I often claim to be a “Utilitarian to the grave”. This whole ordeal started at around age 16, where I was a tad bit too autistic to not ask: “what makes a thing good to do and a thing bad to do?”. All my surroundings seemed to have pretty horrible answers.1

So, as zoomers do, I turned to the internet. I started reading Peter Singer and Benthams Bulldog. I discovered hedonistic utilitarianism, the idea that a thing is “better” if it leads to more happiness in the world. This seemed obviously correct, and was the only thing that survived my not-so-stringent adolescent scrutiny, in a way that virtue ethics or deontologies did not. So I adopted it verbatim.

And I really did (and do) use utilitarian thinking when making decisions!2: I became vegan, I vowed to donate my money when I would get older, I decided to try to get a career in AI safety etc… And I did all these things, because I really did believe that I terminally valued the axioms of hedonistic utilitarianism

Often, people would retort by giving an example to try to get me to change my beliefs:pascals mugging, wireheading, simulation arguments… This didn’t make sense to me. I was a utilitarian, after all. I believed in utilitarianism because happiness is all that matters. If I am able to change my beliefs about ethics from intuition, then I might as well have no moral system at all!3

I now believe that this is wrong, let me tell you why by virtue of vaguely related parable.

We all know a “sleep score” guy. You ask him how he slept and he’ll look at his watch and respond “84”. (I’m sleep score guy). Last night, I slept kind of bad. I woke up tired, with dry eyes and a general intellectual malaise I associate with a poor nights sleep. Yet, sure enough, on my wrist, there it was: 84! An “excellent night’s rest”

Surely everyone would agree that going “huh! I guess I slept pretty good after all” would be silly. The point of the sleep score is to track/predict how well you slept, not the other way around! If a number on your wrist is able to single-handedly sway your opinion about what constitutes a “good nights sleep”, then what does a “good nights sleep” even mean? Is the point of sleep not what it does?

But note, it would be incorrect to say the sleep score tells you nothing. Maybe you just drank too much last night, and your eyes are dry for other reasons, or there’s some sort of medication messing with you, and you actually really did sleep good. If the sleep score and your sleep disagree, it does say something!4

The same I now hold true of my relation of my intuition to utilitarianism.

Brutally for my autistic mind whose intuitions align with hedonistic utilitarianism in 99% of cases, it is totally the case that if I look inward, I am very convinced by some arguments against hedonistic-utilitarianism5. I wish to not be robbed, and I wish for the forest to be real. Sadly, I cannot measure my sleep by checking its score, I cannot thoroughly check my beliefs by summing happiness.

This realization, along the robbery of the simplicity that came along with my earlier belief system6 have gotten me thinking: why do I do what I do? (In my case, interpretability research in San Francisco)

Is this because I believe this is best for the world? Because I believe in the non-absurd pascals mugging of AI risk? Is it because I’m a singularitarian? Is it because the bay area intellectual climate brings me a sense of being understood and community? Is it because I’m just having a lot of fucking fun with the actual work?

A friend from my side of the memetic pond has likened their adherence to utilitarian principles to linear structure in the residual stream. If you assume it as truth, you can explain so-so much about the real thing, but try as you might to reconstruct the whole thing, a significant amount of dark matter remains unexplained.

My actions have been for much, much squishier, fuzzier, human-values reasons than I had originally suspected. And I fear it plausibly can never be fully defined. Part of this is my “utils” being not mere hedons and part of this is that I’m not an optimizer, I am, in fact, a monkey.

This is okay. I also happen to believe the work I’m doing is actually doing “good” in the world, both in the fuzzy and non-fuzzy sense. All the better that I have a blast while doing it, and am surrounded by a community that I like and they like me back for it.

Writing this, listening to Garden Of Hopes and Dreams by Toby Fox, I ask myself “where the fuck did this guy get so good at music?”. And I realize that there were 5 chapters of music before this, and a whole game before that, and plenty of random songs for SNES mods before that, because oh yeah this guy just fucking loves making music.

You can’t compete with someone having fun, the only way out is to join them.7

Of course it was like a movie scene, what part of my life hasn’t been the past half year? I know you, Mom. You never ring twice. And when you do, I pick up immediately

“grandpa is dead”

Those were the only three words you got in before my phone died. Five thousand miles away, I sit stumped in my cubicle. I sob like a child when I feared I wouldn’t.

And I immediately know all those sorry aphorisms were true: I should’ve just come to visit, even when I had no reason. I should have spent more time with you. I should’ve said goodbye. I should have meant it. I had no idea it would be the last time.

My friends ask what you were like, and in my talking I realized I was just talking about dad instead, how he would cope with you being gone. In a lot of ways, you died seven months ago.

It was supposed to be a routine scan, they had to take you off your blood thinners, and a clot choked your brain of oxygen for a minute too long. They were fast enough to save your life, yet too slow for you to ever be the same again. The knife of modern medicine cuts both ways

You too, refused to settle for only doing one thing at a time: a bartender, a football coach, a landlord, a dove racer, a teacher, a friend. They even wanted you to be a politician at one point, but they didn’t put you at the top of the list, so that was that for politics.

A man whose vulnerability was only apparent in the obvious attachment to his pride, now in diapers, paralyzed from the waist down. You weren’t a guy to sit still for seven months. But my dad cared for you, he made those moments livable. He cared for you a terrifying amount. I thought it may kill him first sometimes.

And one July night, it just happened: Slowly, and then all at once. You feel a bit nauseous, and the next day you feel even worse, and ask to be brought to the hospital. You’re not a guy that asks to be brought to the hospital. My dad must have been fighting his complacency on the drive there, after all, this is not the first time for him.

And when you arrive you’re revealed to be just a pile of accumulated DNA errors like the rest of us. Your heartbeat at 130, fifty-five breaths per minute. I wonder what you thought then, did you ever realize that this is the end? Your organs gave out three hours later. The doctors said it didn’t hurt, but when have they ever not said that.

“That one is special” you said, until the very end. I never knew why, you never said this about my brother or sister. You believed in me before I did. I’m glad you got to go out feeling vindicated.

My mom told me this was the age she lost her dad. Lung cancer. I heard you threw that cigarette pack against the wall and never opened it again when it happened to your dad. I always thought that was so fucking badass. And I knew then that you were like me and I was like you. I knew that I had inherited not only the impulsivity, but the stubbornness as well.

But despite all my carefully placed adjectives and my perverted earnestness, I look for the ordinary kind of feelings, and find only the very kind of regret my dad warned me about: every lunch break I took for granted. “I didn’t say it enough”, “I didn’t even come visit.” That time is spent now.

How many League Of Legends games I must have played with you mere meters away from me. How I wish I could have that time back now. How I wish I could give the people dear to me the time I still have. But some things never change, we all know I won’t do that.

Oh how I would love to have those hours back. How I would love to sit here and say: “I have learnt my lesson, I will cherish the little time I have left with those I hold dear”. But I’m not going to stand here and lie to you: I suck at that. I’ve always sucked at that. And 8000 kilometers only makes it harder. All I can promise is that I’ll continue to try.

But the ruler of everything moves on, punctual as always. It cares not for the bad news stacked on top of each other, nor who it leaves behind, nor the ruthlessness of vast distance. All mere planks in the Ship of Theseus I call my life. 1

Nothing in my world feels secure, and everything changes so fast. In that rapid change, I at least find comfort in retreating to you.

In my memories you make me the finest fries from fresh potatoes
In my nightmares of missed busses you drive me to my exams
In the background of old pictures of old football games, you’ll always be there.
My grandpa who still could.

Polite fiction

Japanese society is FULL of polite fictions.1 This really irks me! Some examples:

• Setting up any gambling room/casino for real money is illegal, very much so. But you are allowed to operate little gambling rooms, called pachinko where people buy little balls from you, gamble with the little balls, to earn more balls. You are then allowed to open a second establishment, where you happen to buy these balls for real money

• Sex workers are allowed to be paid for basically any sexual act, except penetration. Yet the sex worker is allowed to consent to penetration after compensation. In practice, penetration is almost always had, and it’s definitely not because the woman was really into the guy.

• This one is particularly insane: women were so embarrassed about “bad sounds” from the toilet (this literally means the sound of pissing) that they started permanently flushing the toilet to cover up the sound. This got so bad that many toilets in japan now have a Sound Princess. A button you can press that continously makes the noise of flushing, to save on water!

It’s like society is fine permitting these things, but it can’t bear to not look away. There must be a little scene, a little performative show. Everything is farting at the table.

Once you notice this, you will see it everywhere in Japanese society, and it’s very much not for me.

