time we got together, the unhinged sexual chemistry we had; how when you got back from a work trip you came to my room and instead of speaking we just started stripping each other naked. The time we both cried with laughter, trying and failing to explain to a housemate the premise of that show where they make cakes that pretend to be things. The time you called me up to your room and you were standing fully nude wanting help with your fake tan before a girls’ trip to Ibiza. How we’d spoon on the sofa and watch Stranger Things and I’d hold you when you jumped.

The time we lay in bed together in the dark as a summer rainstorm battered the windows and talked about whether we had a future. And the next time we talked about it, when you told me that we didn’t.

You came home one night, before we got together, drunk as a skunk on “cab sav” and unable to find your keys. Pounded on the front door until I woke up. I went downstairs, shirtless, and found you slumped on the ground with your phone in your hand. “I’m so cold, Henry.” You held me close, and handed me the phone so I could tell whoever was on the other end that you weren’t going to die of exposure.

In the days leading up to your departure, you packed your stuff, leaving behind things you didn’t want. You offered some of them to me. I took them because they reminded me of you, and got rid of them for the same reason.

I kept the t-shirt of mine you wore the last night we were together until it stopped smelling like you. In the end, all I have left is a business card from that job you hated, a scrunchie you left on my nightstand, and a photo I took of us on the sofa together. Months after you left I printed it out, and on the back of it I wrote maybe it isn’t all hopeless bullshit.

“You’re a real catch,” you once said, peering at me with hazel eyes. “You’ll make someone very happy someday.”

⁂

When you fall in love with someone, you fall in love with how they see the world, and so I found myself going on weekend hiking trips to Snowdonia, to surf school in Newquay to catch waves on the freezing Atlantic, interrailing around Europe; mimicking your adventures in an attempt to capture some of the joy and curiosity with which you moved through life.

But with you gone, I was chasing a ghost.

You once showed me the necklace you liked to wear: a New Zealand fern in 18-carat gold. You bought it with your first paycheque. I loved the symbolism of deep commitment to your adopted home, with its sparkling lakes and piercing mountains. I can’t imagine feeling that way about England, a place where it increasingly feels like my time is up.

I once heard that, as adults, we try to be whatever we weren’t in high school. In your case, that was cool and rich. So in London you courted a set of friends straight out of Mean Girls; people who cared a great deal about being thin, who dated your ex behind your back, who abandoned you on holiday when you were sick, and who kicked you out of their group chats as soon as the wheels were up.

It’s funny because one of the things I loved about you was how conscientious you were, how civic-minded. You cooked us vegan meals, you ran a book club, you helped people on the street. You managed to avoid the hard shell that most Londoners adopt after living here for a while, retreating to their inner worlds when they leave the house and not caring to interact with anyone outside it.

I was an anxious kid. I cared about getting good grades, getting into a good university, getting a top degree. At some point that fell away and I learned not to worry so much, that none of that stuff counts for nearly as much as everyone makes out. And when I realised just how much time I wasted being anxious, I risked becoming cynical and sneering.

Like so much about you, I found that necklace intimidating. It showed how much you cared about building a life from the rubble of your traumatic upbringing, the long shadow of which you were still grappling with. I couldn’t imagine caring that much about anything.

⁂

I didn’t hear much from you once you left. I thought you might want to keep in touch, but I was too proud to reach out. When I finally did message, you said that you weren’t coming to London on your way home as planned, but you’d happily talk on the phone – until something changed and that, too, was out of the question. The scope of our relationship, such as it was, shrank until it could fit into a text message.

So I said goodbye for good. That you meant such a lot to me, that you should look me up if ever you’re passing through. You never did. Your final message: “Thanks, all the best,” and then your profile picture went blank and I didn’t hear from you again.

I ended up moving out of that house. When I opened the storage locker in the garden, I saw the blue bike you left behind. And there, in the helmet hooked on one of the handlebars, was a strand of frizzy auburn hair. Proof you were here, that it really happened, that I loved you once.

A reminder that you’re out there, on the other side of the world, living out your life with people who adore you.

⁂

I’m glad our paths crossed, for a bright, brilliant moment. I’m not sure I could do it again. But part of me will always long for the land of the long white cloud.

Thanks to Lily Blackburn, Jasmine Ren, Matt L., Drew Schorno, Emma Baker and Alexander Wales for their feedback and encouragement to post this after three years in my drafts.

Thanks also to Hauke Hillebrandt for putting up with a summer of me being insufferably heartbroken.

In 2014, I applied for a job at the Centre for Effective Altruism. I don’t remember the actual title – probably a marketing position, as at the time I was working in life science/pharmaceutical public relations.

I’d come across effective altruism not long before, probably initially through Eliezer Yudkowsky’s essay Money is the unit of caring, in which he suggests that someone who is taking doing good seriously would think like a businessperson, and in most cases opt to make money instead of doing the thing themselves.

This really changed my view of how to think about doing good. I discovered effective altruism not long after.

I didn’t get the job at CEA. I was sad not to get it, but it wasn’t a huge knock. I took an 80,000 Hours career consultation around the same time, and they suggested I become a software engineer, a career move I was already leaning towards. I quit my shitty PR job – which was sufficiently mind-numbing that I was already outsourcing much of it to a Nigerian lady on Upwork – and signed up for Makers Academy using money I’d saved from running a separate PR consultancy on the side. I got my first tech job three months later.

I went along to the pub with a few friends for what turned out to be a sort of CEA old lags’ drinks – some folks who had worked there at various times and in various roles (one who was central to organising the first EA Global conferences, another who worked on the community-building programme, another who used to be the CEO).

I found out from one of them that they had, in fact, been on the panel which reviewed my job application and had given me the green light, but some other unnamed person had said no. I had been one ‘yes’ away from getting the job.

The person who did get that job was sitting at the table with me. And without giving too much detail, they’d had a very significant impact on the organisation and its structure in quite a dramatic way. All of the other people at the table had been around for that tumult and were downwind of it.

And of course for each of these people working at CEA was a path dependency for the various things that came next.

One person went on to do career coaching for ambitious people seeking to have an impact with their work. Another stayed in the EA sphere, working on a high-impact crowdfunding platform, flying to the Bahamas to do research under the FTX Foundation umbrella, and receiving grants to do AI safety research. Another ended up working for a crypto hedge fund that branched off from Alameda Research, which later ran as the sister company to FTX.

It felt like – even excepting the aforementioned drama – that their time at CEA had been at once exhilarating and exhausting, and that they all shared a sort of collective trauma from it. But that working at the vanguard of the effective altruism movement when it was but a fledgling was a highlight for all of them; a job whose camaraderie was unmatched before or since. Strange to think I came close to having that.

Getting an intuitive understanding of what animals go into which diets, and their welfare, is difficult. Last week I shared an interactive essay on the subject.

In this instalment, I’ve updated it significantly to include:

• details on how animals live and die when we farm them for food,

• a diet calculator showing how different diets fare in terms of raw animal ‘life-days’ that they comprise, and the number of ‘suffering-days’ when weighted by how bad their lives are,

• welfare estimates for farmed shrimp, and

• a closing section on what these numbers might mean.

Read the updated essay here.

[Updated copy for Inkhaven word count purposes]

Different diets translate into very different numbers. Pick a diet to see its totals, and use “Compare to” if you want to see two diets side by side.

Fish consumption dominates the number of life-days in their diets — 90% coming from fish alone.

Americans consume less fish than Europeans, but much more chicken — 46kg per person per year. This makes chicken the single biggest source of suffering.

Brits eat slightly less fish than on the continent, and slightly less chicken than Americans.

Vegetarians don’t eat meat or fish, cutting out much of the suffering in their diets. But 300 eggs a year still produces over a thousand days of suffering. Replacing beef with eggs is probably bad on net from a welfare perspective, which isn’t totally intuitive.

Vegans don’t consume animal products, so their numbers here are zero by definition. Like all diets, crop farming still has indirect impacts on animals (rodents, insects, displaced wildlife) that aren’t modelled here.

