moltbook roundup, no surprises with that thing given my insider view though. laura deming with some soulfulness. my friend sophia is very cool, she wrote about chess engines.

one of the many things i dream of doing is learning languages to read texts in their original language. foucault est peut-être abordable, but i’ve given up on all the soulful soviet content and chinese texts. a new contender is the memoirs of lady hyegyŏng which is enticing both in content and in the fact that i have principles against learning japanese because of the hell of learning it and hangeul is a masterpiece.

Bill Joy, Why The Future Doesn’t Need Us. This is one of the more prescient pieces written about AI risk (2000). While I have reasonably high numbers for P(doom) and great respect for tail risks and buy into the great personal responsibility of people responsible for the technology, I do resent the luddite framing. For more luddism, there’s this piece from Now-Time featuring William Blake. May God us keep, from single vision and Newton’s sleep.

welcome to 2026! here’s the Dan Wang letter! Dan Wang is one of the most epistemically clean people I know of and it’s been a bit of an inspiration. As a LessWrong Rationalist, my practice is the Truth, but in the past couple years I’ve deviated a lot and had more fun with ideas in a manner where artists might paint still life before doing working in the abstract — but I’ve started to head back a bit. The world as we know it is ending soon (maybe for something worse, maybe for something better) but I’d like to pay my respects through understanding.

im mad about the tasteless

usually when i think about taste it’s in the context of research taste, but the nature of taste generalises pretty well, including to food. there’s some skill in developing taste and you need samples and time — but those are reasons to have no taste, not bad taste. to have bad taste you have to have fallen to a taste serpent. there are generic ones like ego, but there are also more specialised ones. in research it often takes the form of a nerdsnipe, and in food it’s often virality.

i want to tell you about two korean spots in san francisco. sohn (usually styled in all-caps) is disgusting. line out the wazoo, poorly seasoned stale fried chicken and fried rice that is impressively inedible. don’t let the reviews fool you. the serpent has captured those of poor taste. no one i know enjoyed it, not even the tasteless.

on my bedroom wall i hang a menu from a restaurant that now feels like a dream: joodang. fantastic cocktails. creamy seafood kimchi udon. miso crab. the best grilled eel in san francisco somehow. after discovering it i went weekly, to their empty restaurant in the tenderloin. on the fourth week they told me it was their last night.

i hope taste-jesus returns and separates the taste-sheep and the taste-goats. i hope he slays the taste-serpents. i hope joodang is enjoying eternal life with the heavenly taste-father.

marty supreme

i think this movie ought to be able to cure anti-semitism. it was a bit cheap with the anxiety-scenes but the dialogue and charisma in that movie is unmatched. chalamet can’t act as anyone but himself but that is alright. he is obsessive, neurotic, volatility seeking and anhedonic in all the most beautiful and most jewish ways. though it’d be very male of him to fail in grandeur and embrace fatherhood as a backup, i know in my heart he’s abandoning that baby (also very male of him).

no other choice

every time i think i will stop being obsessed with korea they deliver on something new. yes all the big korean movies are about capitalism, which to americans can feel cheap but they do it so much better and deeper (just like they did the zombie movie) in large part due to the fact that they also do capitalism harder in real life. no other choice has absurdist elements and comedy that are familiar from parasite, and the same lack of cheap anti-rich sentiment that americans have caricaturised themselves into. it does a fantastic job at balancing the virtue of a man who cares so much that he will do anything it takes (no other choice) with the nihilism of the whole situation.

internalised fear of pedophilia

Epstein. What a horrible situation, what an exciting release of documentation. If you haven’t already seen jmail.world, I highly recommend it.

I think it was always probable that a nontrivial slice of the population would, given the chance, enjoy the island. As dramatically immoral as Epstein was, the more unusual and the notably dangerous thing about him was the unadulterated grit and obsessive power hungry behaviour. This became apparent in the sheer number of emails he sent.

The fact that pedophilia is so widely regarded as one of the most vile things someone can do is a strange attribute of human morality. It was a relatively recent development, didn’t require any sort of “movement” and perhaps one of the most bipartisan issues. I think we got here in large part through sexual shame — the same self-watching, pre-emptive kind that was used to produce internalized homophobia. The internalised hatred for pedophilia has very positive externalities even though I think its not unusual nor indication of bad character to have pedophilic desires. On net, and even if i’m wrong, I think the fact that we’ve really condemned pedophilia is beautiful achievement for human morality.

The whole article kind of feels like a prank, but read a little bit just to see the rising height of Chinese people. Mandatory Yao Ming mention — this article is surprisingly good at characterising the Party and comes with this hilarious video of his family trying to react to his draft pick. Why the Chinese elite are in Japan, which notably advertises a Chinese-language bookstore in Tokyo I’ll trawl. They rudely don’t include restaurant recommendations whilst talking about how you can get good Szechuan food in Japan. This Chinese vibe coder is doing it right.

Rising wholesale electricity prices. The Face of Ice is a good way to think about this whole ASI-is-coming-are-we-ready thing. I would probably enjoy Ursula K. Le Guin but I’m busy with the whole ASI-is-coming-are-we-ready thing. A beautiful story about electricity and whispers of my socialist-libertarian utopia. There’s really good New Deal posters and other sorts of propaganda at the Poster House in NYC that’s worth seeing. Rocket scaling.

What happened to us? Why isn’t this a thing anymore? Other than the ASI thing I was really born into the wrong era.

Insane acts of kubernetes. Brief message on an AI incident? Rarity: an ML paper I liked. Beautiful plots about newlines.

Bugonia

Many of the Lanthimos movies are kind of pervy, which I love (my favourite continues to be Dogtooth). Bugonia opens the movie by castrating the male characters and shaving Emma Stone’s hair off in a clean establishment that no pervy business will be going on. It was a strong opening for me (and presumably for beloved Avator director, James Cameron) as Jesse Plemons explaining why castration is a good idea strongly resembled a conversation I had with a friend a few weeks prior. Bugonia is about intellectual and class nihilism, the two most important kinds of nihilism! It’s also kind of about being your authentic self — something which one must never be nihilistic about.

Crime and Punishment

I get the hype with Dostoevsky: the characters are consistent embodiments of particular archetypes, the prose is functional, and everything is filled with emotional honesty. There’s a sense that he has a lot of distaste for all his characters though doesn’t allow anyone to be single dimensional.

The first two thirds of the book are slow with the characters psychologically accurately repeating themselves. It does pay off in the end, particularly because I was reading the scene where Raskolnikov most uncouthly expresses his exceptionalist beliefs when I got the text that Caroline Ellison was being released from prison. These themes around intellectually appealing but morally corrosive ideas are familiar to me — they should be familiar to all Bay Area people. Raskolnikov’s feeling isolation and guilt are highly redeeming qualities, in reality people are far more internally avoidant. It makes sense that this is the Dostoevsky that the Silicon valley favours, but I suspect it’s a Power Broker situation where readers somehow land on the wrong side of what the author is trying to communicate.

The foils in Crime and Punishment are obvious and extreme, and would be exceedingly so if not for the fact that they manage to complicate the characters. I’m at least a few standard deviations above the mean for how much romance I read out of it, but Sonya and Raskolnikov’s connection is deeply romantic. It doesn’t really matter if they like (or love) each other, but that they are very much programmed to exist well in each other’s lives. I think that’s romantic.

Frankenstein

I’m so glad this movie was done well, the book is really unmatched in terms of literature about father issues (in both directions). It’s ridiculous that vampires became the goth fixation for teenage girls. I usually consider the angst of a love-hate relationship and clinical loneliness from fatherly abandonment to be kind of a cheap thrill, but Frankenstein dresses it up for me enough that I can enjoy it.

An update on California high speed rail. A briefing on cryonics.

It’s not the incentives, its you. My libertarian mindset is really into seeing the world as a pile of incentives and I love making the incentives good. That being said, I do just think that exceptional people overcome the incentives. A beautiful short story, the Innocent Gene. What are the jhanas? Jose again on doing more things. Reality has a surprising amount of detail, and its a pleasant and epistemically fun thing.

If I was born a lot sooner I would’ve been in a scene with people like Gregory Bateson, they’re spiritually a lot like rationalists. I got a shirt with “Rigor alone is paralytic death, but imagination alone is insanity” from the official Deep Springs merch shop. Expectedly, the Queer had really good publications back in the day. Exhibit A, B and my favourite: C. Wittgenstein as a designer and more Now-Time Wittgenstein content. When I was a kid I was told that different parts of my tongue taste things differently, which I initially didn't believe. So I took a sugar cube and applied it to different parts of my tongue and then saw the truth. This week I learned that I just psyched myself into thinking that and your tongue tastes the same everywhere and empiricism failed me. Never update your beliefs.

Do you think I could get a prescription for this? The average font. The stepped reckoner! Someone should really 3D print these. The Lion of Gripsholm Castle. The NFL is uncrustables georg, eating at least 80,000 uncrustables per year. Nothing can make me understand the British psyche but cheeky nandos got me close. Sadism is named after a guy, who is of course a Frenchman.