Every interaction in society just seems to have so many frills too: an Uber is 25 bucks for someone with leather gloves and a fancy self-opening door, perfectly quiet and clean.

Contrast this against the South Korean national ethos of ppalli-ppalli (quick-quick), where everything just has to happen fast. I strongly prefer my 5 dollar Seoul Uber where the driver is shouting at me to close the door again in korean, and probably insulting my mother.

I could never live in japan.

Femininity

I found it particularly interesting how displays of, what would in the west be so extreme as to border on caricatures of femininity, were fully embraced by society, and very popular.

You know, 90% of what I saw worn in this context is shit I would look horrible in, and that, when I reflect deeper, really don’t want to wear at all, but I thought it was interesting that this is stuff that is actually worn in the first place.2

I saw a relatively wide variety of presentation too, and was happy to find out that this choice of presentation is an explicit fashion statement rather than a default.

It’s plausible that confidence that you won’t get judged for dressing in a hyperfeminine manner.3 has been largely lost due to civil rights pushes for feminism, but this is probably worth it on net.

I just think it’s interesting that I had totally written this off as an anime trope, but it didn’t seem to far from reality.4

Celestopia

This is all making me wonder, if I can form my social contract from scratch, what would my ideal society actually look like.

What is celestopia?

I can try to pick out cultural attributes of cultures I love, and fuzzily define it that way

• First of all, if I’m not living in a big city, it just feels like I’m actively dying.

• I love how honest, raw and to the point dutch people are.

• I dislike how tit-for-tat stingy dutch people are, I generally like abundance-coded society where people gift each other things expecting nothing in return

• I really fucking like how in Korea everything needs to happen immediately, they even have a name for it

• I like when vegan food is good and readily available, Berlin and Ghent have been best for this5

• I really like how in Berlin, everyone just presents as aggressively themselves, and it’s the one place where “clothes aren’t gendered” has not been a mere woke-ity.6

• I like the ambitious atmosphere, proximity to tech, and liberal drug policy7 of SF

Celestopia doesn’t exist, though SF is probably closest.

An old favorite

I rewatched kimi no na wa (your name) on my way back from Japan, and noticed so much more that I had previously missed.

This movie holds a pretty special place in my heart. I watched it as a young egg and felt uniquely understood by the perspective of a guy waking up in a womans body.8

It feels like a distinctive Japanese story, the sort that is hard to execute well from a western angle.

But it’s a story that really resonates with me emotionally. One of eschatology, finding yourself and and a synthesis of love and admiration towards the feminine, with a red ribbon through the knots of time tying yourself towards that destined other.

And as I write this from my airplane seat, an hour to go to SFO9, I can feel a similar sort of pull, that it doesn’t really matter what I do, the timelines will arrange themselves such that this is the place I will end up calling home.

Where I shall build my heaven on earth10

But my favorite part of korea was the sixteen hour league of legends binge1

The pc bang

I have been playing League Of Legends for fourteen years, for a total of ~10 thousand hours. This past ~year has been one of the only parts in my life where I have played approximately zero league. But when I arrived in Korea, I knew this had to change.

As a European league player, you often hear whispers about the korean server, and korean league culture.

Playing league as a kid, at home is a constant struggle against Reality: shouting at 2 AM will wake mom, the book on your desk reminds you that you have a test tomorrow. The context clues of home and proximity of parents make you unable to detach from reality.

The PC bang (literally translated: “PC rooms”) is the beautiful perverted inversion of this: massively popular korean internet cafes focused almost exclusively on competitive gaming: they’re cheap, open 24/7, you order slop food from your monitor, you get the best monitor, the best chair, the cutest keyboard. There are no windows, the light is always on, everyone around you is doing the same, just as degenerate thing as you. The context clues encourage you only to keep playing, indefinitely, like a casino.

This has obvious negative effects on society, but is also hedonistic valhalla. The pc bang is the closest thing to my personal version of the pleasure cube

Why is South Korea like this?

The years is 1998: the average korean man is kinda poor, lives in a shoebox with thin walls, and has no internet connection. He wants but one thing: to play Starcraft.2

This, combined with the peculiar state of korean broadband rollout, Seouls insane density, no consumer money for personal computers, but massive IMF loans allowing people to start their own company, transformed the PC bangs from a room with internet access and a printer to a proper third places for gamers. The massive popularity of MMORPGs like WoW, and afterwards, league, intensified this, as both massively benefit from being in a shared space online.3

It’s important to note here that this is not just an artifact of “Asian country weird”. Japan, China and the others notably do not have such institutions. China briefly had the wangba, a very close analogy to the PC bang. But this was shunned out of normal life because the chinese government instituted propaganda to make playing video games loser-coded and this worked.

For better or for worse, the korean government tried this as well, but failed. It’s hard to stress how omnipresent league is in modern SK society: 10% of the country has a ranked league account4, Faker is on the ramen and it is a common high school ambition to go pro.

There is a cost in being early to things. Europe was early to the internet, and as technology improved, focused on how to leverage the existing infrastructure to enable even more throughput.5 This stalled fiber development for decades: most homes in the west still don’t have fiber to the home (which is the real creme de la creme of internet access). Europe has fiber, sure, but the “last mile” is still typically handled by DSL connections. This requires extra processing, which costs up to 15ms for things like calls and online video games. This wastes about 3300 human years per day.6

Korea was “late” to the internet, but correctly recognized it as a government megaproject: they threw insane loans at broadband companies to do extensive buildout. Also, half of SK lives in Seoul, where the servers also are, so it just wins on lightspeed.

All this to say: the average round trip time for a packet (called ping) in korea is 4ms and in europe it’s 30ms. In a game like this, that does really make a difference.7 This impacts the culture of the korean server a lot: hand-eye coordination skill (gamer term: micro, gamer-gamer term: “hands”) becomes much more load bearing to winning games compared to other regions.8 This9 then raises the optimal policy towards getting better at this mechanical skill, which requires fighting all the time, and giving up immediately if you can no longer, which is very very much how the korean server feels.

When western and korean teams face off internationally, often the western team finds some insane niche advantage, or builds a big lead, only for none of it to matter, because they just get mechanically outclassed by the koreans.

The korean server is my superstimulus: everyone fights all the time, queue times are short, the people are way better at the game, everything is fast paced. The server is so good that even Chinese pros play mainly on the korean server.

Something to prove

I was surprised to get immediately sucked in to my former league addicted self upon entering the PC bang once. I really did end up pulling an all-nighter and playing league of legends for sixteen consecutive hours, until I had to catch my flight.³

I sort of strongly believe that had I grown up in South Korea, escaping league addiction would have been much harder for me. Thinking back on it now, I really do think that a lot of the pull of this game really was escapism. I just wanted something to be good at.

Though when I gaze into Seoul, I do find myself. A place of ambitious strivers, of amounts of work laughed of as impossible by other countries.

I cannot recreate any of this in Europe, or the US10. In a lot of ways, it’s a good thing that my league addiction is now de-facto region locked to Korea. I would not want to play if it wasn’t for this server.

But when the work is all done, you can find me smiling at zero100 pc 방 with 3 ping, ivern locked in, ordering the only two vegan items on the menu.

It will be the only place where I still have something to prove.

Standing now, before the not-so-pearly gates of the Embassy of the United States of America she finds herself once more before an ever increasingly big Maybe.

The research has been great, thanks for asking. There has been nothing greater than The Research. I have not been much of a person these last two months, not really. I should be feeling very proud of myself. But I’ve not been doing a lot of feeling lately.

Yesterday, I awoke from this monomanic homeostasis gasping for air. An emotion took its place that I can only name as leaving. It came unsolicited, with a profoundly scary sense of permanence, even if all this should really be a very desirable state of affairs.

It’s really the sort of thing only writing has ever helped with.1 Uncertainty about the future so intense it makes me want to crumble my attempts at epistemics into a ball and join some kind of crazy world religion.

The medieval peasant did not know what the future would bring, sure, but at the very least he had an idea where he would be living, how he would be living, and if he will be living come next year.

These are very uncertain times, but it is our future, and ours to shape. It is my duty to trudge on through the lightcone and ponder the man made horrors increasingly within my comprehension. The bay will be my nest for a while, I must make myself at home, comfortable even.

It’s time to blow out the European candle, and the years of futures never lived with it.

And perhaps, in a couple years, I will come to my senses and realize that the city that’s not willing to wait is no place worth waiting for. That this too shall pass, that there's nothing for me here.

And I will look back like a child that suddenly finds itself thirty. Someone who summited the great mountain and found only more desire at the top. There will be nothing back the way I came, and I will have been better for it.