These are relatively simple models — they account for how many animals get eaten in a given diet, how many days of each animal’s life that corresponds to, and how much those animals likely suffer. The answers depend on which weighting you trust, and reasonable people will land in different places.

There is a lot we haven’t quantified: the wild fish that get fed to farmed salmon, male chicks which are killed at layer hatcheries, the difference between factory-farmed, free range, and pasture-raised animals, and the thorny question of whether shrimp even feel pain. It’s worth noting that most of these would push the numbers in the direction of _more_ suffering, not less, so the numbers above are somewhat conservative estimates.

Directionally, chicken, eggs and fish make up the bulk of the welfare impact of people’s diets. Beef and dairy make up a much smaller proportion. Switching out chicken, fish and eggs (in decreasing order) will have the biggest impact on one’s diet. Many people switch out beef for chicken for various reasons, including to reduce the climate impact of their diet. But counterintuitively, this is actually one of the worst moves to make in terms of animal welfare.

Eating is very personal and has great cultural significance. A Christmas or Thanksgiving turkey, a Sunday roast, or even a cheese toastie all have strong emotional salience for lots of people. None of the above is intended to be proscriptive. But hopefully it’s given some insight into an otherwise unintuitive — but enormously important — subject.

If the idea that animals’ experiences might differ this much already changes how you think, consider donating to effective organisations which seek to reform the way we raise animals for food, pay for ‘offsets’ to cover the animals you consume, or think about where you could cut back.

Numbers here are tentative.

It also means you’re out $100 a month, which doesn’t leave much left over for hosting. Here’s how you can run your projects on the cheap.

Serious projects – use a PaaS

Anything that needs to work, gets significant traffic, which has paying customers

For serious projects (especially ones you’re charging money for and get significant traffic) you should probably just use a proper platform-as-a-service like Heroku. It’s expensive for the amount of compute you get, but they handle deployments, monitoring, logging, add-ons and so on, leaving you to sleep at night.

Small projects

Projects which are getting some amount of traffic but can tolerate some downtime

Fronend-only

Anything that doesn’t need a backend can just be deployed to GitHub Pages or Vercel for free. Vercel gives you free analytics; on GitHub Pages you can add a Google Analytics (or similar) tracking code. You can run builds locally or just do them using GitHub Actions.

You would also be surprised how many things can be run with just a frontend. retreatsfyi was initially a Rails app with a few features that required backend (like setting up alerts). But for a project that’s so small those features weren’t critical, so now I have Claude scrape retreat centre websites, process the data locally into a static site, and then just deploy that, doing filtering on the frontend.

Projects with a backend

Anything that is backendful will need an actual server to run it. But if you’re a solo operator, you don’t actually need a real CI and deployment pipeline, because Claude can do it all for you locally.

Sign up for Hetzner and buy a cheap box. You can get a surprisingly capable one for about €6 a month. Generate an SSH key locally and launch the machine with the corresponding public key. From there, Claude can interact with it from your local machine.

Claude Code can write you a docker-compose file, SSH onto the box, set it up, pull your repo (or scp it) and run it. The benefit of using a VPS rather than a PaaS like Heroku is you get permanent storage, so you can store files locally and use a database like SQLite.

Set up a ‘/deploy’ skill in your project which runs your tests, and if they pass, SSHes onto the server, does a git pull, and runs docker-compose up.

Multiple projects on the same box

Use Caddy as a reverse-proxy. Claude will set up a Caddyfile for you, with multiple stanzas, one for each project. Just make sure it knows that other projects are running on there.

Backups

Use restic and AWS S3 to back up the box; Claude will write the cron job.

DNS

AWS Route53 can be interacted with using the aws CLI, so you can ask Claude to manage this for you once you’ve added your domain.

If you don’t want to have to buy a domain for each of your projects, just buy a general root domain and use subdomains for each of your projects.

Cloudflare has a free tier if you want to use them for DNS and get DDoS protection out-of-the-box, but that’s probably overkill.

Sending email

Use Postmark. You get 100 free a month.

Happy coding! 🧑‍💻

Charles Mackay’s 1852 book Memoirs of Extraordinary Popular Delusions and the Madness of Crowds chronicles a variety of delusions and fads that gripped people of the time. He notes:

Whole communities suddenly fix their minds upon one object, and go mad in its pursuit; that millions of people become simultaneously impressed with one delusion, and run after it, till their attention is caught by some new folly more captivating than the first.

Mackay puts the boot into a variety of charlatans: alchemists, fortune tellers, modern prophets, tulipmaniacs, and monopolists.

Most intriguing of all are the Magnetisers:

[As] alchymy began to fall into some disrepute … a new delusion, based upon this power of imagination, suddenly arose, and found apostles among all the alchymists. Numbers of them, forsaking their old pursuits, made themselves magnetisers. It appeared first in the shape of mineral, and afterwards of animal, magnetism, under which latter name it survives to this day, and numbers its dupes by thousands.

He deftly draws a through line from Paracelsus (who describes a remedy involving magnets being rubbed on the sick patient, then buried with seeds so as to transfer the disease to them), Kircher the Jesuit (who had a hernia patient swallow powdered magnet while wearing an iron filing poultice), Mesmer (whose salons involved water bottles filled with iron filings laid out in a circle, which caused the attendees to flush and convulse), and further nonsense through the 19th century (a magnetised elm tree, magnetised steel pins to cure gout, “never magnetise before inquisitive persons”).

He divides this history into two sections.

The mineral phase starts with Paracelsus travelling through Persia and Asia to find the legendary “mountain of adamant”, selling a magnetic nostrum, and “transplanting” diseases out of the body and into seeded earth:

If a person suffer from disease, either local or general, let the following remedy be tried. Take a magnet, impregnated with mummy, and mixed with rich earth. In this earth sow some seeds that have a congruity or homogeneity with the disease; then let this earth, well sifted and mixed with mummy, be laid in an earthen vessel; and let the seeds committed to it be watered daily with a lotion in which the diseased limb or body has been washed. Thus will the disease be transplanted from the human body to the seeds which are in the earth.

Kircher the Jesuit was likewise a believer in the power of magnetism.

He directed [a man with a hernia] to swallow a small magnet reduced to powder, while he applied at the same time to the external swelling, a poultice made of filings of iron. He expected that by this means the magnet, when it got to the corresponding place inside, would draw in the iron, and with it the tumour; which would thus, he said, be safely and expeditiously reduced.

As word of these miracle cures spread, a sort of ‘inverse magnetism’ appeared, in which any wound inflicted with a metal weapon could be treated with “weapon-salve”, a concoction made from

moss growing on the head of a thief who has been hanged and left in the air; of real mummy; of human blood, still warm … of human suet, two ounces; of linseed oil, turpentine, and Armenian bole[.]

This mixture was to be saved in a narrow urn, and then the weapon which inflicted the wound should be anointed with same salve while the wound was washed with “fair clean water”.

The remedy was brought to England with much fanfare by Robert Fludd, who at 25 went to Europe and became a disciple of Paracelsus’ teachings. On applying the methods he learned abroad he had much success, although Mackay notes this was likely because of his fastidious wound treatment rather than any magnetic sorcery:

[W]hile he kept up the spirits of his patients by boasting of the great efficacy of the salve, he never neglected those common, but much more important remedies, of washing, bandaging, &c. which the experience of all ages had declared sufficient for the purpose.

The magnet craze continued through the 18th century until we reach Father Hell, a Jesuit who strapped steel plates to his patients’ naked bodies to treat various diseases. In 1774 he passed his methods down to Anthony Mesmer.

Thus began the animal phase: with Mesmer and his theory of “animal magnetism” or “mesmerism” as it would come to be known. Mesmerism revolves around Lebensmagnetismus, an invisible magnetic life force found in all living things (including vegetables). He had Father Hell make him some magnetic plates, which he used on patients apparently to great success. Hell tried to claim these successes as his own, with Mesmer mentioned as a footnote. Mesmer was having none of this and accused Hell of breaching his confidence, claiming the invention as his own. The quarrel was the talk of Vienna for some months, with Hell coming out on top.