Have you ever wondered who had the coolest career progression? It might be King Zog, who became the first King of the Albanians after being President and Prime Minister. Murder Inc, and a beautiful walk through Brownsville where I intend to go grab some Jamaican food and try to spot churches that were synagogues. The Mao Wikipedia shenanigans. On the bright side, some revolutions went spectacularly well, on the downside, it does justify some manner of Deng Xiaoping’s actions. The Orange Alternative is even technically still active. Korea had an incredible show of civic duty protesting beef imports. I really think about being born in the past somewhere in Eurasia a lot.

Theories about Deepseek hype being manufactured. Honestly regret reading it, I don’t need to care if it’s true. A snapshot into OpenAI’s culture. I hope people everywhere are writing lots of documents now so we can look back and compare notes. Never meet your heros and never ask for their AI takes.

Peking Opera Blues was an okay and funky movie! Bugonia was perfect. I love that Yorgos just knows which of his movies will be box office hits and which won't. I love that he opens the movie with castration and balding of Emma Stone. Don't get spoilers.

Claude 4! Good blog post describing megakernel. LLVM PR with clever work done by Claude! He also has a cute lil jailbreak. Steve Klabnik (🦀) on AI discourse, really pleasant to see old software engineering idols weigh in on my life’s work. Some thoughts on how AI will change things, need more of this content with higher specificity and increased number of rollouts to actually understand though. Dario on regulations and transparency. I had never heard of a Rick Rubin! I think he could be an important concept, as a producer who maybe doesn’t have the skills for making music yet still provides real value.

Sam Altman’s kitchen (we use the same olive oil). Scott Aaronson on being a rationalist, like me!

There was a sport for me in high school, I just didn’t know it. Bill Clinton and James Patternson are writing a political thriller?! And it’s not their first!? And it has a funny “About the author” where his presidency isn’t mentioned till near the end. The broad arrow. Looking for good gelato.

World Tour

I’ve been thinking a lot about the world lately, and all I haven’t seen. I think if the world is about to drastically change really fast (and really soon) then I would like to know and understand as much of it as I can.

In historical tradition, I’ll go East to West (ish).

Japan

What can I say about Japan that hasn’t already been said? They’re the only country that that would have an earnest life-sized cardboard cutout of the late pope at a Catholic museum (found in Nagasaki). People are always complaining about tourists, but Kyushu during Golden Week was just Japanese tourists, who are adorable if not too excited about what is an aggressively mediocre Chinatown.

Spoiler! Alert. If You Think There Is Any Chance You Will Be In Japan From Now Till October 13th, Stop Reading!

Absolutely wild to me that not more people know about the World Expo. It’s like the World’s Fair but it’s still happening right now! In a hilarious fall from grace, the spectacle that brought us the likes of the Eiffel Tower and air conditioning, is now full of ESG slop and AI art. There are a lot of reasons for this fall, which can all be summed up as “the world got too big and things had to specialise”.

I took the USA Pavillion tour in Chinese because the line was shorter, and the tour guide made a great deal of delineating the mainland from Taiwan. The exhibit advertised the Fulbright scholarship prominently (rip), and documented every single American space flight (including Blue Origin) except SpaceX. Staffed by State Department interns, it was a true Biden capsule. Taiwan did not get a traditional country pavillion, rather they had a spot outside the ring called Tech World, right next to the Gas Pavillion which kept talking about a mysterious “e-methane”. Countries like Turkmenistan and Uzbekistan brought incredible production value, and upon looking up their GDP I learned they must be dictatorships. All the architecture there is stunning. The Korea pavillion had a huge screen (very conservatively at least 15x30 feet) that played 2016 GANs on loop. There is a cute mascot. The China pavillion had a sanctioned GPU and a section on “friendship” with the Japanese. There was a drone show and water show. Hat tip to Indonesia for having an earnest pavillion, but the other ones were really funny. I would’ve spent at least two full days there if I budgeted for it.

New Hebrides

In a surprising libertarian twist, a condominium colony in Oceania?! They had separate post offices and police forces for the French and British, with equal administrators on each side. Each administration had its own laws, currencies, immigration policies, and even had regulatory arbitrage. I asked people who would break a tie in case of a contention. Some answered the natives because they don’t understand hows colonialism works. My guess was the pope because I forgot about Martin Luther (sorry!). Idiots! Clearly, the King of Spain is the tie breaker.

China

Beijing Silvermine, an archive of negatives from 1985 - 2005. Some truly epic photos in here. Makes me nostalgic somehow, maybe the ancestral memory is in me somewhere. The Chinese Periodic Table.

Do not write madlibs of ancient philosophical and religious texts! I don’t want to claim anything on the grounds of disrespect or offense here (I don’t even like Taoism), just that it’s ugly and crude. That website is the 道德经 with word substitution for it to be about code in ways that don’t make sense. The referenced translator in that production, Stephen Mitchell, claims to have also translated the Bhagavad Gita, the Iliad, and Job which is a clear red flag even if you don’t notice that his translation is in large parts fabrication. Even fanfiction — a far higher form of art — of ancient texts has rarely been an aesthetically successful endeavour. The best is likely the Book of Mormon, which wins by a landslide unless you count the New Testament. If you want to think about Taoism, check out this site.

Korea

Definitely learn hangul which is the most incredible alphabet (there was a time where they almost incorporated traditional Chinese characters, thank goodness we are not in that timeline). They have some fairly unusual politics as well which you can read about on the Blue Roof website. When you visit, you can see a lot of downstream effects of the Chaebols, the War and their economic boom. Also scroll down on the Kimchi page, it goes crazy. The Chinese love to claim that anything that happened in Asia was them (except the things that they really don’t want to have been them).

On the margin, Korea is probably the top country I’d recommend visiting. It’s beautiful, cheap, and full of unique culture that is not going to last long as either the ASI or birth rates will catch up to them. They have Chinese maximalism (artificial waterfalls, spas with waterslides), lack of western software, insane drinking culture and delicious food. There’s a lot of “insert tokens for utils” sort of energy. Their subway stations are stocked with modest numbers of gas masks and every city block has something about Dokdo and Jeju (two important islands for very different reasons, but in a way, the same reason).

USSR

The second-most exciting experience I had on a recent trip to Vegas was talking to a Yugoslavian Lyft driver who I stayed in the car chatting to after arriving at the destination. There’s very little that is like the connection an emigrant has to their now non-existent country. It’s unlikely I’ll ever see the fall of a nation on the scale of the USSR.

The Jewish Autonomous Oblast, highly recommend listening to the anthem. Reminds me of the King of Spain thing where it sounds ridiculous but then you think a little and it starts to make a lot of sense. The Soviets referred to the Jews as “rootless cosmopolitans” in a historical slur, though I’m told “rootless” is somewhat of a bad translation and it’s more like “without people”.

Unfortunately thanks to their misfortune, the Soviets produced a lot of incredible art. Crime and Punishment is on my nightstand and I’m making good progress, and the Idiot remains in mind as I vividly recall a friend informing me that it is about me.

I watched Stalker (my first Tarkovsky film), it was good? I was certainly engaged, affected even. It’s a difficult watch, I don’t know if I have anything to say. I’ve been told I’m more of a Solaris girl. I watched the Hedgehog in the Fog at least six times this month. Miyazaki loved it, but it also so perfectly captures the Soviet mindset. I’m extremely excited to watch Tale of Tales (but saving it) and the Overcoat, which has been in production for uh, 40 years now. I also watched a bit of Cheburashka and it’s pretty good good. I watched the entirety of Vinni Pukh of which there is like 30 minutes total and it’s way better than “Winnie the Pooh” though maybe people who grew up with American cartoons will disagree.

Polish Haitians

America, United States of

Man elevates to 4,900 meters in a lawnchair, “Ah, the difficulty is, ah, this was an unauthorized balloon launch, and, uh, I know I'm in a federal airspace, and, uh, I'm sure my ground crew has alerted the proper authority. But, uh, just call them and tell them I'm okay” he says. Why don’t we use pleasure as an interrogation technique? The Mayflower is underrated as a boat. If there was one thing I could go back in time to see in America, it would be the Civil Rights movement. It was an incredible situation of humanity, politics, heroism and leadership. What better year than 2025 to get into the Civil Rights movement?

I’m always saying that musical theatre is The White American Art. I got the chance to see a local theatre put on Pacific Overtures, a Sondheim musical about the beginning of the Meiji Restoration. “Please Hello!” is likely Sondheim’s lyrical masterpiece, and the show was artfully problematic in every single way all at once (though is straightforwardly pro-colonization).

Ben Kuhn on running major projects. Nadia on doing the Jhanas, highly recommended technical read about a particular kind of meditation. I don’t necessarily endorse it, but I am going on a Jhourney retreat in July. Matt Godbolt (of the famed compiler explorer!) wrote a really good characteristic post about Claude usage.