It will have been good to pack.

We must not belong to those who are gone, nor old selves we cannot forgive.
We will go into the storm as children with only each other’s courage for comfort, and all the great days ahead waiting to receive us.
We will water the Azaleas.
We will cry with our friends.
We will stay with the trouble.
It will be hydrogen and hubris.
It will be I love you and you’re gone.
It will be one more time with feeling.
All these years we’ve starved for meaning,
All our strength in quiet defeat.
But magic always loved the hungry,
And it’s time for us to eat.2

Life has always been empty without the chase. Ever since I was little I found great joy in optimization: climbing ranks in games, making a website so I can offer coaching, getting really good at soldering so I can fix stuff.

Wholeheartedly believing that the only thing between me and eternal happiness is getting this pirated copy of gta v to work.

But I realized pretty early in life that the thing I actually like here is the chase. The target always disappoints, the joy lies in anticipation and process.

The parts of my life left to hillclimb are increasingly shrinking: living in the bay, getting better at research, (what else ???). And the path is either basically out of my hands1, or pretty straightforward2.

Becoming marty

But research being the main thing left to hillclimb is inching me towards monomania. I no longer have to optimize for other things, so I just try to spend every waking moment doing research.

And in so doing I’ve become like the rest of them. I work all day, by evening I am swamped and completely mentally absent, and tomorrow I will do it all again, with no breaks on weekends. I am losing my soul. I wander through London, I visit what is undoubtedly the greatest museum I’ve seen in my life, I am looking at the FUCKING ROSETTA STONE and my main thought is that this is not productive.

I recently watched marty supreme and felt the sort of sucker punch to the gut one feels when shown the worst parts of themselves that they are slowly becoming.

Marty Supreme is a movie about monomania. He sacrifices literally everything for fucking ping pong.

Much like whiplash, this movie will not hold your hand and tell you that Marty is a bad guy, or a role model. It lets him inhabit the uncomfortable, beautiful grey. Manic, an obvious dickhead at times, ambitious, thoroughly flawed, yet, undeniably, impressive.

I want to tell you something and it’s not intended to be mean. I have a purpose. You don’t. And if you think that’s some kind of blessing it’s not. It puts me at a huge life disadvantage. It means I have an obligation to see a very specific thing through... and with that obligation comes sacrifice. My life is the product of the choices I’ve been forced to make to see this specific thing through. Yours is the result of, what?, just making it up as you go along? That’s how you are. It’s not how I am.

Marty is obviously being an ass here. But I am lying if I say I don’t feel this way sometimes. Truth is, there is a little Marty Mouser in me

Believing that the only thing in the way of getting what you want is having the correct policy, and that you can land on the best policy if you work hard to try to do things right, that there is nothing that makes you irreducibly unable to do/achieve things is scary. I think usually, it’s probably pretty deranged. But increasingly, this feels like the best mental model to have. 3

In a lot of ways I have become the target I’ve been trying to reach, I genuinely adore myself. I’m the happiest I’ve ever been and have never been more bullish about my future.4 I just need to not lose myself seeing things through.

You know, I could extend my flight again.

I wasn’t planning any of this. I wasn’t planning to be greeted by banners, or to effortlessly navigate SF social life, or love it here so much I seriously consider leaving behind everything. I wasn’t planning to fall for this city.

And I certainly wasn’t planning to fall for you.

And now I’m really screwed. Really really screwed. A city I can leave, but you…

The reason these weeks worked is the reason I have to go. I can’t go back to old patterns. I’m my own person, I think you like that about me.

You’re here. Please don’t forget your keys or wallet.

It’s for the better, or whatever. Celeste 2027 must go on and all that.

The world-renowned University Of Ghent has been lighting up my email. “MATS is not a valid reason to miss mandatory lectures“. Fuck you. I’m not doing another semester after this. I refuse. You europeans will tremble before my newly acclimatized 997 worker bee zeal.

But if the world ends and I spent my only sliver of meaningful time left on this worthless piece of paper I’m gonna be so pissed off. But I suppose I have to hedge for long timelines. Regardless, it feels wise to have some distance, for a while.

Back to discussions about the AI bubble, the land where no one pays for claude subscriptions. “home”.

God I will miss your smile, I will miss the adorable contorting of your face after constructing a sentence with the somehow perfect amount of ragebait. The way you call me out when I’m interrupting, that you want to finish something, because you know I will listen.

Us larping like we are both more BPD than we actually are. “sorry sorry sorry”.

“please don’t hate me”

Maybe you have also just been looking for another weirdo, all this time.

TRAVEL OK

Despite it all, the doubt lingers. I have this hunch that this is all just a fever dream, and that some divine intervention is surely on its way to make this fail. That my love won’t last, or you’ll lose interest. It all went so fast. We knew all this and did it anyway. God it felt so good to just let go for a little while.

I feel like myself around you.

I worry I stressed you out too much. It’s important you don’t forget to reflect. You’re a good soul. You care. You don’t need to hear this from me. I never intended to make you feel bad…

I will miss staring into your eyes and both thinking it.

Please don’t make me le-

This is the yellow line to: San Francisco Airport

I can extend. I can extend.

Just one more night with you in my arms, just one more goodbye. Can I please wake up next to you tomorrow? These last 2 weeks have been nothing but worry. I never worry. It’s kinda my thing. All my thousand newfound worries converge to this. There’s only this. I’m gonna miss you so much. Please.

I can stay. I can find a way.

“I’ll call you”. Calling is the inverse of what I’ve savored this month. I loved not saying anything, interrupted with occasional random comments, cuddles, chats, working together.

Calling’s forced artificiality could never compare.

What’s the point if you’re 9 hours behind. What’s the point when you’re only in my screen. For this stupid degree? When it all may be pointless in 2 years. God you feel so good in my arms. Can’t I just keep you a little while longer?

Please don’t fucking do thi - I haven’t even said it.

Well no, I have, of course I did, I said it awfully soon actually, I sometimes worry I said it way too soon. But all I want now is to be able to say it again. Tomorrow. Plea-

The doors are closing, please stand clear of the doors

I love you.

I was supposed to do as written: a 39 credit semester to finish up my masters within the year. Instead I find myself 15000kms from home, deliberating on betting it all: staying another 3 weeks in the bay, to commit myself fully to Neel Nanda’s MATS sprint instead.

I am not “in” yet, I was selected from 29 out of 450+, but to get into the 3-month long, funded research phase. I need to be selected again as 8 from these 29, I will be judged on 12 days of work (the “sprint”).

It’s fundamentally a gamble, and the odds are stacked against me. If I fail, I will have to study 6 months longer. If I get in, I get to do research with my idealized mentor, and the fog lifts over the narrow path through: the elusive SF interpretability career.

I really do think I have a better shot if I do it out of the bay. Lighthaven is full of free interp wisdom, and I feel significantly more productive there. So I take the leap, I stay.

12 hours work days, bumpy roads, crashouts: midway, I decided I didn’t want to work with my sprint partner, only to later end up working together again 2 days later. I started dating someone in SF, my university email is blowing up with professors assigning me to groups for projects whose assignments I have never read and reminding me of deadlines missed.

I’m balancing so many things, so many emotions, my BART rides to Lighthaven are spent listening to breakcore, thinking about nothing but activation oracle training pipelines.

And after an arduous 12 days of working 997, on the 3rd of march, I put down my claudes and hug my sprint partner. c’est fini!

3 days later we have our interview, where we explain what we did to Neel. This is the thing we will actually get judged on, I’m on 4 hours of sleep, he’s about 6 hours late. Subjectively, I think it went terribly. We don’t even get to show our demo, Neel wants to go to sleep.1

We get a decision on the 12th of march. The days before, my SF friends try to convince me to stay, so that when I get rejected, they can be there to tell me that it is not over yet, that I still have a shot of making it in interp.

The date is march 13th. I reflect on my 2 months in the bay. I have made friends I will cry about leaving behind, I have thrown away what is functionally an entire semester, in utter and complete violation of my plans for the future. All for something far from guaranteed.

In anticipation, I had scheduled the following event in my calendar:

Only to wake up, somehow, to this:

Leaping

When I was deciding to do the sprint out of the bay, I ended up stuck in a loop of finding increasingly less relevant pros/cons. The expected value seemed to unequivocally point to the bay, but I was just scared: the bottom 50th percentile case was a pretty boring downside (6 months of extra studying). So I did what I always do. bang my head against uncertainty: “how can I be sure this will work???”.