Nonetheless, Mesmer continued his magnetic crusade. He treated a young woman suffering from a rush of blood to the head and attacks of syncope, using the magnetic plates to great effect, but then discovered he could do away with them and simply use his hands:

When trying the metallic plates of Father Hell, he thought their efficacy depended on their form; but he found afterwards that he could produce the same effects without using them at all, merely by passing his hands downwards towards the feet of the patient, even when at a considerable distance.

His methods were not well-received in Vienna, so he travelled to Switzerland where he had much more success, meeting Father Gassner who similarly could heal the sick by the laying on of hands. He learned the ropes from Gassner, and was able (by his own account) to heal some of the patients Gassner passed to him. He returned to Vienna, still without much fanfare, so moved to Paris, where mesmerism would become very much in fashion.

He ran sessions from his “charmingly furnished” home – stained glass, mirrors, the scent of orange blossom. Mackay describes the events:

In the centre of the saloon was placed an oval vessel ... in which were laid a number of wine-bottles, filled with magnetised water ... covered with an iron cover, pierced through with many holes, and was called the baquet. From each hole issued a long movable rod of iron, which the patients were to apply to such parts of their bodies as were afflicted ... Then came in the assistant magnetisers, generally strong, handsome young men, to pour into the patient from their finger-tips fresh streams of the wondrous fluid ... staring them out of countenance to magnetise them by the eye! ... Gradually the cheeks of the ladies began to glow ... and off they went, one after the other, in convulsive fits ... In the midst of it, the chief actor made his appearance ... Dressed in a long robe of lilac-coloured silk richly embroidered with gold flowers, bearing in his hand a white magnetic rod, and with a look of dignity which would have sat well on an eastern caliph, he marched with solemn strides into the room.

He made quite a stir, and even performed a controlled experiment on twenty-four patients at the Faculty of Medicine, half of whom would be treated magnetically, and the other half treated using the usual methods. But he made conditions on the assessment of the treatment’s effectiveness that were strict enough that the faculty did not agree to them.

Mesmer’s reckoning came even without performing the experiment which would surely have debunked his methods. In 1784, Louis XVI assembled a joint commission of the Faculty of Medicine and the Académie des Sciences – whose members included Benjamin Franklin, Antoine Lavoisier and astronomer Jean-Sylvain Bailly – to investigate the claims. Mesmer did not attend the sittings.

The commission ran experiments to investigate Mesmer’s claims. They magnetised some objects and not others, then asked patients to identify which was which. They also tested the effects of telling patients they had been magnetised while doing no such thing (today we would call this a “sham procedure”). When patients knew they were being magnetised, the effects appeared regardless of whether any magnetic chicanery was actually happening.

The commission eventually published their report, drawn up by Bailly:

After detailing the various experiments made, and their results, they came to the conclusion that the only proof advanced in support of animal magnetism was the effects it produced on the human body—that those effects could be produced without passes or other magnetic manipulations—that all these manipulations and passes and ceremonies never produce any effect at all if employed without the patient’s knowledge; and that therefore imagination did, and animal magnetism did not, account for the phenomena.

With his reputation in tatters, Mesmer left Paris (albeit with 340,000 francs of subscription money) and retired comfortably to Swabia. Nonetheless he spawned a dozen imitators of his methods, who promulgated his beliefs to the modern day. His ideas, debunked as they were, comfortably outlived him.

The 1784 commission had run perhaps the first placebo-controlled trial in medical history to debunk the myth of the Magnetisers. Yet a cottage industry continues to push magnetic chicanery, with “magnotherapy” rings, bracelets, wristbands selling tens of millions of units a year. And for those with furry friends even magnetic dog collars are available – although presumably the owner’s imagination is doing more work than the dog’s.

Thanks to Alec Thompson for pointing me towards this subject.

I went to Inkhaven, a writing retreat in Berkeley where you must publish 500 words per day or else.

We actually discovered that or else is a real thing when one of the residents got kicked out for submitting at 12:02. After a dramatic performance where said resident was placed in stocks and put on trial he was accepted back into the cohort.

One resident did get kicked out for not posting, although he decided to leave for personal reasons, so he left overnight.

General

• You are here to write, but not just to write. You’re in Berkeley on the incredible Lighthaven campus (which you should explore all of), and surrounded by 50+ clever and beautiful people from all walks of life. Get to know them, have deep conversations with them, make out with them (or not, I don’t care).

• Batch 2 was much more diverse than batch 1. Batch 1 was very rationalist/EA/AI-pilled; batch 2 was still all of those things but with more variety – we had “real writers” too. Poets, novelists, playwrights, famous bloggers. It’s possible batch 3 will go further in that direction.

Social

• Meet everyone! Do this sooner rather than later; you don’t want to meet someone right at the end and find you don’t have time to get to know them properly.

• Avoid falling into routine conversations as time goes on. You’re not going to be around these people for long; don’t just ask them what they’re writing about. Probe a bit more deeply, ask about their home lives, about why they’re at Inkhaven and how they found out about it, about their personal stuff.

  • I learned some of the most interesting and surprising things about people in the last few days of the residency. You never know what someone is like, so don’t make assumptions!

  • Everyone here is smart and weird and wonderful, and there’s so much to find out about them and so many ways to connect with them.

  • Don’t limit your social connections to the people who are generally in your sphere. There’s much to be learned from the residents you wouldn’t normally interact with.

• Some people will just never be around. They’re either not living at Lighthaven (or the nearby accommodation) – perhaps they have a family – or they just don’t want to hang out. So you won’t be able to actually meet everyone.

• Live on campus. The nearby accommodation is nicer for the money, and isn’t very far away, but there’s something magic about being able to open your door, walk ten yards, and be in the throng right away.

  • My mornings looked like: wake up, put on sweatpants, walk to the kitchen, pour myself a coffee, talk to whoever was up (usually quite a lot of people).

  • It’s possible your writing will be better if you don’t live on campus – the resident who produced some of the best-liked writing was living in one of the off-site houses – but I still think it’s a tradeoff worth making.

• Get a feel for the rhythm of the place. Some people have quite different daily routines.

  • I was surprised how quiet it was after dinner until late, having published my piece for the day. Lots of people worked right up to the line so were often just not around in the evenings. If that’s the case for your batch, maybe aim to do more socialising during the day.

  • Lunch was the most sociable meal. If you’re going to order one meal a day and just snack or eat the free bagels the rest of the time, order lunch.

  • There was a person I hadn’t met; another resident said: “she’s the one who’s always sat at a booth in the kitchen by 8am with an empty French press and fifty poems written.”

• Play social connection games. Someone organised a game of Hot Seat, where someone is nominated to answer questions from the audience of varying levels of spiciness (they can of course decline to answer a given question). Make your questions silly, probing, or just straightforward.

Writing

• You do actually have to write and publish 500 words per day or you get kicked out. There was lots of legalesque debating about whether 12:00:59 the following day was acceptable, so the rules have since been updated to specify that 12:00:00 is the deadline.

  • What actually happens if you get kicked out is undefined, annoyingly. If I get kicked out do I have to actually leave the next morning? Do I get a refund on the accommodation I paid for?

  • But also, please don’t submit stuff right at the deadline. Lots of people did this (30% of posts went up in the last hour of the day) and I just couldn’t understand it. It’s pointless: it means you don’t have time to read other people’s writing, have less time to socialise in the evening, and probably will end up writing less well.

• Show up with a list of things to write about. Put them in the Inkhaven Tracker I made (not affiliated with Inkhaven)!

• Show up with a bunch of drafts. You’re allowed to do this so long as you do at least 500 words of editing on them. I had a piece I’d been sitting on for about three years which turned out to be the best I published (if you’re reading this on April 29, it’s not been published yet!)

• You are allowed to get ahead and write drafts for other days, which I hadn’t realised. Definitely do this, preferably early on when you have more stamina. Having even one or two in the chamber means you can just hit publish if you don’t have anything in you, which really takes the pressure off. I had another commitment mid-way through which took most of the day and it was good to be able to just press the button.

• Don’t use your buffer to coast. Use it to allow yourself a day off if you’re burned out, or to work on a post for several days which you otherwise wouldn’t be able to focus on.