OpenAI is making big swings. Strong piece of writing; Superintelligence Strategy from Hendrycks and Alexandr Wang. I consider this to be in the same category as the Gradual Disempowerment post I read earlier this year, and to a lesser extent AI 2027, as a piece of writing that attempts to capture predictions of ASI outcomes now that we’re in the final few years and doing that is quite tractable now. I left out Eric Schmidt from that author list intentionally. I’ve read and skimmed a number of works with his name on it, and I’m fairly confident “Eric Schmidt” is more or less a pen name. If you want a good example of this, check out “The Age of AI: And Our Human Future”, which I have a strong suspicion was written by an eighteen-year-old. Also a good time to look at some of Kissinger’s work, much of which is worth reading and very likely written by Kissinger.

I have an upcoming very short trip to Türkiye, which of course means I need to read up! Korean politics (recent trip I took) and history is pretty wild, and I got a lot of mileage out of that but I suspect Türkiye might hit harder. The PKK and Öcalan are must-knows, with the weirdest Öcalan situation being his at least 1,000 guards as the lone prisoner on an island. Former Turkish PM Claims Gaza Still Part of Ottoman Empire, Gazans Turkish Citizens. Türkiye is in many ways, the center of the world after all. San Francisco’s long-dead Suicide Club.

Fraternal birth order and male sexual orientation is a mildly interesting article, mostly because when you combine those statistics with number of children per mother, you would expect there to have been at least 10% more gay men 200 years ago. Band of Thebes anyone? The Rippling Deel funny story. Ancient warfare lawnmower. What if we nuked the moon, for morale?

There is finally a new pass at understanding what ASI misalignment or misuse looks like in detail. It’s probably extremely hard to make predictions around “this is how it’ll go”, but I think there’s a lot of clear predictions you can make in the shape of “given this plausible outcome of a particular variable, this is how it’ll go from there”. The Anthropic Economic Index serves a similar goal in providing bits that prepare us for different futures. I’m way less worried about how ASI will fit into our economy than our governance though. Anthropic jailbreaking paper — it works! You can just stop jailbreaking!

Good time to learn about the original Thinking Machines which seemed utterly incredible. The MiniMax attention paper. Another great eval (Humanity’s Last Exam), though the name is far too ambitiously great. I’m pleased with calibration reporting, though I asked for a bit more. OpenAI comes forth with computer use. The frontier math drama was, in my opinion, not a huge deal in of itself (especially after considering all the nuances) but is pretty strong foreshadowing about what’s to come. A simple sama post, with an extremely cute last sentence — but I’m still waiting to someone come forth with a strong pitch for neo-georgism in an ASI regime and it seems like sama might miss that train despite all the setup he’s done. OpenAI releases a lot of good content about evaluating on contest programming. Maybe now’s a good time to share that dream I had two+ years ago where the only way to defeat the unaligned AGI was to beat it at Codeforces and so I started a camp in the Canadian forest where we had to be awake at weird times to do the Russian contests?

I’ve always noticed a ton of evidence that people these days in many way aren’t as competent, agentic or even as emotionally healthy as they used to be. I’ve never attributed it to the way we raise children, but this essay was poignant on the matter. Weird that an American political faction can win being anti-child sexual exploitation by digging up huge news from a decade ago, from across the world. The Rotherham child sexual exploitation was uniquely horrific though. On a brighter note of things Indo-Aryans did that we didn’t pay enough attention to is Kumbh Mela 2025. At least 400 million attendees over 45 days, with an estimated 7M present on any given day. I think it’s an insane, sordid and beautiful human feat. It’s a super special place, with drone shows, an android app, 40,000 police, 150,000 toilets, 2,300 cameras, and underwater drones? Even though I think religious events likely have much lower rates of crowdrush, they definitely cover up deaths with only 30 reported. Lauren Powell Jobs went but did not bathe for health reasons (though supposedly, the water Does Not contain feces). I rant about it here because of how underreported it is.

Jane Street wrote too little on Dune, their build system! It used to be my dream to work on build systems and performance of devtools, and is still sort of beautiful to think about. I will definitely be rewriting Bazel recreationally after the singularity. An analysis on strategy and decisiveness in organisations.

Why our butts jut out. I learn about why I never want to try to buy a house. Reminder that UA Flight 93 happened and should be a huge part of the terrorism-zeitgeist. Year-old Matt Levine has a really good bit about AI researchers that aged well. Really impressed with how accurately Matty can clock the way our industry works from the outside. The Hiroshima Maidens happened? They just sent disfigured women to America for plastic surgery and press (not in that order), with the end results being pretty good and one accusation of doing human experiments (haha). I learned that one of the campaigns into China was called Operation Ichi-Go which is not militarily interesting it’s just funny that it sounds like “strawberry”. EGG PAPER! Which for some reason is in the Communications Engineering section of Nature! 🥚

DigitalOcean, never change. OpenAI had a good, open post-mortem for an outage caused by Kube API overload. Jane Street continues to win on culture. Debugging stories; 500-mile email limit and the magic position. I’ve always complained about the inability of AIs to write well, AI ensloppification is a weaker claim but Gwern makes it, which is nice. Claude on being a simulator. OpenAI published alignment research which isn’t particularly exciting but so little research is published that it’s worth a read.

Extremely long, sometimes really opinionated but still fun-and-informative post about Nigeria. I was thinking about coups because of the wildly incompetent Korean coup. How do successful coups look? 2021 in Myanmar and Sudan. 2009 in Honduras which barely counts it was more of a Napoleon situation. 2013 in Egypt. Also all the Nigerian coups which seem like they were all led by this guy Abacha. I followed that into reading about the Igbo Jews, confusingly not recognized by the Israeli Supreme Court despite immigration to Israel, plausible historical ties and circumcision practice. The Bnei Menashe (India) did officially get declared one of the lost tribes of Israel, which they figured out due to Christian missionaries.

The Seventh-Day Adventists are protestants who do Saturday, and there are supposedly more of them than there are Jews and they funded a bunch of veganism and cereal stuff which involved a sanitarium? Poster House in NYC is a great museum, I really liked the Lester Beall works (too young to be a New Dealer). A snapshot into Chinese slang and culture.

Cryptic Crosswords

I did the last two weeks of London Times Quick Cryptics! These crosswords are really really beautiful to do because;

• the exploration space is really high

• unlike concise crosswords, there’s only one correct answer and you can tell when you have the correct answer

• you can solve one clue collaboratively, ie someone can parse the wordplay and someone else can do the definition, or you can even have someone parse it correctly and the other person actually solve it

• anagrams are really fun

• you can know you have the correct answer without understanding why

• arcane Britishism

There are much gentler cryptics at minutecryptic, or the New York(er| Times) ones or even Guardian, but the Times serves as a particularly violent place. I hope to be able to do the Quick Cryptics in under half an hour, and maybe eventually get to some regular size Times Cryptics. The Times also publishes a Mephisto (extra hard crossie) which is so arcance that Claude assumes that it has simply stopped being.

It’s also really funny to watch the Times report on American affairs, and to visit the crossword blogs which seem mostly full of sixty year old British men who do their crosswords on paper but log onto a computer to talk to people about it.

100M context windows, someone please make use of it. sama will never top the American equity fund, but at least the website for the intelligence age is still good. Best tech drama lately? Definitely Automattic. See also this and well, the orange site.

SSC vintage on reversing advice which I quite like the underlying model for. Probably I should write some text that is advice on taking advice. We love Baudrillard, and we love Baudrillard gloves! This years Lesswrong Petrov day is iconic as usual. People are still stealing and doing a good job at it but are scummy? Tony Soprano was right after all I suppose. There’s an Italian fugitive who like definitely faked his death in what seems like a wild story? His name is Samuele Landi, he was a seasteader who supposedly passed in February (this article seems quite biased but other ones are paywalled) though his body does not seem properly identified. He also has ties to Liberland and some Saudi foresting thing. There’s an upcoming documentary too, so maybe we’ll learn.

I read about the Bayh-Dole Act because of certain non-profit. People used to just do things like go to the arctic in a balloon which is kind of awesome despite the unfortunate result, but some things are just bad? Like trying to aim missiles by putting a pigeon inside of it. Past few weeks I’ve just been generally frustrated with people being excited to to unusual and strange things but forgetting that while it’s true that being unusual is the only way to be better, there are way more ways to be worse than better. African Americans responsible for Coca-Cola not containing cocaine? Rare coke flavour (new!) Marriott hotels and their LDS origins, specifically about how they decided to do alcohol. Your lymphatic system does do your brain, how did we not know this? Chinese shoegaze music.

Cute article about an SF panhandler, like seriously cute. I went to the Bronx Zoo and read up on about it ahead of time. The otter-murder was cute and fun but this was harrowing. NYPD is notorious for parking in bike lanes, and it seems like they probably got ticketed a lot before the traffic agents became a part of the NYPD in 1996. 10e9 dollars is the annual budget of the NYPD, a figure that has been steadily growing. In comparison, North Korea is estimated to spend about 4e9 dollars a year. The NYPD did millions of dollars of ticketing legally parked cars. New Karl Marx translation!