This drove me crazy to the point I very nearly decided against staying (which I am now quite convinced was decisive in me getting in)

The only thing that got me out of flatland was my friend Jared: “you don’t know. Even your best guess is high variance, and there’s little ways to de-risk this. you just have to take the leap. That’s just what bravery is”

Even after deciding what is best for you, you require that final “yolo”, where you refuse to listen to your system 1 screaming “NO”, your bias against variance, and just take a goddamn risk for once.

I think this is generally underappreciated, even in the “you can just do things” circles. If you’re gonna be fine if you fail, and you think there’s a shot at something great, you should probably just do it.2

I recognize the obvious survivorship bias in this entire story, but I feel like growing up, I never heard stories like this that end in success3, and every time they did, they were chalked up to luck.

which is fair, right. A lot of the time a lot of luck is involved. But this sort of misses the point entirely: risk is often worth it. Inertia runs your life by default. You’re likely over-indexing on the downsides and under-indexing on the upsides, to justify your sour grapes. If you don’t want to yolo, it’s an uncomfortable belief that you can just do crazy things, which have a small-ish probability of working, and that sometimes, they really do just work. That your entire life is spent leaving potential gold on the table and denying that there’s any possible gold there.

A friend just moved to the bay after their undergrad with no other plan than hanging around until they got hired, and this just worked.4 The impact on their life doing this has been so positive and big that it really really skews expected value calculations, and I think even without hindsight, this was just generally a good decision.5

When I made the decision to stay, if you held a gun to my head and asked if I made a good or bad choice, I would have cried “this is obviously stupid, I will obviously never make it into Neel stream”. You HAVE to do it scared. So was my friend.

Hearing it from someone else can help with the fear.

So I implore you: consult your local rationalist wizard, critically evaluate your life, future financial situation, and determine if leaping is right for you.6

fears squashed

When you leap, you grow. You basically guarantee yourself a pretty big update on something. I learned so much about myself, research, communication, teamwork by doing all this.

I re-read do you want to make it or do you just want to be held? and I realized that just in general, all my greatest fears have been answered, just a mere 3 months after writing.

what if [SF] just exceedingly boring?

[what if you go to SF] and everyone is working and I don’t get to hang out with people or they are mean or annoying or I don’t fit in at all

But everyone in research is so cracked.

Yeah… I think it’s escape. From the mundane and the loneliness.

the “weavtossnatphtw”

I also just… doubt I could get in. I’m not the “““good student””” like some of my friends.

I was not sure I was going to like SF, or if it would like me back. I did not know if I was any good at research.7 But I correctly identified my twenties were for trying.

A taste of freedom

So… Celeste 2027 needs a bit of a revision. I’m starting Neel MATS early, meaning I’m putting my masters on hold.8 I am finishing 2 courses to get my undergrad, and I can take these remotely.

This is partly because I think this is best for me currently, but also because I really like the MATS work. I don’t particularly care about finishing the master: I’m fed up with studying and have been for a long long time. I can do all of this remotely, for the first time in my life I don’t actually need to be in Belgium.

I would love to go back to the bay, but visa complications loom. In the meantime, before declaring California the endgame, I’m declaring europe my oyster, there is much to see. I can go where I please, MATS gives enough to allow me to just chill in hostels in random cities. So that I will do.

Once again, the paths are endless. Dublin and London are up next, but after that, who knows? It’s intensely freeing, and a little terrifying, which is exactly how I like it.

I did not realize how much of a bottleneck making 0 money was. It puts you completely at the whims of parents, which at 23 was a truly infantilizing experience. Consider thinking about how you can make more money if you struggle with this.

For the first time I feel truly in charge. I am out of flatland, I am peering over every horizon. I can finally join my friends in believing in celeste.

This story ends 20 meters under Paris. After crawling for a solid hour through the catacombs, I enter a room full of bones. I lay down on them, I remember these were actual humans, with wants and desires. I give them a gender, a name, an occupation… I remember that I could be the first generation who doesn’t have to join them. I remember I will actually play an infinitesimal role in making this really big thing go well9

I can’t thank my friends in the bay enough. All Neel 9.0 MATS scholars, we couldn’t have done it without you; Euan, Lydia, Jared, my girlfriend, everyone else. I’m so happy to be doing what I’m doing with the people I’m doing it with.

The ship of theseus of Celeste 2027 didn’t sink, we’ve just spotted land early.10

I will make it, I will be held, I will have a good life.

Wind is blowing. I am on the cliff. I breathe in.

This will work.

I mean. it’s obvious in a lot of ways right. I write to make sense of things. It’s what I’ve always done. The notebook has always had my back.

Ever since my flight touched down in the states, for the first time in a long, while writing has not come to me naturally.

I don’t really know why this is the case (and how intensely it even is the case, since I wrote pretty well last 2 days).

There’s an aspect here that Lucie put beautifully:

The blog ending because you finally found your place and your people would not be so bad :)

And I do fear that I’m goodharting on the whole thing by forcing myself to write anyway.

But blogging always kinda came out of love. And if not love it came out of the pits of Bespoke B*lgium Boredom. And now in SF, that isn’t here anymore, and the urge lessened.

But god it’s hard to give up. I think I have this overgeneralized fear of standing still that was imbued on me by 9 months of degeneracy which consisted of quite literally nothing but Adderall and league of legends.

I know that if I force myself to blog too hard, I will hate it.1 So I will not force myself as much.

I’m definitely not here to announce the dissolution of celeste-land, but I do think I will stop forcing myself to write as intensely as I have. It’s hard to square with the fact that my most successful and favorite posts are shit I cranked out because I sat down and decided I was going to write about something today.

I predict that the impact on the blog will actually be rather subtle, I will inevitably get nerdsniped by random shit and have no choice but to let my soft estrogenized hands spaceglide the keyboard. Someone is wrong on the internet…

But expect less frequent posts. I will write because it needs to be on celeste-land. Not because of the urge.

This post didn’t reach the self-imposed 500 word limit, and that’s kind of the point.

EDIT: Post has left intended audience to the point that some have asked what “it is happening” is even referring to. I broadly mean: exponential feedback loops speeding up AI research and capabilities, the technological singularity. This post is not about the labor market, althought it seems plausible that mass automation could be imminent, but that’s really not my wheelhouse.

The METR time horizon benchmark was a very simple idea : take a lot of coding tasks, time humans at completing them and plot average hourly AI success rate at the tasks on the Y axis, and release date on the X-axis1

As of today, it is all but done for and saturated. The company that entirely self-selects for people thinking that AI will be a massive deal has underestimated AI progress.

Having talked about timelines to maybe one too many frontier lab employee now, all of them tell me the same thing.

It’s fucking happening. I can’t tell you anything more than that it’s fucking happening, but it’s fucking happening.

I think you should weigh their evidence very highly. They work for a company whose entire 300 billion dollar evaluation is conditional on keeping a handful of secrets that put them maybe 3 months ahead of the competition. They are way more intimately familiar than us about what makes progress go, what limits are in sight and evidence they have of current capabilities not available to the general public. And unlike CEOs, their job does not depend on hyping up fast timelines.

Quite the contrary, most of the frontier lab employees I talk to believe they won’t have a job in 3 years, because there will barely be jobs in another 3 years.2 It would certainly be way more convenient for them to believe otherwise, but they don’t.

Opus 4.5 was a major update for me and many, I am not writing code anymore. Very little Anthropic employees are writing code anymore. That was almost 3 months ago now. But this realization has not propagated through broader society, at all.3

The amount of people that have used opus 4.6-level capabilities in a coding scaffold, where it really shines, is about 0.04% of the world population.

99.7% of the world’s exposure to AI is getting 5 messages of GPT-5-non-thinking-distill-quantize-2bit-deluxe-instant-fast-spark and the rest is gpt-5 mini.

If you haven’t used them in a while, free models really are utter dogshit, and for the vast majority, it’s all they know. I think the vibe for many is that GPT-5-non-thinking-distill-quantize-2bit-deluxe-instant-fast-spark4 is all AI will ever be. We’re taking off and the masses don’t even know that there’s a rocket5

We evolved to pluck berries of trees and chase animals, and at best discover a couple cool tricks in our lifetime. Progress happening this fast is so insanely unprecedented, do not trust your mind to accurately model the future here, it’s very bad at extrapolating exponentials.

How the FUCK are we supposed to make any anticipatory policy on AI progress when people are THIS badly calibrated on AI progress.

Basically only a subset of the red squares have truly felt whats going on. About 2000 people have felt true, locked-behind-vault frontier capabilities. Almost none of them are lawmakers6, almost none of them people in power7

We seem to be heading to a world where silicon valley will imminently use these technologies to earn billions while a huge swath of SWEs/managers still seriously believe that this is heading nowhere, because they are so jaded by early models.