• Experiment with different kinds of writing. Get out of your comfort zone; if you always write in-depth research, try writing some personal non-fiction (aka squishy personal stuff; really plumb the depths). If you always write persuasive rhetoric, try writing a short story.

  • The team added a challenge to write one piece in each of seven categories: empirical (a survey or experiment), informational, persuasive, humour, advice, fiction, and personal. If you hit all seven you got a cool enamelled pin.

• Dig deep. Expose yourself more than you would normally. Some people are here to get over their fear of hitting the ‘publish’ button. The fear of hitting that button when you’ve just written something about your last two breakups, or about catching the eye of a cute girl on the street, or about lost love – that’s ten times worse. Learn to overcome it. Those might be some of your best posts.

• The kinds of things you post will be shaped by how the other residents respond to them. I think this is inevitable. I was surprised by how quickly I forgot that my friends back home were subscribers and ended up writing stuff I thought would impress other residents (and get lots of likes). Probably fine; you can reset your reward function when you leave and it’s fun to compete for attention locally.

• Pay for a Claude Max subscription. You’ll get a lot of mileage out of it – bouncing ideas back and forth, getting line edits on final drafts, asking for topic recommendations based on what you’ve already written.

  • I used Claude Code to write a script which scraped all of my Substack posts. I then set up a Claude Cowork project in that folder so it always had context on what I’d written.

Misc

• As much as you can, read the writing of other residents. I spent hours doing this in the first couple of days but it’s unsustainable to read the output of 54 other people. That said, skimming the recent posts of the day is a great idea and gives you a way to connect with others through their writing.

• Go for a walk now and then. There were stretches of several days where I didn’t leave the campus or exercise, and that’s probably not great. The UC Berkeley campus is nice for a stroll.

• Inkhaven can be many *-havens. Slophaven – in the middle of the month when people’s energy dips and they just pump out low-effort posts. Brinkhaven – the inevitable deadline panic that people seemed to weirdly love. Drinkhaven – although the stock of alcohol by the end was dire, and avoid overdoing it and being hungover. Situationhaven – I will say no more about this. Kinkhaven okay this didn’t actually happen but I’m hyperstitioning it.

Inkhaven has been one of the best months of my life. Some people got financial aid to be here; I paid full whack and it wasn’t cheap, but it was absolutely worth it. If you’re on the fence, go for it. I would do it again in a heartbeat.

Scientists are a strange bunch. In the name of advancing the frontier of human knowledge, they sometimes put themselves in harm’s way: trying unknown substances on themselves, getting stung by various poisonous animals, drinking Petri dishes of bacteria, and threading catheters into their own hearts.

Here are a few I particularly enjoyed.

Discovery that a melanocortin regulates sexual functions in male and female humans

Mac E. Hadley, Peptides, 2005.

Melanocortin agonists bind to a range of melanocortin receptors, which regulate appetite, melanin production, and sexual arousal. So they make you thin, tanned, and horny.

The author, having developed a superpotent melanocortin agonist (“melanotan”) and tried it on himself, reports:

Unlike MTI, however, MTII caused a rather immediate, unexpected response: nausea and, to my great surprise, an erection (no figure provided). While I lay in bed with an emesis pan close by, I had an unrelenting erection (about 8 h duration) which could not be subdued even with a cold pack. When my wife came upon the scene, she proclaimed that I “must be crazy.” In response, I raised my arm feebly into the air and answered, “I think we may become rich.”

Effects of Sexual Activity on Beard Growth in Men

Anonymous, Nature, 1970.

An anonymous author writes about the effects of isolation for several weeks on a remote island. He noticed that during these stints, his beard seemed to grow more slowly, as shown by a reduced weight of beard clippings during this period.

It’s an n=1 study but found some interesting things: beard growth had a diurnal pattern (presumably related to the diurnal pattern of testosterone secretion?), beard growth was stimulated by both exogenous testosterone and cortisone, and “even the presence of particular female company in the absence of intercourse … usually caused an obvious increase in beard growth.”

That said, he also observed that shaving more often caused his beard to grow more, which is a common folk tale (how would the hairs, which are made of dead keratin, “tell” the follicles to grow more quickly?)

The pronounced effect of sexual intercourse on beard growth may therefore be caused by an increase in testosterone secretion, which is known to occur in response to coitus.

Versuche über Cocainisierung des Rückenmarkes

(Experiments on the Cocainisation of the Spinal Cord), Bier, 1899.

At the time this paper was published, cocaine was used for local anaesthesia (as it still is today in ENT surgery, although has largely been supplanted by other anaesthetics). Bier essentially invents spinal anaesthesia, performing a lumbar puncture into which cocaine is injected.

The paper documents several medical cases in which “cocainisation” was performed: to anaesthetise a man while his ankle joint was sawed through and resected, chiselling open the tibia, bringing out fractured ends of a broken femur to drain a suppurating would.

The most interesting part of the paper comes at the end, when he injects his colleague in the same way, and then does various horrible things to him to observe his pain response: crushing the skin with forceps, putting a lit cigar on his leg, ripping out his pubic hair and crushing his testicles.

After 11 minutes: in the arm, the sensation of pain is markedly diminished.

After 13 minutes: a burning cigar applied to the legs is felt as heat but not as pain. Ether produces a sensation of cold.

After 15 minutes: tickling of the sole of the foot is no longer perceived as such, but only as touch. Pinching of the leg is felt as light pressure; pinching of the upper parts of the chest, as severe pain.

After 18 minutes: from the nipples downwards, even strong pinching is scarcely felt any longer.

After 20 minutes: the pulling-out of pubic hairs is felt only as the lifting of a skin fold; that of chest hairs above the nipples, by contrast, as vivid pain. Forced bending-back of the toes is not unpleasant.

After 23 minutes: a heavy blow with an iron hammer against the shinbone is not felt as pain.

After 25 minutes: strong pressing and pulling on the testicle is not painful.

Who needs enemies when you have friends like this? The paper describes how the colleague has trouble sleeping and “felt very wretched” the next day.

On being one’s own Rabbit

J.B.S. Haldane, Possible Worlds, 1929.

The whole chapter in Haldane’s Possible Worlds is extremely quotable, and Haldane was a mensch.

On biochemists:

The biochemist provides, so to say, chemical splints for damaged organs.

He moved away from animal experimentation to self-experimentation, hence the title of the chapter:

One might, of course, have tried experiments on a rabbit first, and some work had been done along those lines; but it is difficult to be sure how a rabbit feels at any time. Indeed, many rabbits make no serious attempt to co-operate with one. I except always a large buck called Boanerges (which is, being interpreted, the Son of Thunder). Boanerges had to breathe carbon monoxide every day. He sat on the table with his nose in a well-greased funnel. When he got bored he stamped. If one took no notice of his first stamp he proceeded to walk off. However, he was always willing to co-operate for such a period as he thought reasonable; but most rabbits get frightened … A human colleague and I therefore began experiments on one another.

Curiously he and his colleagues ended up doing a lot of weird breathing experiments, trying to reduce the level of carbon dioxide in their blood to show that it regulates normal breathing. These experiments mostly involve breathing fast and deep.

As any freediver will know, this purges that carbon dioxide in your blood, allowing you to hold your breath for longer, although this has the effect of blunting your hypercapnic response, so you can easily pass out doing this.

Poor Goldman on one occasion, after about half an hour, uttered a shrill cry and went into a general convulsion. Every muscle in his body was contracted, his limbs stretched out stiff, and his back arched.

In this vein he also exposed himself to airtight rooms with high concentrations of carbon dioxide, finding that it makes you rapidly feel out of breath even as you’re getting enough oxygen (the hypercapnic response in action). This escalated to drinking weak solutions of hydrochloric acid in order to keep the blood acidic, although getting his hands on it required “[thinking] of a dodge for getting the hydrochloric acid under false pretences.”

The convulsing that his subjects (colleagues) experienced were oddly similar to the symptoms of tetanus. As dissolved carbon dioxide is acidic, this breathing technique makes your blood more alkaline, suggesting a link between alkalosis and tetany. The effects were reversible by administering ammonium chloride, which was converted into urea in the body and releasing hydrochloric acid in the process. Modern-day treatments have moved on to treat the root cause of the alkalosis, but Haldane’s work is still taught in medicine to show the physiological principles of tetany.