I finally read some Lovecraft, and it makes so much sense how the concepts are so memetically fit now. If we erased all historical and cultural context in the world, Lovecraft would still a classic.

Book: Path to Power and Means of Ascent (first two LBJ Caro books)

Earlier this year I visited New York to see the Columbia protests, after which I went to a bookstore to pick up a book on Nixon (when in Rome). Expectedly I got stressed out about which Nixon biography I would enjoy most and figured I’d accept the off-by-one-presidency error and get LBJ instead — where I was certain I wanted the Caro. I tried to purchase Master of the Senate, the third book in the series which documents Johnson’s senatorship. The bookstore clerk proceeded to ask me whether I had read the first two books and talked me into getting the first. Big Biography wins again!

LBJ is unfortunately a large part of my personality now, the books are phenomenal. They don’t make biographers like Caro anymore. His research involved moving to Johnson’s hometown and his interviews are so skillfully conducting, not to even mention the prose itself. I adore and admire Caro, but LBJ is a horrible person? (I hear there’s a plot-twist at the end, but no spoilers please!) In real life, utilitarianism and consequentialism are usually the same thing, but not for LBJ. He arguably did net good, but most of the bad things he did were not necessary to achieve those positive outcomes. In general being into Caro is a green flag and being into LBJ (or Moses) is a red flag.

The Power Broker is the more popular series, but it’s kind of an East Coast story about an idealist who is polarised by the system and takes it over. LBJ is a West Coast story. A man born believing that power was his birthright and dedicates his life to acquiring it through a blatant lack of regard for convention. A narrative where one can acquire all the power and then figure out something to do with it later. He is the worst of both gender stereotypes / tendencies. Manipulative, power-hungry, short-tempered. Insecure, bipolar 2, daddy issues, tummy hurt. From LBJ’s childhood up to his second senate campaign he exhibits relatively little character development. LBJ is so compelling as a human, in large part due to Caro’s writing.

I had moments of doubt. Caro goes on these side quests, and in between the lines he’s saying “trust me, you’ll understand why I’m telling you about this later”. The first time was in the opening fifty pages, which all took place well before Lyndon was even born. My favourite parts are likely the mini-biographies of Ladybird, Rayburn and Stevenson. Caro also devotes a chapter to describing the pains of laundry and other household chores to provide context for what electricity meant. I do sometimes worry that Caro takes on more narrative license than I’d prefer when reading about a morally controversial figure, but his writing is so effective at bringing you into the universe and fully grokking even the smallest of situations.

Book: Lolita

I turned away from fiction years ago because I thought it was offensive to reality, and this year I’ve three-sixtied into reading fiction escape reality. Lolita is doubly inconsequential as it’s not only fiction, but fiction written by an unreliable narrator.

“Oh Lolita, it’s about you” was the recommendation I got for it while scanning a friend’s bookshelf. Sure, I guess. The plot wasn’t the most special part of the book though, it’s the fact that Nabakov wrote it in a non-native language — a fact that I thankfully encountered in the first quarter of the text. Nabakov did a Russian translation too, which he claimed to be worse despite him learning English well after his where any pedagogist would consider optimal (and of course also wrote in French). My favourite part of his writing style was constant original metaphors without invoking cringe or deferring to reader interpretation.

Objectively speaking, the plot is unfortunately probably the notable feature of Lolita. It reminded me a lot of this book that contained a segment titled The Young-Girl As A Commodity. "And they are realistic, even in love” (Tiqqun, 2012). The controversy is in the hebephilia, the commentary is in the specific kind of sexualisation of young women we do, especially in the western world. Most characteristic was how Lo could be bought off so her love (or at least, good graces) could be retained.

As a girl, every song is about me! The way HH uses his intelligence to so seamlessly rationalise something barbaric is something I’m as familiar with as one can be. It’s very popular in my locale (San Francisco) and culture (rationality). People say this shit about this all the time but it’s quite special to have to captured so beautifully in fiction, since nothing teaches this lesson like carefully watching someone else make the mistake.

vitalik on techno-optimism, he’s amazing and this is well reasoned and written but also i am mildly annoyed that everyone stopped to look at it when nothing in it is that novel. why you’ve never been in a plane crash. there’s this, which i obsessed about for days — i don’t want to say anything more because i couldn’t possibly do it justice. incredible combination of funny, beautiful and educational. the vesuvius prize has winners but also future plans.

the openai press release about a lawsuit, which is weirdly entertaining? perhaps a bias of my occupation. sleeper agents paper! i’ve been waiting for that one to drop for a while.

short story that really straddles eerie and cute. this excerpt from “surely you’re joking mr. feynman” (brilliant scientist makes pithy quote). want insight? check out this article by ricky about how mistakes are made, particularly how multi-billion dollar crypto exchange mistakes (and their surrounding mistakes, because with one mistake always comes another) occur. coconut deaths! my friend with the most insane journalist name wrote a great article about ai. a scientific cure for suffering? solution for wireheading? a 500M typo from my favourite federal organisation.

i only studied one paper meaningfully, but was mamba, so that’s fine. here’s a mamba accessory!

it’s elon season? it started with Elon Musk is Misunderstood, then there was Elon Musk is Overpaid (and the rest of the ongoing matt levine saga). the handmer-levine pairing (unrelated: the mars helicopter) was incredible. i’m so sick of “if you liked x, try y”! i’m now into things to enjoy simultaneously. 1984 x brave new world. atlas shrugged x progress and poverty. making of the atomic bomb x american prometheus.

Room for Thought

i am a fan of the boot boyz biz, and the douglas englebart brick pencil shirt i have from them. problem: their apparel has becoming increasingly text-dense over the years. solution: a four-hundred page philosophy picture book. it’s beautiful and thought provoking and overall an enjoyable read. my favourite quote — though much of the bits were contained in the visual — was: There Will Be People Who Will Say “You Don’t Mix This with That.” And You Will Say: “Watch Me (Watch Me)”.

imagine everything one might enjoy about a naval tweet and manifest it into full-thoughts represented in beautiful imagery and typography. this is it!

More qntm shorts and The Metamorphisis of Prime Intellect

for my entire life i didn’t read scifi out a notion that it’s disrespectful to the beauty of the science of our universe, either existing or prospective (that’s a lie, it’s because it’d take up too much time). i don’t believe that less at all, but alas i was bullied into reading scifi, and expectedly enjoyed it.

it started with shorts — my last blog post had a few qntm, i read five more this year. honestly some were underwhelming but not boring because they were making me think a lot. some (some) were quant, as if the slice-of-life equivalent of scifi. some were creative and exciting in a traditional way. i loved them all.

metamorphisis of prime intellect was technically my first full-feature and it is perfect. it does an incredible job capturing all the nuances of immortality and how human desires would respond to it without wasting time on things i’ve already thought about (cries in the first two thirds of permutation city). the world it created was compelling and the pornography was nothing but the finest.

Another good data paper, though good data papers don’t need to have big insights or takeaways. The best Gwern posts are ones where I can’t tell if he’s joking, like this one on doing a giant MLP for everything. A paper on baking in CoT processes into the model which I generally like because there’s no reason for number of tokens to correlate with how much compute it’s needed to do something. . Um. A way to test if models were trained on the test set, but requires knowing the exact strings and so isn’t that exciting. Using temperature scaling for calibration of DNNs. Understanding KL Divergence a bit better though I’ll probably never need to think about that again. Trustworthy paper on Gemini evals.

The x.ai grok announcement, which is impressive though they evaluated other models at weird temperatures. Gemini report dropped! They generally did good, though I’m annoyed by eval milking even though it’s not unfair. New Yorker profile that mentions a researcher I respect but is mostly about the model. I want to do profiles on all my favourite people but it would take time and embarrass them I guess. Incredible tweet thread about Chinese nicknames for semi-conductor manufacturing companies.

Read this book review for a book I have no intention of reading ever. I didn’t really need the financial advice, but I generally felt like exploring the documents of the great Mr. Money Moustache. The US government and their 7,500 watercolour fruits. It’s awesome that the Church of Jesus Christ of Latter Day Saints was the first website that showed me results when I was googling about this bok choy. Ant eusocial behaviour does have nanobot swarm energy. Turns out Wikipedia edit histories are a great source of funny material, I was refreshing this page on a certain news page and there are a lot of good commit messages in there. The Scientologists had a weird jazz-y band. A branch of Christianity named after milk because they drank milk at lent. Another weird guy to remind you that serial extreme lying is really common and not that hard. I should make a list of these guys for whenever someone tries to dismiss the possibility of a pathological liar.