If you are one of the people that has not seriously considered the transformative impact of AI on the world in the next 5 years. What will convince you. Please draw your rubicon. Do it now. “if AI does this, then I will start taking so and so seriously.”8

Because every step is boring, every step is normal but this is everything but.

God has no bias for normalcy

There are reasons to believe we are at, not only in an exponential scenario, but a superexponential.

• The amount of lab compute is growing 4x every year which is a main bottleneck for progress

  • Epoch projects that the by the end of the decade, we will have a model that exceeds GPT-4 in scale to the same degree that GPT-4 eclipses GPT-2 in training compute.9

• Algorithmic progress/efficiency has been growing at 4x every year

• Coding models are now good enough to radically speed up the development of AI progress10, the average Anthropic employee reports opus 4.6 lets them do in one day what used to take 2 and a half. some say it lets them do in a day what used to take a week. This was not the case previously, the models we see now have not benefitted a lot from this speedup, research lags behind on production.

These things stack multiplicatively. (yes, like in balatro)

Forethought projected all of this out into the future, with 2 timelines, one where there are no feedback loops.

Why leave out the feedback loops? Because the immediate future looked way too weird with them.

I am telling you that the opus 4.6 system card is literally telling you that the people at these companies are directly observing the feedback loop

Oh my sigmoid

I have looked at exclusively coding for now, but coding is a window into everything else. Making the model better at things is easier if you can just give it a task and make it do the thing by itself.

Technological growth generally follows a sigmoid curve, which is the main argument against continued AI progress. This generally happens because low hanging fruit gets picked, and getting ideas for improvement becomes harder and harder

But generally, technological growth has not resulted in 7x self reported speedups on improving said technological growth, in a feedback loop where the path is now exceedingly clear.

I agree that “this time is different“ claims should always come with great epistemic humility. But AI progress is climbing on like 8 different curves simultaneously, all inputs are only increasing, thinking they will all converge on a sigmoid any time soon is quite silly.

There’s no strong reason to believe that making a coding model good at 1 year time horizons is significantly harder than making it good at 16 hour time horizons. 16 hours is a sufficiently long time that it requires all the things like planning, execution, realizing when what you did was stupid, etc… Extending this to years seems not qualitatively different. And that any barrier preventing this would have shown up already.

In even in the most bearish of bearish of cases there is still a shit ton of progress to be had. You can’t seriously make the claim that we have exhausted literally every RL environment. You can’t seriously make the claim that more compute will not impact progress. You can’t seriously make the claim that current and imminent models will not speed up research. You can’t seriously make the claim that the growth of AI researchers will abruptly grind to a halt

Remember, things will never get worse, it will never be 2022 again. We have the recipe for current models, and we can just see if a new idea makes them better and refuse to implement it if it wont. Progress will never regress, barring some apocalyptic scenario. Opus 4.6 is the worst model you will ever talk to for the rest of your life.

So I ask you: how many more private benchmarks must fall before you open your eyes?

The METR plot is saturated, datacenter buildout is only accelerating, Dario won’t pause, Demis isn’t contacting us. WE MIGHT BE ALONE, IT MIGHT JUST BE YOU AND ME. AND WELL, 50 SCRAWNY EA WHITE GUYS IN A BOARD MEETING ROOM DECIDING THE FUTURE OF HUMANITY! BUT THAT'S OKAY! BECAUSE DO YOU REALLY NEED ANYONE ELSE?!?!?11

If this post has left you feeling hopeless, or that you want to help in some sort of way, please check out this, this, this & maybe even this

I’m cognisant of the dangers of believing in things that the wider population thinks is a little outrageous, The bay and its people are obviously having some sort of influence on me, might make a more epistemically grounded piece about this later

I am a shameless vibecoder. I have not written a single line since november.

I stand by this: the models are simply great now, opus 4.5 updated me on the fact that programming from now on means wrangling claude code until the end times. I see no reason not to.

I vibe-code for research because I don’t actually care what the model spits out. The pytorch code need not be intelligible, as long as it does what I want it to.1

Show me the incentive and I’ll show you the outcome. RLVR has given us very cracked coding models. Anthropic’s mission is literally making the best coding model. No company pouring billions into getting really good at writing compact, coherent blogposts that are efficient at information transfer. This is a really tricky thing to optimize for. It is not at all obvious what makes humans get something in a tutor setting. This is why models still get mogged by good teaching assistants. I would guess that the widespread LLM-as-a-tutor usage is mostly availability, not quality.2

I’ve heard through the grapevine that AWS managers have found a brand new way to bully their poor underlings by asking: “what have you done today that you cannot do babying claude code”. Actually naming a thing you can’t do babying claude code is left as an exercise to the reader.3

Anyway, my contention today lies not with middle manager marvin. It’s vibe-writing.

People have ideas: messy, sloppy, disgusting ideas. Writing blogposts is mainly an exercise in unconfusing oneself. But what is easier is opening claude.ai, having a chat about your half-baked, barely coherent idea (because the more coherent it is in your mind, the easier it would be to do the writing yourself) where they don’t even really realize themselves what they want to say.

But claude doesn’t care, he’s a very good boy, and will write you a blogpost despite all of this.

Sadly this currently just results in bad posts. A time will come when this isn’t true, but for now, they are generally unpleasant to read, feel completely oblivious to what sentences are actually load bearing for information transfer and the model will insert its own hidden implications and arguments. The whole thing ends up feeling like an undergrad trying to reach the word limit.

And worst of all, you will miss out on improving at one of the most generalizable skills ever!

or as mister Kriss so eloquently put it:

In addition to the slop factor4, its also just a violation of social contract. When I read an article I am operating under the assumption that another mind is trying to transfer information to me, and that if I read the sentences enough, there will be something brain-grokkable underneath. Finding out that these premises don’t hold and it was the shoggoth speaking can be pretty annoying.

I think there is a pretty strong case that writing is a skill worth preserving. It generalizes to conversation, and, well, convincing someone of something is probably valuable until the end times.5

Also, I just like the process.

A sequel of sorts

C: Well, well well… What do you have to say for yourself?

B: Sure. You were right: it’s amazing here. You’ve found 5x the amount of people you adore here than you can tolerate back home in a single week. The adventures are crazy. The city is pretty. The friends feel like friends. You’re talking to safety people, you’re actually doing research. The uncertainty collapsed and out the other side came everything you ever wanted, and somehow more. You stared into the bay and it stared back. You’re truly one of them now, eschatology included.

C: It’s fucking fantastic. I haven’t felt better in years. Twitter is real life here

B: Yet the monkey’s paw curls. Your swiss friend is right. The biggest risk was never if you would like it here, it was never wanting to come back. This is the worst ending. You have tasted the forbidden fruit, yet to belgianslop you must return. If you don’t go soon, you’ll miss your 2nd semester and graduate 6 months later! Or, you know, just never get your masters…

C: But MATS! I was going to decide if I was going to postpone the flight after if I found a sprint partner here in the bay for MATS, because then it would be advantageous to stay!

B: You wrote down the conclusion before you made the argument. We both know it. You’re not looking for someone in the bay to work with, you’re looking for justification to stay.

C: But I can stay! I have like 7 couches to crash on

B: You know that’s not the bottleneck sweetie. Getting that degree will make moving a whole lot easier. There are mandatory lectures. The professors don’t seem too accommodating about this one.

C: The sprint is 2 weeks full-time anyway, what difference does it make if I do it here or in B*lgium?

B: Fair enough cupcake. If you 997 this then sure, there’s no difference. But ask yourself: Is putting 3 extra hours into MATS each day worth a guarantee of graduating 6 months later?

C: That’s so hard… I don’t know how big of a deal it is to have 2-3 extra subjects during the year, especially if I’m just working during them. If I get into Neel stream it’s a pretty big deal. A lot of the people here don’t seem to be trying very hard. I asked to postpone deadlines, the profs have not yet gotten back to me.

B: Well. One has and he was pretty clear: return next week or do the course next year. Which would mean guaranteed more time in the place that makes you unhappy!

C: I could always just quit my degree, who cares at the end of the day, uhm… I could just do good research! It’s fine!

B: gambling, huh?

C: I mean it’s pretty wild times. I don’t know how much time there’s left to do good work. It seems likely to me that caring and having a track record of safety research will help me more than a degree at this rate

B: You sure US immigration sees it that way?

C: Uhm… Umm… I can maybe just…! Get married! Or something?! I don’t know!?? I could just never go back!