There was also the non-execution of the one Inkhaven resident to be thrown out for posting at 12:00:48. Make an example out of him, I say.

Speedhaven

I helped run Speedhaven, which saw four people who had to speed-write head-to-head on a specific prompt. For this I wanted to show everyone’s work in real-time on a big display, and let the audience vote on their top picks.

Writers were given a laptop pre-loaded with the ‘writer’ page, from which they could enter their name to join. Once everyone was in, I could start the round using an admin interface on my phone, and each was shown an editor which would update the view on the big screen in real time. (The writers were protected from heckling by wearing noise-cancelling headphones; turns out it’s basically impossible to write coherently when a group of people is shouting at you.)

Claude did a pretty good job here although it needed a few bugfixes and layout tweaks. I gave it screenshots of the existing Inkhaven website to copy the style, which it diligently did.

Aaron Kaufman (who has an amazing blog to which you should subscribe right now) did a great job MCing and polling the audience for great prompts to torture our writers, including “Should cat ears be mandatory?”, “Yelp review of the Garden of Eden”, and “Bottomless pit supervisor”.

You can read the essays here.

Brinkhaven

I was perplexed by how many people worked late at night on their posts, either because they were putting a shitton of effort in (dope) or because they started really late (cringe) or because they just need to feel something and submitting at 11:59pm gives them a rush (based).

Luckily you can export all the posts from the writers’ dashboard, so I put together a one-page analysis of people’s posts to see how many people like to ride the ragged edge of destruction. Some of them came in three seconds before the deadline. Those people are crazy.

Witness the madness here.

Inkbook

One I wish I’d built – a ‘facebook’ in the traditional sense of a book of faces of everyone here. You get shown one face, you have to recall the name of the person and type it in, and if you get it wrong you get shown the face again until you remember it, spaced-repetition style.

I actually did build a prototype of this, but so many people on the portal didn’t use a picture of themselves that it didn’t really work. I think future Inkhavens should require people to use an actual photo!

Most rentals have whitewashed walls and rising damp; this one was owned by an old lady who had moved to Spain in her dotage to escape the bleak London winters. It was well-appointed, with an open-plan living room and kitchen, trinkets on shelves accumulated over decades, and a patio for barbecues in the summer.

The residents at Larkhill were friendly. Noah and Olivia, a couple from California, and Richard, a quiet man with narcolepsy who no-one saw much of. We had all chatted briefly and they seemed happy to have me for the next few weeks while I figured out my work situation.

It was time for the Sunday journalling session – they did them every week. We retired to different corners of the house to write, and then returned to the patchwork of sofas in the living room. While I was upstairs they had been pushed together to make a single, large surface, on which we all lay. The narcoleptic did not join us.

Olivia went first, sharing some of the things that happened during her week – a few clients at her therapy practice, some disagreements with her boyfriend Noah.

Noah was next. He mentioned the “Art of Accomplishment” course he was doing. I googled it later. The head coach was a man called Joe Hudson, a man in his fifties with a perfect white smile and close-cropped hair. The website showed him on the Modern Wisdom with Chris Williamson podcast, two gleaming smiles beaming at each other. There was a carousel of shortform videos, each with a title like Beyond discipline, Do the scary thing, Anxiety is excitement.

The website had a testimonial from Sam Altman, extolling Joe’s virtues. This did not strike me as a good thing.

“I’ll be doing my anger release at 7pm tomorrow, so you might want to wear headphones,” Noah said.

We moved on before I had time to ask what this involved, or why I had to wear headphones. Not wanting to seem perturbed, I didn’t return to it.

The following day, my phone buzzed as I walked through the door – a calendar event, anger release wear headphones. I rummaged around for my headphones and put them on.

I didn’t hear anything from upstairs. Noah and Olivia shared the attic room at the top of the house, so perhaps the headphones and distance meant I wasn’t going to hear whatever shouting was involved.

After about twenty minutes, I felt a bump on the floor, then another – like something heavy was being lifted up and dropped on the floor over and over. I lifted the headphones off one ear. I could hear shouting from upstairs but couldn’t make out any words. I figured this is what anger release meant – getting your emotions out by making some loud noises and throwing a medicine ball around or something. The sounds stopped after a few minutes.

Next Sunday, the same notice from Noah – anger release at 7pm, wear headphones. Was this going to be every week? I mentioned this, mostly out of curiosity, to Olivia. What’s this about? Why does Noah need to do all this screaming, and why in the house? She cut me off, and asked me what trauma I’d encountered that meant I didn’t like people authentically experiencing their emotions, so I left it at that.

This time the noises were much louder. The slamming continued, but I could hear screaming, a kind of guttural roar, a voice breaking as it was pushed to its limits. The slams from above got closer and closer together. More screams, but this time a second voice. It couldn’t be Olivia; she was out seeing a client. Then a thud which shook the house. And then silence.

My heart raced. I didn’t want to go upstairs; interrupting because the sounds were scary seemed incredibly feeble. I am the kind of person who encourages people to authentically express their emotions.

Eventually my curiosity got the better of me. I crept up to the top of the house. The loft door was ajar, the room dark. “Noah?” No reply. I waited a minute before carefully pushing the door open.

The room was destroyed – holes punched in the walls, bloodstains on the carpet, the wooden bedframe splintered. The narcoleptic was lying face down on the ground, beaten to a pulp. He was naked, emaciated, ligature marks around his wrists. How long had he been in here?

Noah was standing motionless, shoulders slumped. His eyes glinted in the light from the laptop screen. On it, a Zoom call was open, Joe Hudson’s beaming smile on screen. I could see at least a dozen other squares, each with someone watching intently at the destruction. Noah turned to me, eyes glazed, saying nothing.

“What the fuck is going on here?” I shouted. Noah breathed raggedly but said nothing. From the screen, Joe said, toothily: do you not like people authentically expressing their emotions?

ART OF ACCOMPLISHMENT – INTERNAL ONLY
2 Apr 2026

Team,

Q1 performance has exceeded expectations:

• 78% year-on-year increase in unique web visitors

• 3,200 Connection Course paid memberships ($200 ea.)

• 150 Master Class tier members onboarded ($10,000 ea.)

As we move into Q2, capitalising on growth remains essential. Our Connector Course members represent our highest-leverage opportunity for authentic emotional release.

Master Class members are increasingly asking for curated viewing experiences calibrated to their new nervous system thresholds. The new invite-only Vision Tier, launched last quarter, has proved enormously successful, with over 100 livestreams run to date. Reception has been extremely positive.

The enhanced Connection Course protocol has approximately doubled the conversion rate from member to integration participant, providing more capacity for curated Vision Tier experiences. The Larkhill intervention exceeded engagement metrics, setting a new benchmark for our member experiences.

Remember: together, we’re rewiring how people relate to themselves, to others, and to the inner journey.

Onwards,

Joe

When I founded Mast, I had a few motivations: I’d never started a startup and wanted to try my hand at it; it sounded more interesting that being a regular software engineer which I’d done for many years; I would get to build stuff that adds value, that people would use.

But the primary motivation was to make a bunch of money and donate it to effective charities – much more money than I could make in a job. It would be dishonest to say that making money for myself wasn’t also a motivation, and starting a startup is one (admittedly unreliable) way of doing that.

I want to make a difference to animal welfare, and my own skillset isn’t relevant (though I have tried to shoehorn it in). As with Earning to Give, Founding to Give almost always involves decoupling your job and the impact you have, although a lot of FtG discourse tries to claim otherwise.

Being a startup founder, if successful, could net me millions. It’s an outstanding opportunity to do good. And if it failed, it would hopefully fail early, and I could just return to my software engineer job.

That was five years ago. I took the Founders Pledge, a commitment to donate some portion of your proceeds to charity when you exit your startup. Since then, a whole ecosystem has popped up around ‘founding to give’ (as opposed to ‘earning to give’, where you take a well-paid job to give some of your salary to charity).