A review on earning to give. Tom O’Donnell’s writing titled Libertarian Police Department. A thorough intro to the pharma industry that is really well-written even if you don’t care about pharma. The whole regulation thing is sort of new, and makes me skeptical of the whole “governments can’t react quickly to big technological developments” notion — it’s more like governments probably won’t react well. The better than Beatle’s effect is fun, and it’s generally nice as an AI risk-worrier to see that it’s possible to be damagingly risk averse. Have you heard about the tooth sauce that makes you never get cavities again? I’m still waiting for them to come out with an N/M dentists hate this stat, more info here. A heartfelt shutdown of omegle, morality is hard these days.

The best fucking post-mortem I’ve ever read about the Cloudflare outage. I get a bit of schadenfreude for outages just because I get excited about the post-mortem. This one was special because it was an enjoyable read, but I didn’t quite appreciate the attitude of blame to their data center manager (who is definitely getting fired) especially when they missed some good practices and were out for far longer after the data centers were back up. It reminded me to reread this old classic. I finally looked up the origin of the term “cargo-cult” and it’s really fucked, that term will never be the same to me. Imagine someone has written a hit-piece about your company which consists entirely of quotes your engineers said in the company’s engineering podcast. A cute post about forming emotional attachment with your servers.

Matt Levine really hit the mark on the whole OpenAI situation, my favourite was this day’s.

Reread Lena, I still don’t get creeped out by it. If You Are Reading This is incredibly utterly relatable. Another qntm that starts out really cute and takes a morbid but funny turn. The Difference did manage to creep me out, as did Gorge — those two are definitely my favourite of the listed qntm.

Partial Book Review: History of Western Philosophy by Bertrand Russell

I started reading this book like six months ago and am only just halfway done because I’m bad at reading. The first two hundred and seventeen pages are about Ancient Philosophy and the following two hundred and seventy pages are about Catholic Philosophy, so I thought I’d write some notes before venturing into the three hundred and forty six pages of Modern Philosophy. Russell is a fantastic philosopher, really smart and seems underrated. He inserts opinion in isolated spots, and I at least always appreciated them. I don’t really need to know anything about ancient or catholic philosophy, but here are some of the things my brain retained (which are only a little correlated with how interesting they are, but very correlated with how recently I read them):

• Some great Catholic philosophers (like Occam) who did good thinking were perverse in that they weren’t in some truth-seeking regime, but rationalising existing beliefs. There were some goofy things, like Roger Bacon saying the first cause of ignorance is bad authority but specifying that it does not include the church. This seems rarer in this day in age where truth-seeking is more common and virtuous but I wonder where it’s happening that I’m not seeing? I feel like some research is locally like this in a good way.

• There’s a funny event in the 14th century where the French lost the papacy when there were two popes declared (one Roman and one Avignon). The funny part is that it was resolved by just… asking a council? Cultures of absolute chain of command are good like this I guess. Russell writes about this with “therefore a power superior to a legitimate pope had to be found” and described the Avignon pope (who was closely associated with the French monarch) as “addicted to favouritism and nepotism”. It’s a surprisingly funny book.

• Saint Augustine semi-anticipated the cogito ergo sum, he also believed that it was morally important to live in solitude and did so for a number of years I don’t remember

• In the middle of all the boring Catholic philosophy stuff, Russell drops this, which I don’t endorse as fully accurate but is entertaining:
  

  Yahweh=Dialectical Materialism
  

  The Messiah=Marx
  

  The Elect=The Proletariat
  

  The Church=The Communist Party
  

  The Second Coming=The Revolution
  

  Hell=Punishment of the Capitalists
  

  The Millennium=The Communist Commonwealth
  

• There was joke about how the Roman army figured out they could take bribes when choosing emperor and then assassinate the emperor and repeat

• Aristotle is really pro-slavery, and has some weird reasoning about how some people are better off when ruled by superiors (like animals) and it justifies every single war as morally good. I only remember this because there was a funny comment from Russell: “Very satisfactory!”

• Reading through the Ancient philosophy part was interesting as it reminds me how much standard logic and intuition I take as common sense had to be derived at one point. Simple things like how some opinions can be better than others even if you don’t know that they’re true or how knowledge is useful for morality

• A lot of the content was context for how the works of specific philosophers were passed down, and may have been poorly documented by followers of the philosophers.

Book Review: A Very Short Introduction to Hegel by Peter Singer

I found this book at the store and I only picked it up because I saw “Peter Singer” and I was like no way! Not The Most Good You Can Do Peter Singer? It was! I was biased against the book because Hegel feels like a meme and the book opens with Singer saying that Hegel was the most “impactful” philosopher of the 20th century. Of course I thought to myself “Marx erasure?” but Singer had thought of it and followed it up with something about how because Hegel influenced Marx it doesn’t count or something.

I have never… read Hegel, but the only coherent takeaway I could pull was in regards to Hegel’s philosophy of history. The Hegelian notion of freedom actually really clicked with me, it’s what we today call “agency” (even though we use it in the definition of it’s original use) and he describes it with things like how loyalty to your nation isn’t freedom because it’s too much of a default, convenient belief. The history part is uncompelling though. Hegel describes the development of history as gain of freedom, which is true in uninteresting ways and he fails to make a coherent claim about what the end state is and claims that his 19th century Germany was it.

There was a bunch of stuff that didn’t seem fully cogent though I understand bits and pieces. Probably there is more good stuff in there? Though it seems sufficiently hard to study that I will simply not.

I have a lot of priors against Hegel. To start, Hegel is notoriously hard to understand (countless of introductory attempts) and background says I should worry that it’s hard to understand because there isn’t much to understand. Also a lot of what is attributed to Hegel is notes from his students. More bias against Hegel is that I tried to read Zizek’s jokes and he has horrible jokes, continential humour just doesn’t click with me.

Read more

Article explaining the Llama.cpp performance, though I find myself asking “how is it so slow” infinitely more than “how is it so fast”. I asked myself “how is it so slow” with the 3.5-turbo size-leak(?), and I probably would’ve asked that even for a 200B param dense model. The GPT-V system-card is kind of exciting, I can’t wait to move out of the chat modality. A study on training stability (and a surprising publish from Google!). Data comp work from the people who did Red Pajama. Cool multi-modal architecture from Adept. Using sink tokens to put the attention somewhere when it’s not using the most recent window of tokens.

Incredible Hillel Wayne on JavaScript history, I really want to write the one about why there are three JS engines that were all developed at around the same time. They’ve read the first letters of the Herculaneum scrolls! 2015 ML wins again! I caught up on Rails World, which also introduces the “renaissance developer”. Everyone definitely needs to spend time exploring the catalog of the Computer History Museum — I found a “palm pilop” shirt while I was searching for Research in Motion items.

I miss the peak of the housing discourse, this 2018 post was a fantastic throwback. Also a throwback, 2021 Sam Altman was fucking amazing, I hope he still believes those things. I finally asked myself why people don’t trade on Hollywood, and uh, it’s because it’s illegal (along with onion futures). Apparently it was shot down (by lobbyists) because it would be too easily manipulated and wouldn’t help film companies, but there’s gotta be something more right?

There’s this activity called “high-pointing”, which is for people who like mountaineering and checklists. My favourite high point should maybe be the side of the road on Florida but the best read is definitely the Nunavut high point. Maybe my favourite high point is the Saudi Arabia one because it comes with this great story about how it was corrected in 2018 with a GPS by high-pointers. Also in eldritch physical feats, there’s Everesting. We have CNCed lace braiding (lace is more advanced than knitting by centuries) now! How to make temporary Lichtenberg figures at “home”. The University of Minnesota is phenomenal at apples, the UWashington ones are much worse and you can really tell that the Cosmic Crisp was their attempt at a premium apple but it’s kind of mealy and the skin is too thick. I am so excited for Triumph!! A cute lil newsletter with good items — I bought a D-BROS vase off ebay. There are so many pasta shapes.

Check out franciscosan.org !

An attempt at a new architecture, but immediately opens with a plot showing they couldn’t scale their baseline transformer properly, an inspirational quote and an impossible triangle that is used as a diagram? A good explanation for why GPT-4 is so non-deterministic. Testings of think tokens (filler text) to see if it improves performance, and the answer is no except maybe for GPT-4?

Three year old text on administration markets, which references a much more historically loaded piece on the decline of administrations. Excuse the loading time from webarchive, but this content about Uber, Maoism, and Georgism. Muji Manifesto, “this will do”.

TinyStories! I can’t believe the same organisation put out both this paper and the Longformer paper. The TinyStories doesn’t focus on what should be the primary application of its technique, which is of course mech interp. The autointerpretability work was quite nifty, OpenAI alignment lives on after all (I haven’t read it yet, but DeepMind interp is also alive!) I am a big fan of line of research around measuring faithfulness in chain-of-thought reasoning, which includes checks like filler text (think tokens), early answering, adding mistakes and the always-choose-a trick from this other paper. That paper was published with the decomposition paper, which does two kinds of decomp and finds that it on average decreases performance (especially the factored-decomp) but increases faithfulness. nostalgebraist couldn’t find as much sycophancy in GPT models, my best guess is that this is a difference in RL data. Positional embeddings are a helix, but also I still don’t get how they can be important.