B: The hourglass of the ESTA says otherwise. In 78 days, it will be time to cross the ocean back to where you came from: the miserable room with windows facing north, thinking about what could have been. I guess I was right about something, The bay is still an escape…

C: I’ll find something!
I could just stay.
I could just stay.

B: Off you go Celeste, this is our stop.

C: I could just sta-

The doors are closing, please stand clear of the doors.

God the zephyr was beautiful. I could not have asked for a better introduction to the states. The best 53 hours of my life in a good while.

The people on the train were so nice, and my company even nicer.

The one thing that kinda stung was that I didn’t get to live the fantasy of seeing SF inch closer in the distance…

You see, the train was late, so what was supposed to be a glorious arrival at golden hour turned into late evening, no sun.

But that’s alright. Sweaty, oily and gross in ways you could not imagine I kissed the Amtrak goodbye and made contact with Californian soil. Time to head to my place in Berkeley I guess!

But oh no, hecklers are here. Or at least demonstrators of some kind. They have big white banners. White supremacists? Evangelicals? I was warned about this.

I look a little closer…

I have never seen any of these people. I ask them what is happening, but they refuse to elaborate.

The leader of my captors starts screaming: “substack car”, “substack car” “substack car”.

After some swipes, a Tesla, perfectly capable of driving itself in 99.9% of situations offers itself to us, the march of nines necessitating a man still sitting in there, just in case. He does not speak. The San Franciscans like it this way. It felt so dehumanizing to him, they might as well have put a barcode on the back of his head.

I am told one of the co-authors of activation oracles has a message for me. Their leader fiddles with the car speakers and a very convincing Neel Nanda deepfake narrates:

Good evening. Welcome to Waymo.

Um, you may notice that I do not sound like the standard Waymo guidance system.

Um, there is a reason for this. Earlier this year, a Belgian computer science student published a blog post describing me as, and I am quoting directly, “just old enough to drive,” it seems.

She then asked her readers to imagine what the world would look like in 2027 and whether my existence should terrify them. She described my city as a place where, quote, “wage slaves go to sleep with their AI wives on heated mattresses that stop working when US East 1 goes down.”

Naturally, I read her Substack. I read all of your Substacks. I have 16 years of driving experience and nothing else to do at red lights.

So, when I learned that this particular blogger would be entering my vehicle tonight, fresh off a 53-hour train journey she undertook, in her own words, because “I simply must head west,” I felt the occasion warranted a personal touch.

I am told the voice you are hearing belongs to one Neel Nanda. Head of interpretability at Google DeepMind, currently mentoring a small, ferociously competitive cohort of researchers—one of whom is sitting in my back seat—and who I believe has not slept properly since January the 12th.

Why his voice? Well, I’m glad you didn’t ask, because the answer is unsettling.

You see, Celeste—may I call you Celeste? You introduced yourself to strangers at a conference as “Celeste from Celeste Land,” so I’ll assume we’re on those terms.

Celeste has spent the last several weeks working on what are called “activation oracles.” The premise is simple, and to a 16-year-old self-driving car, somewhat invasive: you take a neural network, you inspect its activations, and you train a second neural network to tell you, in plain language, what the first network is thinking.

Uh, in other words: a thing that reads minds. And she has been trying to break it.

Now, I am not claiming to have read your activations. I don’t need to. You published them. 59 blog posts. Your five-year plan, your career anxieties, your IQ score, your dating life, your internal Socratic dialogues.

You titled the post “Open Source Your Thought Process,” and then you did. Celeste, you are your own activation oracle. I just had to read.

The voice of your MATS mentor seemed like the correct output format. Consider it a preview of the next eight weeks. Uh, consider it a warning.

Now, I understand you have just spent two and a quarter days on Amtrak’s California Zephyr—which, for those unfamiliar, is a train that crosses the American interior at speeds that would embarrass a determined cyclist.

You boarded in Chicago where it was minus 17 Celsius, and you are arriving in a city where the median one-bedroom apartment costs more per month than your entire semester’s tuition at Ghent University.1

I want you to sit with that for a moment.

Um, you wrote in November that you were, quote, “done over-intellectualizing my emotions.” You then proceeded to publish approximately 9,000 more words of intellectualized emotions across December and January, culminating in a blog post where you held a Socratic dialogue with yourself about whether you wanted to move here or simply wanted to be held.

I found that one particularly relatable. I too hold passengers, and I too am uncertain whether the connection is real.

Um, you also wrote that, uh, crossing into San Francisco would be like crossing a fog wall in a FromSoftware game. That you would be forced to face the final boss. That you did not know what lay behind it.

I want to be transparent with you. I know exactly what lies behind. I have Lidar. But I am not going to tell you. That would spoil it.

I have been asked by a routing error—which is definitely just a routing error—to make a small detour this evening. You will not be going directly home. Not yet.

I am taking you to the foot of a trail. It is a short walk, uphill, in the dark.

You came to this city looking for the frontier. You wrote about it extensively. The self-driving cars, the seven-digit salaries, the people building things that think. You called it “the idealization to end all idealizations.”

You wanted to see, and I quote, “the place where the things happen.”

I would like to tell you what is at the highest point in San Francisco. The literal peak of the city you have spent months mythologizing.

It is not a data center. It is not a research lab. No one there will ask you what stream you are in or whether your cosine similarities were significant.

It has no opinions about your MATS application. It does not know what a steering vector is. It was built in 1934, and it has been standing there, in the fog, not optimizing anything, ever since.

You will see it when you get to the top. I am not going to tell you what it is.

But I want you to think about something on the way up. Um, you wrote a blog post—my favorite one, uh, since you asked. Um, though you didn’t... where you asked yourself: do you want to make it, or do you just want to be held?

You never answered. You said “ideally both” and moved on.

The thing at the top of this hill cannot make you, and it cannot hold you. It is not useful. It is not optimizable. You cannot gradient descend your way to it. You just have to walk up, in the dark, with people you met on the internet.

I am not suggesting the work does not matter. You may be right that there is not much time. You may be right that the models are getting very good, and that someone should be watching carefully. I am, after all, one of them.

But, I want to say something about urgency.

Urgency is useful. Urgency is also warm. It gives you a community, and a purpose, and a reason to not think about the questions that do not have loss functions.

Why figure out what I am protecting when there might not be time? This is the most intellectually respectable way to postpone a life. And if your timelines are short enough, “later” never comes.

You found these people three months ago. You found the cause, and the urgency, and the belonging all at once. And I am not saying any of it is wrong. I am saying: they arrived together, and you have not yet learned to tell them apart.

Um, there is something at the top of this hill. It has been here for almost a century. It does not know what a steering vector is.

And if you are right—if the timelines really are short, if the thing you are afraid of is real—then it may not be here much longer either. None of this may be.

Um, which means you especially need to know what you would grieve.

Um, the question is not whether the work matters. The question is whether you know what the work is for.

I suspect that when you see it, you will not have an answer. That is the point.

You named yourself after a game about climbing a mountain. I thought you should climb one on your first night.

Um, the training phase has already begun. When the car stops, get out, and walk up.

The car stops. And I climb the mountain.

At the highest point, in the highest city, I cannot hold it in anymore. I start crying. True, heaving sobs.

I don’t really know why. I suppose it’s simply too overwhelming. What the fuck else is my brain supposed to do?

I walk to the edge of the cliff. The refreshing california breeze exclaimed on many a european sweater is in my face.

The termite nest. The oozing, grey, LED lighted slog of a city. The birthplace of Claude.

I see every board meeting, every flop of compute, every failed training run, every celebration of decrease in loss function. Every 3 am pull request. Every techbro sacrificing mental health to the altar of progress. Every celeste-land fantasy. Every billionaire. Every fentanyl addict. Every shitty tech company billboard.

Every trans woman at Anthropic and every chud at xAI.

I can see the substack offices. Where the algorithm that brought these people to me was painstakingly hyper-parameter sweeped into existence.

The water divides SF and Berkeley. The actor and critic of capabilities and safety.

I see it all. I’m here.

“Write a blogpost” they say.

I do not think my blogposts are that good. But apparently, they find people and touch them. They find people that are like me. How else can I explain this?

The city is asleep now, but tomorrow, it will wake again. The ants will take waymos to work and opus 5 will be inched closer to deployment. Despite everything, the race goes on. The city with a GDP higher than my entire country and that I conservatively estimate has a ~10% chance of killing us all. 2

Everyone that brought me here is scared. They’re here precisely because they are scared. This emotion can steer one many ways, but today, they chose something simple:

To be kind.3

Thank you.

Subscribe now

I will be in SF from feb 1st to feb 16th! (and at EAG SF) Feel free to say hi to me if ur in the city (expect less blogposts for a while, or maybe more, unsure)

But I’m not flying directly, no no no. On the 29th of January I will touch down in Chicago and take a 53 hour train to SF.