Optimise for money and edge, not impact

It’s unlikely that you’ll be able to found a company that’s both very valuable and very impactful. It would be a suspicious convergence if you were able to do this and not have to trade off one against the other. Almost certainly the most profit-maximising and impact-maximising ideas are going to be different.

You also want to find a startup idea that you’re actually good at executing. I’m a big believer in doing what you know – which is not to say that people can’t learn new skills (and surely being a founder forces you to do so) but founders need to have an ‘edge’, some set of skills which makes them unusually good at something. Picking an idea you think sounds impactful usually involves ignoring this. Founder–market fit is a real thing.

A common failure mode is to venture far outside of your range of expertise in search of an idea. You can convince yourself that you’ve found something that will turn an enormous profit and also benefit the world. But often it requires very large amounts of capital and skills you probably don’t have. I saw this a lot in the latest AIM Founding to Give cohort: they often focused on impact-focused ideas that are crowded and not especially tractable.1

This is especially true if your cause area and your edge are just very different – it would be difficult for me to shoehorn my particular set of skills into direct work in animal welfare and have anything like the EV of founding a for-profit and donating.

(There are some notable exceptions, like Sendwave – the canonical example of a high-impact for-profit, a startup which offered an alternative to Western Union’s enormous remittance fees for users in various African countries.)

Accept that FtG is a massive bet

Although startups fail at a shocking rate, I predicted the expected value was higher than working as a software engineer – especially in the UK, where salaries are good but nothing compared to what a skilled SWE can earn in major US cities. Think $150,000 at the very top end in the UK versus up to $400,000 in the US, rising to $500,000 or more at frontier AI labs (though getting a job there is very competitive).

Ambitious Impact (AIM) now runs a Founding to Give program which brings together EA-minded founders to find problems to work on, team up, and form companies. Unlike other incubators they don’t offer funding (I went through Antler, which offers a small amount of initial funding), but they also don’t take any equity. It’s too early to tell how successful the incubated startups will be, but I’m optimistic, although I think many of them don’t aim high enough in terms of potential market size.

Hugo Walrand has written sceptically about whether Founding to Give is as promising as it looks. His core arguments: startups take a long time to exit (he uses 7.5 years as a typical figure), which makes donations heavily discounted in a short-AI-timelines world; founders may renege on their pledges over that long horizon; earning-to-give donors get short feedback loops on their giving while founders don't; recurring annual donations get more favourable tax treatment and often attract employer matching.

An AIM analysis of Y Combinator startup exits suggests:

• the median time to exit is 7.5 years

• the median exit is worth $7.8M to the founders (so $3.9M each for a two-person founding team)

• the 90th percentile exit is ~$30M

I think the median numbers are doing too much work here. The impact is in the tails: for a 90th percentile exit, if the founders hold even 25% of that, they can donate $7.5M. No SWE (or two or three SWEs depending on the size of the founding team) can donate anything close to that.

But ultimately these numbers are not enormous. You probably need to be aiming for a very large exit, and so you should pick a startup idea accordingly. For startups that do extraordinarily well, the donation opportunity is enormous. The cofounders of Canva have 30% equity in the company and have pledged the majority of it to charity, which works out at at least $6bn at its current valuation.

You don’t have to do it for 7–10 years

Startups are a slog. An exit is uncertain, and usually takes many years. The NPV of your exit money needs to be priced in when comparing it to EtG.

Despite this, secondaries allow you to extract value before an exit. When you raise capital, especially in later-stage rounds from institutional investors, the founders are often able to sell some of their equity to investors directly and take the cash directly.

Investors like this because it gets founders bought in for the long haul; although they expect you to take some of the risk of founding, they also don’t want you to be on a ramen-level salary forever and just give up. This gives you a liquidity event and a chance to donate earlier than your exit.

This also reduces the risk of value drift. If you get some liquidity without having to wait years to exit, you presumably have less time to talk yourself into reneging on your donation pledge (I don’t know the base rate for this happening but I imagine it’s very low).2

One other thing to note: it usually doesn’t take 7.5 years to discover if your startup is going to work. Most startups fail well before that point; not dying is the most important part. The longer your startup has existed the higher the chance of an exit. If it blows up in year 2 because you don’t have enough customers, the opportunity cost is much smaller.

Pick a lane: you can’t FtG and EtG simultaneously

Walrand suggests you can derisk by founding a startup and then paying yourself a big salary – founding and earning at the same time. This is almost certainly not possible.

The expectation is that as a founder you will be paying yourself less than what you can get in industry because they’re being compensated by the large equity stake they’ve taken. The whole point is that you’re taking on some risk in exchange for upside.

In the early stages, you won’t have enough money to do this anyway. And later on, investors are simply not going to allow you to take a huge salary. They don’t want their money being doled out on your fat salary! You’re expected to take on some risk – and doing so gives them confidence that you’re bought in to your startup succeeding.

Investors can enforce this by putting limits on what you can pay yourself as a condition of their investment, especially if they’re an institutional investor like a VC (angels typically don’t have this level of control), or by putting in place a requirement that you get approval from them to pay yourself more than a certain amount.

I think Founding to Give is very defensible in a particular form: as a way to make money to donate, not having direct impact. Finding a startup idea that does both requires a suspicious convergence of which I am not convinced (except in very rare cases). For most causes, especially animal welfare, there aren’t many for-profit options that move the needle, especially for tech startup founders.

A big consideration is when you think the money is going to be needed. If your timelines on transformative AI are short, it might be that the money is simply not going to arrive fast enough to make a difference.

Hugo undersells the tail. Most founders won’t become billionaires or even multi-millionaires, but enough will to make the expected value work, provided you’re at least somewhat honest about whether you’re in that distribution. And even if you don’t end up at the 99th percentile of exits, founders sometimes end up with millions and gain a huge amount of experience that they can parlay into their next role. Go forth and found!

People routinely underestimate the rate of AI progress. AI now has superhuman skills in coding, design, writing, research, diagnostics, realistic image generation, and much more. Many of these abilities were confidently proclaimed to be impossible just months before AI gained these capabilities.

Still people think “it can do that job, but it can’t do my job.” In some cases this is pure delusion: yes it can do your job at the slide farm, you are not special.

(In many cases this opinion arises because people are rarely using frontier models. A friend working at a major multinational insurer complained that her AI-generated meeting minutes were full of hallucinations. Turns out they were using GPT 4.1 nano, a model that is small and months old.)

A more reasonable objection is that AI lacks taste. Yes, it can pull together PhD-level research reports and write functional code and write emails. But if you’re a tastemaker, you have an edge on an LLM which is post-trained to produce text which is somewhat milquetoast, aimed to appease the median user of a helpful, harmless and honest AI.

Craft taste and editorial taste

Craft taste refers to judgements about form once the target and context are fixed. Typography, prose rhythm, tight design, code that works. Image models couldn’t reliably produce hands with the right number of fingers a couple of years ago. This is a load-bearing part of many people’s confident claims that AI will never replace humans in specific domains – specifically the combination of aesthetic taste and competent execution – and is now largely dead.

Editorial taste refers to judgements about content. In a blog post or a report, it might be what gets kept or emphasised versus what ends up on the cutting room floor. How to design a website so it appeals to your target audience and speaks their language, or marketing copy which emphasises the right set of features in the right register.

AI can sort of do this. It can exercise good judgement about interesting writing and copy. People lean on it heavily as a sparring partner when thinking through a tricky problem. Sometimes people even have a revealed preference for LLM slop – notice how every post on LinkedIn now apes ChatGPT’s particular style as though it’s a new brand of corporate-speak. Most professional writing is taste-neutral and LLMs are perfectly good at replicating it.

People whose roles lean heavily on editorial taste are harder to ape. These roles require having some insight into their domain in a way that lets them make picks, sometimes even unpopular ones. They’re using knowledge which is hard to capture cleanly and so hard to replicate, to find things which they think will appeal to their particular ingroup. At the extreme end they can pick unpopular things that are on the edge of what’s appealing and bring them into the zeitgeist.

Fashion is a good example of this. Fashion houses make design choices which are outré and even derided by most people. But these choices, in time, propagate into the common memeplex and start to take over clothing designs for regular people.