Paper doing a decent job outlining how one would manage AI public safety risks, though details feel a bit messy. I brought my rabbit, bundle, to the office for his second New York Times photoshoot but sadly the bunny didn’t support the AI doomerism narrative. India’s excerpt on a guide to guided missiles. The frontier model forum launched, lower cased (fingers crossed) until it goes well. Anthropic publishes a bit about frontier threats red-teaming, a bit because it’s mostly focused on biorisk. Good luck, everyone! Paul coming out swinging with a very compelling pitch on sharing LLM capabilities, with a subheader that accelerating agents in particular is neutral.

The highlight of this months Hillel Wayne is in favour of defenses of design. I read a bit of SemiAnalysis this month too (Gemini and TPU shenanigans, sigh), and have shamefully declared Matt Levine bankruptcy and only followed for the Sculptor Saga. Spencer Greenberg published this month on smart people with dumb beliefs, for which one must also see this video of Feynman singing about orange juice to make fun of Pauling’s vitamin C phase. Man being an olympic host basically turns your city into a charter city for a few years? They’re spending a billion dollars to clean the Seine. I learned about the wood-bending techniques for the very expensive Artek Aalto stools, and I’m into it.

Tree of Thought paper, which was well done and nifty though I don’t think the premise will be fruitful. A pretty good shot at replacements for tokenizers with (mega)byte-level models. Thorough data paper with really awesome figures. Gwern comment on mode collapse with CAI

PaLM2’s tech report, which doesn’t contain the unfortunate leak that it was 3x undertrained. I’m kind of glad to see tokenizer work happening. Multilingual chain-of-thought. Evaluating bilingualism in PaLM. Loose thinking about data mixes. Thoughts of designing progress for AGI. DeepMind optimises their lil mov instructions. I reread the GLM paper in desperation of inspiration for multilingual. Another paper towards “data is all you need”. Inflection’s perfectly typeset tech “memo” on their LMs.

The Blackberry movie was the best thing I’ve ever seen, it so perfectly captured how startup founders go insane and start believing deranged things about their company. Also had an iconic line where when Jim asked Mike something about “i thought these were the best engineers in the world” and the quote from the movie was “I didn’t say they were the best engineers in the world, I said they were the best engineers in Canada.” Anyway, I combed through the Blackberry wikipedia page after that to keep my facts straight. Unrelated wikipedia pages: model organisms, the aquatic ape hypothesis and Copy Exactly! Design for more reasonable dataframes in python.

Investing in pandemic prevention. Learning about communal baths, from the archives of boot boyz. Reddit still comes through with this guy who commuted to UC Berkeley from LA. I attempted to read a few of Zizek’s Jokes, it was the worst fucking book. Short story about morality in the grand schemes: The ants and the grasshopper. Manifold Markets leaked the AI extinction letter, and purposely has it be this way and I like it? My leaving Cohere was also leaked by Manifold lol. Some fairly safe forecasts on what GPT 2030 will look like from Jacob Steindhart. Devon Zuegel on why all the good urbanism in the US is in the south. A really cute short story about dogs that made me cry. Agnes Callard’s profile in the New Yorker is so based, I love her. Do you want a living room that looks exactly like mine? Check out the guide to rationalist interior decorating.

Book: Rats

Upon moving to Sat Francisco, I read Cool Gray City of Love and immediately wished for an equivalent for New York (I lived there on and off). It would be hard to do as a book about New York, as the cuteness of CGCL came from the fact that most of the chapters weren’t about the Gold Rush or Summer of Love but about local history that could only be engaging to someone sitting on top of it. Maybe a book about a specific neighbourhood would cut it, but how could I choose a neighbourhood?

I found Rats (Robert Sullivan) at a Battery Park City housewarming in September where the book was not just recommended to me, but physically handed to me (which is the ultimate purpose of keeping one’s books collection). I lived in the Robert Moses part of town (Alphabet City) in May and begrudgingly thought about giving the Power Broker an attempt but then realised now was definitely the time to pick up Rats — only to realise I had left it in San Francisco. Thankfully I returned in June and finished the book in a week. It lived up to my CGCL standard.

Sullivan and I disagree, in that I rather like rats, but the book could've surely had me mistaken. He never quite explains how someone who wouldn’t even fancy a fancy rat would spend a year researching and watching them — even traveling to Chicago to meet an exterminator celebrity.

The book pops back in forth between the present of Sullivans rat adventures and history involving rats ranging from that Manhattan dentist who tried to shoot the rats on Rikers to the role of rats in plague over centuries. It covers the animal rights activists who tried to stop rat fights and the sanitation worker strike of NYC. Rats ends with a beautiful memo to the type it uses and has the greatest cover — look it up!

Derivative of top-p (nucleus) sampling, with really good theoretical reasoning as to why this should work better. GPT-4 Technical Report, amazing author list format and impressive results and safety measures. A clean and nifty paper studying memorization in llms, with a notable finding that some things are memorized “right away”. Sander doing a really good explanation of diffusion models as autoencoders which definitely went over my head. Gist Tokens paper, which introduces a great technique to compress a prompt into a single token via masking (it’s so good I read it four times). Transformer Math 101.

Yay, a paper on dataset selection, but it relies on having a target distribution to mimic. Paper on poisoning datasets on the web, though it makes a strange assumption that people don’t keep their crawls and doesn’t make a significant impact. Data mixes in relation to in-context learning, though the paper is explicitly wrong about a few things. Code language mix in relation to performance, comes with some good eval and good but not super impactful results. I really enjoyed that paper because it calls Rust a “low resource” language lol.

Overview of Bing Chat mishaps. Janus on cyborgism, which was impressive in that I thought we were at the boundaries of theoretical safety work. Tangential, the hot mess theory of AI misalignment, which is well-formed though I disagree with it. Anthropic research on moral self-correction came out! Turns out too many RLHF steps actually result in discriminating in favour of minorities. OpenAI on how AI Systems should behave.

Reviewed Universal Distribution Absolute Self-Selection Assumption, from yashkaf and Paul Christiano and it's still not super deep in my head. Ben Kuhn goes to Anthropic, bringing with him tips for new managers. I love to be reminded that diet restriction for climate and welfare are inversely correlated and shrimp suffer.

The History of the Sears Catalog, I also read the catalog of course. Highlight for me was that the low-tier appliance line was called "challenge" as in "challenge toaster", "challenge waffle iron", etc. A surprisingly coherent Boot Boyz Biz article on oddly shaped buildings. BBB cards I failed to acquire for trying times.

A document that is not public and not related to AI, if we’re friends let’s talk about it — it rocked.

Short Books: Letters to a Young Poet and Preliminary Materials for a Theory of the Young-Girl

People cringed at me for reading this commonly remixed text while not being a poetry enjoyer or having read any of Rilke’s writings at all, but I just can’t enjoy poetry! Poetry is one of those things where it hits really hard at the right place and right time, and that comes much less easily to me (or you care about literature). Letters to a Young Poet was really good, my favourite part was the care in all the correspondence, I don’t think there was ever a time or person during which I exercised that care. All the remixes are attempts to provide guidance, but I’d kind of like to see attempts that are more about nostalgia, reflection and reminiscence.

Insofar as I can enjoy poetry, it’s because it sort of performs the secondary purpose of philosophy (the first of which is obviously, being correct) which is to express particular feelings felt in particular times. The Theory of the Young-Girl is certainly not correct, but it really does capture particular feelings at particular times. Every line has its own font!! It lands so well, and does a great job at capturing some particular feelings I have, including the ones that aren’t correct.

Some lines I enjoyed:

Love for the Young-Girl is just autism for two

Look on the bright side, since History’s happening on the dark side [this one really needs to typography to land]

The Young-Girl is never worried about herself, but only about her value. Thus when she encounters hatred, she is struck by doubt: Has her market value dropped?

1. Models

If a property is unspecified it’s either undisclosed or follows approximately the standard GPT recipe.

GPT-3

[paper] — 175B params, 96 layers, 12288 embd dim, 96 heads — OpenAI May 2020

This was a seminal paper for large language models, following the GPT-2 paper (2018) and the scaling laws paper. It was trained on a 300B token dataset consisting mostly of filtered Common Crawl, along with some books, webtext and Wikipedia. BPE tokenizer (same from GPT-2). 2048 context length. Alternates dense and sparse attention layers. Warms up to 0.6 × 10^−4 learning rate in the first 375M toks, cosine decayed to 10% after 260B toks. Batch size ramp from 32k toks to 3.2M toks over the first 12B tokens. 4x MLP projection ratio as done in the 2017 transformer paper. 50k vocab size. Many of these characteristics (e.g. embd dim = 128 * layers, 4x MLP projection ratio, and LR and batch size ramp) form a standard recipe that has been reused by later models.

there’s a probably-typo in Table 2.1 that documents the hyperparameters, where GPT-3 13B is labelled as having an embedding dimension of 5140 which should probably be 5120

GPT-4

[technical report] — Released March 2023, finished pre-training August 2022

Man, feels awkward to write a pathetic summary of something this big, but here goes: GPT-4 is a model available through OpenAI of unknown architecture (other than that it’s GPT-like, though they only technically specify transformer-like). The technical report contains mostly evals (which performed well of course), as well as the results of their continued scaling which are accurately extrapolated from smaller models. The report also documents safety mitigation and has a demo of their multi-modal capabilities of GPT-4 which seem trained in à la Flamingo. It also has the best Acknowledgements section of all time.