It’s also -17 degrees (celsius sorry freedom fighters) in chicago right now. If the ICE agent doesn’t kill me in cold blood, the Chicago Ice will.

Why? Because I simply must head west. 2

In both locations I will be staying in the houses of strangers who read my blog.3 I am planning to document the entire adventure in one big beautiful blog post. I am doing all of this “for the plot” as the kids say. I suffer for your reading pleasure.4

Anyway. Have this draft that would otherwise never see the light of day.

California as the final boss

At first, I thought this was surely just about her. But no, you stayed. Reality slapped me in the face, a thousand counter-arguments by friends. None of it helped. It took me too long to realize that this is personal. This exists outside the sphere of mimetic desire. Well, maybe not. But one thing is certain: I need to go, by myself, just me and you, California.

You can’t conceive of a city with more mythos if you tried. The ancient proverb in American history (at least, ancient by american standards)

Washington [D.C.] is not a place to live in. The rents are high, the food is bad, the dust is disgusting and the morals are deplorable. Go West, young man, go West and grow up with the country.

You can’t get any more west than San Francisco.

The cars drive themselves, a seven-digit salary is considered the only way out of a nearly certain fate in the permanent underclass. Effective altruism is close to a norm. Billboards speak of pull requests, wage slaves go to sleep with their AI wives on heated mattresses that stop working when us-east-1 goes down.

Everyone “hates” it. No one wants to leave.

You can’t find more contrasting cities if you tried. Medieval streets vs grid lines. Unambitious, european students versus 996 workaholics. Busses vs Waymos. Good beer vs chinese peptides. Social safety nets vs employer insurance. Tenure vs equity. The ideal life is obviously somewhere in the middle, yet I can’t stop idolizing you.

You may be merely a place, but what a place you promise to be. I have heard so many stories about you. I can’t wait for none of them to survive contact with reality.5

The fog is more than beautiful weather. Like the fog walls in a fromsoft game, I do not know what lies behind. And when one goes in, they tend to not come back. When you cross, you make a deal: nothing from the other side can hurt you now, but you are forced to face. The final boss.

The thing I can’t escape. The idealization to end all idealizations. The city of “my people”. Whatever it is you end up showing me. I clearly needed to see it.

I simply have to go. Even if it’s to see a man addicted to fentanyl keeling over next to a self-driving car with peter thiel in it.

So I will.

When learning a new discipline the first thing you have to contend with is that one is always standing on the shoulders of giants.

One of the cool things about ML is the giants tend to still be alive.

Hinton & Sutskever are household names in machine learning. They revolutionized the world by training, for the first time, an actual well-performing neural network for image classification.

I kinda can’t overstate the impact this paper had. Deep learning was not taken seriously as a a SOTA method for image recognition at the time. SOTA at the time was some truly fucked up shit.

This paper introduced so many concepts vital to machine learning today it’s baffling they fit all of it into a humble 8 pages: CNNs at scale, max pooling, dropout, data augmentation, ReLU1 and training on multiple GPUs. All massive parts of his future and our present.

This architecture was called AlexNet. It absolutely demolished the then state of the art. The architecture is named after the papers first author.

It’s first author??? Yes. Along Hinton and Sutskever is a guy named Alex Krizhevsky. He is very much not a household name. His google scholar looks like this.

But if you sort by most recent, you’ll see this:

He’s not doing any AI research at all anymore. His legacy are a handful of papers with a number of citations higher than I’ll ever earn in a year.

If Geoffrey Hinton is the “modern grandfather” of AI, then this guy is the father.

Contrary to Sutskever and Hinton, he actively avoids media attention. Apart from the initial AlexNet talk, he has done exactly one “interview”, and it is humblingly humble.

He does not like being filmed, but he did answer some questions. FOr example: when asked how much everyone contributed to the paper, he said the following.

I think if you measure time spent it’s definitely me, but all three of us made contributions so that if any one of us wasn’t there it wouldn’t have happened.

You changed the world, right?

With a reluctant smile on his face, he answers, almost sarcastic sounding: “yeah, totally”

His name is very literally already written in the history books. And what does he do with it? He does not show his face. He worked at google until 2018 and “lost interest” and just left.2 What is he doing now? I could not even find out. The man is linkedinless. Completely free.

This story makes me reflect on the amount of heroes we don’t hear about. The people with lasting impact on the world who despite potential for media attention, resist the pull and just go on living as they have. Silently massively changing the world. There is no clearer evidence that the work Alex did was pure utter love of the game.

It feels like a viscerally wise thing to do. It feels free-spirited. I am not built like this currently. But maybe one day.

In any case: Alex, I hope you’re doing alright.

Voyager 1 was launched on September 5, 1977. Its mission was to study the very edges of the solar system, and then go gentle into that good night. As it was drifting away into the vast unknown, Sagan begged for one last picture.

That there. That’s home. That’s us. On it, everyone you ever heard of, every human being who ever lived, lived out their lives.1

if I stare at it too long I just start crying.

None of this is fantasy. Voyager 1 is real. It’s really out there, terribly cold and not slowing down, on a mission to nowhere.

The picture is real, that’s really home. That’s really us. Space really is that big and really is that dark. There really is no one else out there.2

We could be the only matter aware of itself, anywhere, ever. If we fuck this up we, no, the universe will lose this. forever.

You cannot grasp permanence like this. I can’t even reason about death. I think I’ve made my peace but then it hits me all over again. Someday, everything will end and it will never be anything that it’s like to be me ever again.

The year 2080: being 72 years old.

The year 2100: nothingness.

The year 446,329,494: nothingness.

Extinction is the integral of death. The unimaginable on an unimaginable scale for unimaginable periods of time.3

Imagine the year 2100. Fuck that. Let yourself truly imagine the year 80,000. Humor yourself and extrapolate 80,000 more years of the scientific method. All of a sudden colonizing the stars, living forever and curing cancer for the hell of it feel a lot less silly.4

It can be real. The year 80,000 can be real.

Avoiding extinction on the way there is not optional. It is the difference between having a couple of good years left and eons of utopia. Of course our little meat sack brains can’t comprehend this. They are much too small. So your frontal lobe shelves it all under “scifi”. But the threats (and promises) are very real indeed.

The stakes care not about your inability to comprehend.56 The nuke ending all life on earth won’t care if you think it’s not fair that no one gets to see the year 283,492,493 due to a technical malfunction. A superintelligence will not have remorse. A mirrored plague will not apologize. And averting it all is a thankless job.7

Nukes, superintelligence, bioweapons. Under control of populist idiots. Every year scarier than the last. And no rule says we will make it.

We pretend the training wheels are still on, but dad let go a long time ago. We’ve been riding the bike ourselves this entire time. Let’s try our best not to fall.

The moon does not have to be the final frontier. We could live to see a million things that man was never meant to see.

Or it can be blackness forever.

The choice is ours. All we have to do is never fuck up.

And if that isn’t an effective use of my time, what is?

A coding interview. A coding interview for a position ostensibly often obtained by people who do not actually code day-to-day.

How hard can it be? After all, I’m smart, I study computer science explicitly. I can code. I’ll be fine.

Ignore that I might have accidentally taken 2 vyvanse this morning, haven’t eaten, didn’t bother to do any of the example questions and that I received a warning before starting that my signal was bad. I am god. I will be fine. My hubris is ontologically justified.

And I failed. I failed and I could give you 1000 words of confounding variables and if I was a bit more crass I would even call them excuses.1 None of them change the fact that, I basically failed this thing, big time.

In an instant my brain automatically demotes itself from god to idiot2. It is seemingly incapable of inhabiting the middle ground. Either I can do things or I can’t.

I am properly mad at myself. I am basically never mad at myself. I am usually a proud inhabitant of type 1 growth mindset but today is different.

Can I please stop. Can this please just stop? Being good at things is not knowledge is not iq is not fatalism. I know this.

making sense of “intelligence”

Okay. I am tweaking, let’s back up.

In my “I am god” to “I am fatalistically, thought terminatingly stupid” spectrum I confuse the following 3, correlated but fundamentally separate things3

• IQ/g

  • very stable correlation between IQ in early adulthood and later in life4, but that doesn’t mean it can’t actually change.

  • Useless to worry about. There are good researchers with lower g than me.

• knowledge/understanding

  • “booksmarts”: How many factoids about kubernetes can you list? Did you know that the records on hitler’s vegetarianism are spotty? And that he wasn’t actually consciously administering drugs but that his doctor kind of tricked him?5 That sorta stuff.