Scott Alexander suggests that fashion choices are about attempting to look like a member of a higher class, and avoiding being seen as a member of a lower class. The upper class starts to wear something new. In time this propagate down to the middle class, who want to look like the upper classes, so copy their outfits. The lower classes do this in turn, and the upper class, not wishing to be seen as lower class, change their fashion again and the cycle repeats.

Taste as zeitgeist

Having “good taste” often means being plugged into the zeitgeist of what groups think specific things are fashionable and then ensuring you don’t seem gauche by association with them. This happens in domains which aren’t drawn across class lines: having “good taste” in music is often about having eclectic and non-mainstream taste, signalling you’re not into the shit that’s on the charts.

This is also the case in e.g. visual design. New and interesting designs start to be used on websites (parallax scroll, how cool!), they get reproduced everywhere (god, another website using parallax scroll) and so become passé, designers start to create new and unusual (but still mostly aesthetically pleasing) designs to distinguish themselves, and so on.

Being plugged into the zeitgeist is something that humans are extremely good at. AI could also be good at this: as each model is trained and finetuned, it starts to get a sense of what’s stylish and what isn’t, and with some shrewd prompting you can get designs which are both aesthetically pleasing and editorially tasteful.

As long as LLMs are trained to be maximally inoffensive they won’t ever be able to compete here, as what’s tasteful and in fashion is often deliberately out of distribution of what the average person likes. The post-training that models are subject to makes them too milquetoast for them to do this; these picks are anti-mass appeal.

But it’s not clear to me that AI couldn’t do what tastemakers are doing. They’re trained on a gigantic corpus which includes essentially all of the tastemaker content that’s ever been created. They’ve read all of Vogue, the Pitchfork archives, weird Substacks, academic crits, Sontag. ‘Tastemaker output’ is well within the capabilities of these models but is locked behind the RLHF layer that pushes them to be mid.

Tastemakers survive, for now

The consultant who needs to write decent marketing copy or produce a slide deck in keeping with her company’s brand guidelines is fully replaceable. Those jobs are about craft and good aesthetics, and the taste they were aiming for is exactly the kind of thing that LLMs are optimised for: tasteful designs, pleasant copy, and superhuman execution.

If your role involves you being a tastemaker, AI hasn’t yet caught up to you. Think: the fashion designer who needs to be at the cutting edge of trends, A-list creative directors, editorial directors. LLMs’ inability to ape this kind of curation is protecting these jobs.

But the number of such jobs is much smaller than most people think. Most who raise the objection of “taste” are referring to the kind of taste which is now fully replicable by AI.

Plenty of mass media doesn’t even require a great deal of cutting-edge taste – see every reality TV show of the last few years which are essentially the same set of conceits (put a bunch of hot people in a confined environment and get them to do stuff) with slight tweaks (on a boat, on an island, with their exes, while singing/dancing/baking).

And where taste really is still required, the surface area of what’s truly unique will shrink – the role of tastemakers will be to prompt LLMs in the right direction. Doing this still requires people with taste to guide the model in the right direction – you can’t prompt your way to taste you don’t have (for now). Tastemakers provide a signal that models can’t predict.

But as tastemakers are increasingly able to get tasteful output from LLMs, we’ll just need fewer of them. Readers’ attention is finite and the world only needs so many hot takes on the latest A24 release. If one creative director with Claude can do the work of a creative director plus three juniors, the juniors are the ones who are cooked.

Alcohol and tobacco are poisons of antiquity we still willingly drug ourselves with. They’ve been around forever and may well be with us forever, such are their far-reaching cultural tentacles.

Of course there are plenty of poisons we willingly ingest – other drugs of all flavours, natural and synthetic, old and new. But we’ve also invented a uniquely modern kind of poison: digital ones.

I opened Instagram after being away from it for a few weeks. It’s a pattern that will be familiar to many: the app sucks you in; thanks to very shrewd UX choices a single tap on a friend’s video opens up reels, and then suddenly four hours has passed and your arm has gone numb from lying on it for so long. You feel a sense of gross residue from spending so long staring, idly consuming fifteen-second slop injections.

This shit is poison, you think. You’re disgusted with yourself for getting sucked back in again. You delete the app.

I reinstalled it to share one of my posts. The hit to my mood – not severe, but noticeable – happened almost instantly as my brain processed the various elements shown to me.

At the top: a series of profile pictures of people I don’t really know. Two of them are girls I met while travelling who didn’t want to sleep with me (actually one had terrible diarrhoea so that stung less). Scrolling sideways I find a London friend, not someone I know super well, planting some seedlings with friends. Though perfectly harmless, it jolts me out of my Californian reverie and makes me realise that life is moving on without me back in London.

The first post in my feed: my brother, grinning after having finished the London Marathon (two years ago; he’s reposting it as this year’s is coming up this weekend). It has half as many likes as my entire follower count. A bunch of comments, one of which talks about his cute smile. I don’t much like my smile.

The friendship paradox means that, for most people, your friends will have more friends than you do. Nowhere is this more obvious, rubbed in your face quite as much, as social media.

I want to delete my Instagram account but can’t quite bring myself to.

Maybe this is because I don’t want to eliminate myself from some people’s memories. Facebook is dead, none of my friends really use Twitter apart from me, so Instagram is the remaining thread between me and a few dozen people. I (mostly) like seeing their stories, seeing what they’re up to, things they wouldn’t think to share with me directly.

By leaving you bear a personal cost: you no longer get the ambient exposure to your friends, nor they you. You’re noping out in a way which leaves you invisible to these people. It’s a collective action problem – it’s impossible to get everyone to leave for a better platform at the same time. And what platform would you even bring people to?

It makes sense then that young people would pay money to have others not use the platform at all.

The other reason these things are poison: everything’s reels.

Reels are the final form of online media.

Everything ends up going the same way. Instagram became reel slop. YouTube became Shorts, which is isomorphic to reel slop. Snapchat has streaks which keep tweens hooked on sending selfies to each other at least once a day for 800 days straight. I’m surprised there aren’t reels on YouPorn at this point.

At a Google conference they boasted about serving 200 billion hours of short-form video last year. We should shun these people the way we shun Philip Morris executives.

TikTok is the final form of the social media slot machine. They didn’t add shorts backhandedly as a feature which is embedded in another app, or which you slide into effortlessly with a rogue tap. Shorts are the whole thing. That app is like having a gun in your house. A sign of things to come; industrial-grade digital poison on tap.

With poisons of antiquity we built up cultural antibodies, legislating and educating people on their harms. Only now are we starting to catch up with the damaging effects of social media, with more schools banning phones and more countries enacting social media bans.

I’m conflicted as to how beneficial bans on social media for young people are. The Online Safety Act is a great example of how this can go wrong – where adults are now required to submit to punitive age-gating to view e.g. a subreddit about cider. But I can certainly see a case for regulating apps which 1) have a clear danger of social contagion and 2) use weaponised algorithms to be maximally addictive. I think they have the potential to be uniquely harmful in a way that e.g. porn is not.

There’s something very sad about the way social media has gone. Early Facebook felt like a warm place: you shared photos with your friends, you posted on each other’s Walls, you planned parties and could look people up you met at events to flirt with them online because you were too shy to do it in person.

The writing was on the wall when people started to friend everyone on Facebook they ever met – the definition of friend changed to anyone you had the merest hint of a connection with. This has a cooling effect; suddenly you don’t want to share anything personal because you have a network of a thousand people who you barely know (including your parents; maybe that was the real death knell).

Some social media platforms tried to address this. The one that stands out to me is Google+, which let you organise your friends into ‘circles’ of decreasing connectedness, with the people you barely knew in the outer ring, letting you pick an audience with which to share depending on how intimate the content was.

Path is another. It was launched in 2010, co-founded by Shawn Fanning (the Napster guy) and ex-Facebook product manager Dave Morin, which capped your friend count to 150, aping Dunbar’s number. Unfortunately before it died they lifted this limit, the pressure to grow being greater than the pressure to stay small.