Gopher

[paper] — 280B params, 260B non-embedding params, 80 layers, 16384 embd dim, 128 heads — DeepMind Dec 2021

DeepMind’s first large language model release in 2021. It uses an RMSNorm instead of a LayerNorm, uses a relative positional encoding scheme from Transformer-XL instead of an absolute positional encoding, which is why there are so many embedding parameters. Tokenizes with SentencePiece, vocab size 32k. Trained on 300B tokens, with half being from MassiveText which was collected for Gopher, along with books, Common Crawl, Wikipedia, news and Github. Note that Gopher was actually trained end of 2020 and released a year later.

AlphaCode

[paper] — 41B, 8 encoder layers, 56 decoder layers, 6144 embd dim — DeepMind Feb 2022

A model trained on 715GB(967B tokens) of code to do competitive programming. The only model on this list with an encoder-decoder architecture, it treated contest programming as a translation task (problem statement → solution) to gain bidirectionality. It uses 1536 tokens in the encoder and 768 tokens in the decoder. Uses multi-query attention, and generates thousands of samples at inference time and then selects a subset of solutions to submit.

RETRO

[paper] — 7B parameters — DeepMind Feb 2022

Retrieval is the general technique if giving a model a database to look up while doing inference. RETRO was the inaugural retrieval paper for transformers, using a 2T token database. It embeds the token-database in chunks using a pretrained BERT-style model and then performs chunked cross-attention to nearest neighbors in the database during training and inference

GPT-3.5

[docs] — architecture unknown — OpenAI Mar 2022 OpenAI delineates three models as GPT-3.5, specifically anything in the davinci-002 or davinci-003 family. code-davinci-002 is the base model, text-davinci-002 is a version with FeedME non-RL instruction tuning, and text-davinci-003 is an InstructGPT with RLHF. There is an InstructGPT paper that trains an RLHF model and does not mention FeedME, and though text-davinci-002 is an InstructGPT model it does not use RLHF. The davinci model on the OpenAI API is noted to be the 175B model in the 2020 paper, but it’s never confirmed whether davinci-002 is the same size.

Chinchilla

[paper] — 70B params, 80 layers, 8192 embd dim, 64 heads — DeepMind Mar 2022

With the paper titled "Training Compute-Optimal Large Language Models”, new and improved scaling laws were introduced. Chinchilla is trained with 1.5T tokens (similar dataset as Gopher) and same amount of compute as Gopher, yet outperforms it. Results in scaling laws that have parameters and tokens linearly increase at a 20:1 token to parameter ratio. Learning rate adjusts with a cosine schedule. Megatron Turing NLG and Jurassic J-1 Jumbo are two other large models that aren’t documented here as they are not Chinchilla optimal and aren’t uniquely significant.

Flamingo

[paper] — 80B params — DeepMind Apr 2022

Flamingo is a multi-modal (text and image) model. It only generates text, and image inputs are run through a vision encoder (435M params), and cross-attention is used to attend to those outputs. It also uses a resampler (194M params) after the vision encoder to produce a fixed (small) number of visual tokens no matter the number of input features. They build on frozen Chinchilla models, the 80B params come from the cross-attention layers added to the 70B Chinchilla model. PaLI is a Google model that follows up on image/language multimodal.

Gato

[paper] — 1.18B params — May 2022

Gato is a generalist agent, sort of a follow up to Flamingo with more modalities. It uses images and text, as well as button-press data formatted into tokens, as well as encodings of continuous data from robotics propioception, trying to use as little data as possible for additional tasks. The tasks include robotics stacking tests, image captioning, and Atari.

Anthropic LM

[paper] — 52B params, 64 layers, 8192 embd dim — Anthropic Dec 2021

Trained on 400B tokens, though in a later, post-Chinchilla paper, Anthropic used a model with the same architecture trained for 850B tokens. And in yet another later paper on moral self-correction, a 175B with no other specified properties is used.

PaLM

[paper] — 540B params, 118 layers, 18432 embd dim, 48 heads — Google Apr 2022

Current (as of Jan 2023) largest publicly known dense language model, unfortunately pre-Chinchilla. PaLM activates with SwiGLU, uses parallel attention, multi-query attention, rotary embeddings and uses the same matrices for input and output embeddings. No biases were used and a SentencePiece tokenizer with 256k tokens was used. PaLM was trained on 780B tokens, on a similar dataset as LaMDA and GLaM.

GPT-NeoX

[github][paper] — 20B params — Eleuther AI Feb 2022

An Eleuther open-sourced model, trained on GPUs with DeepSpeed (microsoft) and Nvidia Megatron. It uses the same architectural modifications that GPT-J had and is trained on the entirety of Pile, 400B tokens.

GPT-J

[github] — 6.7B params — Eleuther AI Jul 2021

Notable for being a fully open-sourced model, while matching the 6.7B performance from the GPT-3 paper. Trained on TPUs, and done with rotary embeddings, parallel attention. Only dense attention layers are used to reduce complexity. It was trained on the Pile, an open dataset created by Eleuther AI which contains 22 smaller datasets including Common Crawl, OpenWebText, books and papers.

GLaM

[paper] — 1.2T parameters — Google Dec 2021

Named “Generalist Language Model”, GLaM is a Mixture-of-Experts (MoE) model, where parameters are sparsely activated. It has 64 experts per layer, with each token activating 96.6B parameters. Each layer has a gating unit which selects one two of the 64 MLPs per each token

LaMDA

[paper] — 137B params, 64 layers, 8192 embd dim, 128 heads — Google (demoed at I/O May 2021; paper posted Jan 2022)

Dialog model made to follow Meena. A 2.81T dataset with a lot of dialog/forums (encoded with a 32k vocab size SentencePiece tokenizer) is specified. The base model is sometimes called LaMDA GLM or GLM-137B; LaMDA itself adds a lot of dialog finetuning on top.

Though it’s explicit how many tokens the model was trained for. It does specify 1024 TPUv3 chips at 56.5% utilisation for 57.7 days, batch size 256k, probably bf16, and arithmetic says that would be about 900B of the 2.81T tokens.

Switch

[paper] — 1T parameters — Google Jun 2022

An improvement on GLaM, SwitchTransformer only routes to one expert, reducing the amount of compute. It using a different routing mechanism, with the main update being that routing to a single expert works.

BLOOM

[paper] — 176B params, 70 layers, 14336 embd dim, 112 heads — HuggingFace July 2022

Current largest open-source model. Trained on a HuggingFace corpus called ROOTS, which is 498 HuggingFace datasets. The model was trained for 366B tokens. Positional encodings was done with ALiBi. 250k vocab size BPE tokenizer, to help accommodate for multilingual data.

Galactica

[paper] — 120B parameters — Meta Nov 2022

Galactica is a science model pretrained mostly on papers, along with small amounts of code, other knowledge-based data and a bit of common crawl. It uses a <work> token to encode working memory, as well as special tokens for citations.

LLaMa

[paper] — 65B parameters — Meta Feb 2023

Chinchilla replication. Fairly standard training mix of mostly Common Crawl.

Jurassic J1-Grande v2

[paper for v1][helm evals] — 17B parameters — AI21 Dec 2022

No information other than the Helm results, which look really good for the size!

OPT

[paper][train logbook] — 175B params, same arch as GPT-3 — Meta May 2022

Meta replication of GPT-3. Trains on the Pile and PushShift reddit, for only 180B tokens.

The Meta papers aren’t at all connected projects. LLama, OPT and Galactica share only one author of 41.

GLM-130B

[paper] — 130B params — Tsinghua University Oct 2022

GLM is an open-sourced bilingual (Chinese and English) model. It uses rotary embeddings, DeepNorm, and activates the MLP with GeGLU. It notably inferenced in INT4 (where other models like BLOOM and OPT had quantized to INT8). It also includes prompts in pretraining Instead of the standard GPT architecture, it uses GLM for bidirectional attention.

2. Architectural Changes

Multi-Query Attention

This Noam Shazeer solo paper, where the key and values are shared across heads, greatly reducing the amount of memory required at inference time, improving latency and throughput. It’s a perfectly concise barely 9 page paper complete with code and results so it feels silly to describe further. AlphaCode and PaLM both use multi-query.