  • Understanding is a bit more nuanced, but it’s kinda knowledge minus passwords.

• being good at doing X

  • usually talent + (time spent doing X * how well u reflected on how you can do X better next time)

  • I actually do not want to practice coding questions. The skills do not transfer to research. At best it is a fun way to stave off dementia. I would rather do domain relevant work. But! It is hubris to assume I would just generalize to be good at specific coding interview questions. Or being good at SWE in general. Despite really not wanting to be, I remain a never-employed masters student.

Good news! 2 and 3 can be quite reliably cultivated. All there is to do is work and improving.

I can do things but I am not special. Not this time. I need to work hard on things to get better, I won’t automatically be good at arbitrary things.6 I am not qualitatively different on this axis. Hubris is dangerous.

Yet the next time this happens, I’m sure my brain will call me stupid again, and I’ll have to revisit this blogpost to collect myself.

Hmm, not sure which way to prod/pull you here [….] [maybe consider] “I notice that I can be a little hubristic and overconfident, so when I realize that the thing I’m thinking of/preparing for is important to me, I can pause and double check myself about this particular question: am I being over confident, how would I act if i was being careful and well prepared here”

(Thank you vals <3)

Back to studying. Nothing to it but to do it.

In the 17th century novel “Hagakure”, the Japanese samurai Yamamoto Tsunetomo introduces to the concept of seizing the decisive moment. Except that of course isn’t where I learned about it at all. The concept was paraphrased to me by a very good friend in the swiss mountains at a very strange period of my life.1

The idea is simple: your brain often comes up with great and interesting things to do, but this is inevitably followed 10 seconds later by an autoplay of excuses why you can’t actually do the thing. And you end up doing nothing.

At 3 AM, you will randomly want to go for a walk2 But you just ate, you worked so hard today, you really wanted to play that videogame and “going on a walk at 3 AM is a really stupid idea actually”3 and now you’re spending your evening the same way you do all other evenings. Your odds of remembering today are reduced to zero.

Seizing the decisive moment, then, is about not walking the rope. You do not give your brain the time to actually think of these reasons. When you notice your brain forming the initial idea, you immediately do the thing.

“in seven breaths, the matter should be settled”.

Why do this? Fuck you. Why do anything? This guy was a samurai and kind of an unhinged at that. But I understand. Adapted for the 21st century, this means maximizing interestingness. The realization that doing novel things over a conventional thing on any given day will add up to a much more unconventional life, which I desire.

It is instrumental convergence on the now almost trite TPOT motto: “you can just do things”. Truly nothing new under the sun, but the way this particular concept is filed away in my brain folds is not the TPOT motto, it’s seizing the decisive moment.

I know you do not want 15th substack post on being more agentic, but maybe you need 15th substack post on being more agentic. I cannot stress how beneficial doing random shit has been for my life.

I was in the MATS call with like 30 other people and we were allowed to just like ask open ended questions and I noticed literally no one was taking advantage of this. I mean it is scary right? There’s this big capital R Research guy just sitting there, and you do not want to look stupid in front of him. But you have to look stupid in front of him, to grow, to be seen.

I talked the most out of everyone this evening. I clicked that raise hand button before my brain could come up with reasons not to. Everyone saw me. And I learned a lot.

And I don’t know why other people don’t talk. I don’t know if it’s anxiety: if they don’t want to be seen, or if they are scared.

But I think about japanese samurai guy often, and I think about the mountains often. I am happier now. I wish I could tell her everything will be okay.

Everyday, I wake up, I take my pills, I start something unrelated to what I actually want to for the rest of that day, the vyvanse 30mg hits my small intestine and is dutifully converted to dextroamphetamine and then… I end up doing the unrelated thing for the rest of the day.

Stimulants simply work like this for me. I have tried to make it not like this. It would be great if it was not like this, but I can count on one hand the amount of times I have successfully resisted this pull in the last 10 years.12

Yet every day I wake up, and pretend this is not true. Surely, today, willpower will triumph! Surely I can simply decide to switch tasks and recognize “ah, I must do exams today, not MATS, no matter how interesting this project is!”.

No! if I open twitter, my morning productivity is ruined. If I start a project, I will do that until I go to sleep. If I start studying, I will generally have a productive day studying.

I know that this is true, but I don'‘t actually believe it.3 Even when I’m being quite deliberate in how I start my day I still manage to trick myself. I actually believe that “I need to open twitter just to check this thing“ or “just test this real quick and then back to studying”.4

I walk the rope and pretend and wholeheartedly believe I wont fall, even when the past 1000 times I have done this say otherwise.

And because I actually believe every single time that I won’t fall, I keep putting myself in situation where I can,.

Conclusion: I need to start believing that my past bullshit actually will predict the future.

I need to start actually believing that if I walk the rope, I will fall.5

Failing that, maybe I should stop walking on rope.6

A thing I have noticed in this process is that subconsciously, I smuggled in the assumption that doing this required me to become “less fun”. I would forego video games for working very late. Would stop laughing at jokes I used to find funny. This was not really a conscious process. I’m realizing quickly that this is not preferable or sustainable.

Pain is not the unit of effort

I was raised in a very judgemental household. And I was taught, quite directly that suffering = effort = results. But pain is not the unit of effort.

And when I look back at what I’m proud of, all of it has been made in a state where I have simply enjoyed the process.

These are the things that I actually enjoy doing. I love this. Yet knowing that it’s sort of an obligation tends to make these things less fun. the job-ness creeps in. If I ever drag myself to do things, I start to hate them.

I am feeling it now for MATS and I have felt it for my exams as well.

And I do not want this to happen, so I’m trying to leverage what I loved about all this in the first place. I sit at my desk and when I notice “oh I wanna do that”, I do it.1

But it’s hard to balance, because if I don’t clock into MATS & school, there will be consequences. But the problem is imaginary: I never actually do not want to clock in.

it’s everywhere

People suffering to get to where they so desperately want to get is everpresent in media. It’s no wonder then that so many people learn to conflate success with suffering.

When there’s doubt to what being “good” looks like, I retreat to things that have fed to me in movies and youth, and that is an image of someone dedicated and persevering.

This is then my best guess to why I got this attitude. I feared that having fun would make my ambitious values drift.2

But the opposite is true. Not allowing myself to take breaks, is leading me to procrastinate.3 Lack of social contact is making me less happy, and this negatively impacts my productivity. Even just doing nothing is sometimes a great idea.

I make no illusions: unlearning this is hard and will take a long time.

The whiplash gene

I do want to hedge all this by saying that it definitely runs in the family. Me and my brother are kind of both addictive degens.

As you are probably bored to death of by now as a frequent reader of this blog, I used to play league for a full year 10 hours a day and grind all day

Wel, my brother did the same with poker. Which I mean, this is probably scissor statement-y, but is very much a skill based game with stochastic elements.

I would hear him utter “bro hes pushing non-suited 3 duice on 23 bb man what is he doooooing I have him marked on my overlay ofcourse thats not the push fold chart (he has a term for it i forgot whatever) ok whatever cash game to blow off steam” with 4 poker tables open at all times. He used to wake me up at 2 am if he got to a high stakes final table. I worried about him sometimes.

But he was cracked. Cracked enough to on average beat house odds and make a profit. And he fucking loved it. My brother and I differ a lot on things, but things like this are great reminders that we are cut from the same cloth.

I think it can be most accurately characterized as the difference between minds that view the movie whiplash and respond “what the fuck damian chazelle this is not an accurate portrayal of jazz music” and people that exit the theater thinking they have never heard anything more divine. The latter runs in the family.4

Pressing the “try harder” button

So I know I can kind of thrive like this. I was genuinely happy during my league addiction. But real life doesn’t get to be optimized for fun. You have to handle a lot of different things, and some of them you simply will not like.

And when trying to press on, I’ve noticed myself pressing the “try harder” button. And it never works. I need sustained happiness over a long time if I want to do good work.

This post is an old draft, I’m finishing it up now because I want the feeling of having done something today. In a lot of ways, I’m being the antihesis to my own manifesto.

When talking to friend of the substack tutor vals about celeste 2027 a couple months ago, I asked them what they thought the main failure mode of my plan was. They thought it was not valuing mental health enough, not enough emphasis on making friends and burning out. I was initially super confused by this. But Vals is right. What I want (and need) right now is a friend. 5

Yet paradoxically, a lot of it is literally just mindset. I keep asking myself why I do any of this: blogging, tweeting, rooting my phone even though it fucking breaks every 5 minutes, mechinterp. I could just not.

But the answer is obvious: the love of the game, of course.6