I think it’s an unsolved problem. I long for the old days where we could share things to an intimate, private audience, without your feed being injected with influencers on apps designed to maximise addictiveness on which you are “connected” to more people than any person can be friends with. Something like a more humane online garden.

Forecasting has become enormously better over the last fifteen years. Tournaments, superforecasters, and prediction markets, and now LLMs closing in on the best forecasters. Metaculus, Manifold, Polymarket, and Kalshi have billions of dollars moving through them.

But the point of a prediction is to change a decision. Forecasters predicted COVID and yet the actual response to it was slow and fragmented. Making good predictions is legible and measurable, but how those predictions impact the world is not. We optimised the thing that was measurable but gave no thought to how those predictions were used in the real world.

Goodharting on predictions

The only possible purpose of predictions could be for them to have some impact on the real world, otherwise they’re just navel gazing. But making good predictions is much more legible than coupling them to real-world impact. What gets measured gets improved, so people just optimised for on prediction accuracy without paying attention to how those predictions get used.

Actual impact is much harder to quantify, and the delay from a prediction being made to that prediction being operationalised is long and hard to measure. No-one gets a career boost from focusing on this, so few people do. As a result the impact of accurate predictions on the world is hard to detect.

Outcomes are a second thought

IARPA’s ACE program was intended to “dramatically enhance the accuracy, precision, and timeliness of intelligence forecasts”, which it surely did. The Good Judgement team did so much better than IARPA’s prediction goal that they stopped funding the other teams after two years such was their accuracy. But when asked if their work influences policy decisions, Mellers said “we certainly hope it does.”

Everything inherited the frame of this tournament. Over the following years, forecasting and prediction markets entered the common consciousness and a half dozen prediction markets appeared, many of them trading real money. But ultimately people were trading either on status or on making money off their predictions.

Ahead of the curve on COVID

A Metaculus user flagged the Wuhan coronavirus as a plausible starting point for a pandemic in January 2020, and by the end of that month the platform’s predictions were pricing in over 100,000 infections. Good Judgement similarly was predicting 100,000–200,000 cases by March 20.

This was a hit. The prediction markets were genuinely ahead. Some individuals did update, especially in the EA and rationalist spheres – some individuals started stocking up and preparing themselves and their families for what was coming.

At a large scale, however, nothing happened. No-one who mattered actually updated on any of these predictions. Lots of things could have happened: public health decision-making pipelines could have included a step which included forecasts, government bodies could have been clued in to what decision markets were saying. There were some prediction market-pilled people in key roles, like Dominic Cummings who pushed an evidence-based approach to the pandemic response in the UK government.

A notable exception to this tells a story: Shannon Gifford was both a superforecaster and a projects officer in the Denver mayor’s office. She pushed her colleagues in January to listen to these predictions and succeeded. She was essentially someone who could bridge these two worlds and actually get something done with the predictions being made.

Essentially the bottleneck wasn’t accuracy; solid predictions existed and turned out to be directionally correct. But there was no mechanism for coupling those predictions to a response among the people who mattered.

Coupling forecasts to impact

In some sense forecasting is a solved problem, and doesn’t need more resources directed to it – predictions will inevitably continue to improve as AI starts to consistently beat even the best forecasters. But cheap LLM forecasting makes the prediction–action gap even worse; there will be more and better predictions but without a working pipeline to translate those predictions into outcomes they will have no effect.

Instead of status being a function of prediction accuracy, forecasters should also be judged on their real-world impact. Informing the conversation or being widely cited are not sufficient – forecasters ideally should be able to point to specific outcome or shift in government policy (or even just the meaningful inclusion of their predictions into a decision-making process with teeth).

So instead of focusing on increasing the accuracy of predictions the focus needs to move to making sure those predictions get operationalised. Decisionmakers need to be persuaded to precommit to taking action in response to particular forecasts, e.g. if a pandemic-related prediction market moves past some threshold, the body in question is committed to acting on it. The CDC could have policy ‘triggers’ which are tied to Metaculus prediction markets or confident Good Judgement predictions.

We should also try to get forecasters embedded in decision-making orgs to bridge the gap between prediction and action – we need more Shannon Giffords. Chief Forecasting Officer should be a prominent job title in government agencies, a specific point person who is across the prediction markets and superforecaster predictions to ensure they’re included in that agency’s decisionmaking process. Cummings attempted to do this in the UK government and failed, so this needs to happen at the agency level and within the civil service instead of the upper echelons of government.

To reduce Goodharting, we can move the stated outcome of prediction markets to bake in the real-world effect of those predictions, pricing in the implementation of the prediction. A funder could pilot funding-specific prediction markets looking at e.g. whether if funding is granted to an org, there will be a specific outcome in the real world.

Decision audits could be used to build accountability on the side of the decisionmakers – public retrospectives asking “did a relevant, high-quality forecast exist and was it acted on?” This creates the missing career incentive for government bodies to incorporate forecasts into their decisionmaking process.

For the very longest tail risks, catastrophe-bond style instruments (like the bonds insurers issue against hurricane risk) could incorporate prediction market data and pay out if the specific prediction pans out. This essentially joins the prediction to an actual consequence to the underwriter.

Without real-world outcomes, forecasting is just a hobby. As a field it needs to aim higher: influencing better decisions among those who matter and have the biggest impact. We need more Shannon Giffords, not more Tetlocks.

A few weeks later, after a frankly ridiculous amount of medical bullshit, playing phone tag, chasing people around, being kept on hold, I was in the rTMS room. A helmet on the end of an arm was put onto my head and the chin strap cinched down. “If you’re ready, we’ll start the stimulation.”

When the session starts, ping – you get delivered 20 short pulses over two seconds, followed by a 20 second break. Ping – another ‘train’ of pulses. You have 55 trains in total, so the session lasts about 20 minutes.

The first time you go in is the calibration session. The technician puts on a cap with a grid on it. She puts the helmet on and uses a foot pedal to apply a single pulse, and watches your ‘motor response’ – basically how strongly your face and hand twitch (on your right side, due to the brain’s cross-wiring). When she’s found the place that gives the largest twitch, that’s marked on the cap and that’s where it’ll be placed in future sessions.

The downside – apart from the enormous cost, thousands of pounds for one course – is that it takes time. The sessions are 20 minutes but you have to go in five days a week for twenty or thirty treatment days (I did twenty).

But as someone who’s taking a career break and is somewhat aimless (how many loss-making side projects can one vibe code really?) it was nice to have the structure. Out of bed, tube to Marylebone, 30 minutes of magnets, and then the rest of the day to write, plan, see friends, see London on a crisp February day.

In rTMS (repetitive transcranial magnetic stimulation), focused magnetic pulses are repeatedly applied to the front-left part of your brain – the dorsolateral prefrontal cortex. This part of the brain is typically under-active in people with depression. The magnetic field induces an electric current in your brain, causing the neurons in a specific area to fire.

Most people haven’t heard of it and are surprised to hear it’s been around for decades – the first TMS device was developed in 1985; the FDA approved the first device for marketing in 2009.

It’s also remarkably effective.

With antidepressants – which I remain enthusiastic about in spite of the prevailing anti-SSRI zeitgeist – the numbers look okay at first (28% remission on your first SSRI). But then the response falls off a cliff – 20% on the second drug, 14% on the third, 10% on the fourth. Side effects which are not trifling, and withdrawal which can mean it takes months to wean off. 40% of people quit their SSRI within three months.

For standard rTMS, 47% of patients show a response, and 22% go into remission. With deep TMS, which uses different magnets to penetrate further into the brain (and which I had, hence the silly helmet), the response rate was 74% and remission 58%.

Medication is a vital part of our psychiatric armamentarium, but sometimes you have to bring out the big guns.

After a week or two, slowly and inconsistently, things started to shift. Things around me were more interesting. I would stop to look at the trees, or the contrails of a plane in the sky. The frantic anxiety waned. I was able to gradually reduce the heroic dose of medication I was on.

Now I’m doing a month-long writing retreat (which I feared would be an emotional nightmare; I checked the price of one-way tickets home before I left), and moving to Berlin for the summer. Things are looking up.

10/10 would recommend.