Sparse Attention

[sparse transformer paper] — Sparse attention is a mechanism where attention is not applied to all previous tokens. It describes two styles of the SparseTransformer, strided where it looks at the last N tokens, and then fixed where sections of tokens in the sequence are attended to. In the GPT-3 paper, the model is described to have alternating dense and “locally banded” sparse layers.

Mixture-of-Experts

There’s a lot more lore on MoE, and I already gave the one-liner in describing GLaM and Switch so here I’ll just give an good initial literature list!

• the original MoE paper from 2017 on LSTMs

• Deepmind Scaling Laws paper for MoE

• Meta paper that trains a 1.1T param MoE

• A large pool of Google papers

FlashAttention

FlashAttention is an architectural change to do attention with less memory access (most of costs in most cases). It tiles and incrementally performs the softmax reduction and avoids storing the whole intermediate attention matrix for the backwards pass. The paper cites 1.7x training speedup compared to megatron and up to over 4x on inference (with the multiplier increasing with longer context lengths). The same sort of approach achieving O(log_n) memory was done earlier on TPUs in this paper.

Encoder+Decoder A la original transformer paper, the encoder decoder architecture was originally made for translation tasks. Where the classic GPT architecture are alternating attention and mlp blocks, the original transformer had an encoder block which was attention → mlp and a decoder block which was masked attention → encoder-decoder attention → mlp. This is still a reasonable architecture to many kinds of sequence-to-sequence tasks, such as AlphaCode or T5 (Google, 2019, 11B params).

Parallel Attention

PaLM uses parallel attention (poorly named) where the model is trained with the attention and MLP layers run in parallel, taking the same vectors. This makes it so that you can do your attention and feed-forward matmuls together to increase arithmetic intensity for better performance (15% on PaLM). GPT-J also uses it.

Activation Alternatives: GeGLU, SwiGLU, SoLU

The original transformer paper uses ReLU (Rectified Linear Unit) to activate the MLP block. It does the simple x if > x = 0 else 0 in between the two linear transformations (matmuls). Intuitively, this is a bit too no-brained. GeLU (Gaussian error) is similar to ReLU but smooths it out a bit. SoLU (Softmax) introduced by this Anthropic paper, is simply x*softmax(x) and is used to improve the interpretability of models. SwiGLU is the most sophisticated of the listed, and is a Noam Shazeer solo paper, as it came through “divine benevolence”. It builds upon gated linear units (meant to be more stable than ReLU) and does the swish operation before the GLU. Like GeLU, it softens out the ReLU and allows some values to be under zero.

LayerNorm Alternatives: DeepNorm, RMSNorm

LLMs norm twice per block (once for attention, once to feed-forward), which does some normalisation function to improve training. DeepNorm and RMSNorm are alternatives. RMSNorm (Root Mean Square) is simply the square root of the mean of the values. There’s also a batch norm that’s inefficient and seems silly to use.

RoPE

[paper][blog post] — I don’t want to try to summarize this one because there’s a good tl;dr in the blog post.

BPE vs SentencePiece Tokenizers [[bpe](https://huggingface.co/course/chapter6/5?fw=pt)][sentence piece] — Byte Pair Encodings are the default for most language models, being used by the original GPT paper, GPT-3 and presumably (based on the API) GPT-3.5. An obvious reason to not use plain BPE (and instead use SentencePiece) is if your distribution doesn’t contain space separated words, as AlphaCode, GLM (Chinese) and PaLM (explicitly because multilingual) did.

ALiBi

Attention with Linear Biases is a long context positional embedding scheme to support extrapolation to longer lengths, by biasing (linearly) the qk scores according to their distance. BLOOM uses ALiBi and Galactica tried it though didn’t go through with it.

3. Post-Pre-Training Techniques

RLHF with PPO

In RLHF, a reward model is trained, where the labeler evaluates an array of model generations. Then the PPO (proximal policy optimization) is used for the RL, where the policy generates an output evaluated by the reward model to improve on the policy.

Deepmind’s Sparrow, as well as Anthropic’s LMs are trained with RL(AI|H)F are have dialog interfaces. WebGPT was was trained with RLHF, as was GopherCite (which called RLHPreferences). I’d say the origination was Christiano 2017, preceding any LLM stuff, followed by 2020 summarizing from human feedback, along with the PPO paper.

Constitutional

An extension of RLHF, Constitutional is basically RLAIF, though actually called “CAI”. It has a supervised learning phase where a helpful-only AI is used to generate adversarial prompts. The assistant then iterates on its own response based on the provided constitution (a short set of values for the model to follow in the form of a string). Then finetuning is done on those responses. The second stage then is like RLHF with PPO, except substituting the AI feedback.

Minerva

Released in 2022 June from the Blueshift team, Minerva is a finetuned model on math and science data, particularly well-executed. It’s a 62/540B finetuned model from PaLM, with datasets from ArXiV and some websites that were carefully preprocessed to preserve mathematical formatting.

Codex

Launched in July 2021 (and resulted in Github Copilot), Codex is a finetune on 100B tokens of code (in this case, publicly available Github code). The paper also debuted HumanEval, human written code evals. This paper most notably demonstrates that code data is really important for code performance, as GPT-J was outperforming 3 at code. They also added some tokens for code, which improved the compression by 30%.

Just Finetune on CoTed Outputs

I forgot which paper did this but its like they finetuned their model on chain of thought outputs from the model, and it did better. Expected, but notable result.

FeedME (SFT)

Described in Instruct GPT paper (though it is not necessarily the origination, which is closer to this). Supervised Fine-Tuning uses human-generated content which is then used to fine-tune the pre-trained model. The paper finds that SFT performs better than base pre-trained models but RLHF performs better than SFT.

FLAN

Flan is an instruction-tuned model (finetuned on instruction-formatted nlp tasks) that results in improved zero-shot performance.

4. Training Techniques

Being Good at Setting Hyperparameters There is obviously no one paper for this, but obviously getting the hyperparameters right is pretty important. Some baseline is available by reading papers, with the most notable probably being Chinchilla or the Scaling Laws paper. There are also a bunch of good theory-based papers, though the one I am familiar with is actually this Jane Street blog post on understanding batch size.

Pre-training with Human Feedback

Pre-training tends to have a very unsupervised format, though PHF(Feb 2023) applies a simple technique to label data at pretraining. It uses two conditioning tokens (good and bad) prepended to samples at training and then samples with them at inference. They tried various other objectives (notably, filtering out bad data) that all performed worse, evaluated on python styling, PII and toxicity.

MuP

Maximal Update Parameterization is a method of parameterization that makes hyperparameters (the ones related to learning rates and optimisers) predictable (consistent) across model sizes. It not only saves the parameter sweep compute but should also be closer to optimal. The paper does a really good job getting into the theory of why this works.

None of the Above

Chain of Thought

This is a technique where it makes the model think “step-by-step” and yielding better results. That name originated in this paper, which describes a specific application of the technique described in this February 2021 paper which describes ways to do prompting that aren’t just few-shotting. The phrase now is sometimes used to describe techniques that aren’t just prompting.

Tool Use

A good canonical tool use paper is probably the Dec 2021 WebGPT paper (though the earliest paper I can find is probably this 2017 Karpathy paper), in which capabilities are greatly enhanced by giving GPT-3 access to the web. It is finetuned with some RL and SL, though I put this not as a training or post-pretraining technique since the concept is not dependent on that. DeepMind also trained RL tool use agents, and Meta has toolformer which does finetuning focused on API usage.

Fill In the Middle

This July 2022 paper describes a simple data transformation which moves a substring from the middle of a text to the end, and asks the model to fill in the middle. This allows the model to gain a capability that is really useful for tasks like code completion without damage to performance on strictly left to right tasks.

Sampling Techniques: Top-k, Top-p (nucleus), Beam Search

The output of language models is fundamentally logits for every possible token, which are then softmaxed into becoming probabilities. The most naive way of turning your logits into tokens, is to take the most likely token. When there are temperature controls with language models, it’s dividing the logits by the temperature, which makes the model more/less confident in its top choice. Top-K sampling takes the top K tokens and samples from that distribution. Top-P sampling (it has a paper but it’s probably useless), or nucleus sampling, uses the top P percentage (think CDFs) of tokens and samples from there.

Tail Free Sampling

Tail Free Sampling takes the derivative of top-P sampling, and is named as such to find the “tail”, as top-P sampling could fail in cases of being cut off at a point where many tokens have similar probabilities. The post linked details the theoretical reasons this should result in better sampling, but when it comes to improving creativity and range in the models there are no good benchmarks.

This feels particularly good to publish now since i feel like its ~the last time something like this could be helpful, given how close we are and the diminishing amount of published research.

Edited by Claude <3

Hillel on universal SE topics, dynamic typing the etymology of “boilerplate” and his notes on Strangeloop. Asterisk mag launched, and has a good introduction to chips and China.

Went on a meditation bender, here’s a book review that correctly characterizes how Buddhist philosophy is bipolar. Book review about meditation. Another book review about meditation. Meditation and sex scandals. Buddhism in relation to rationality. My experience with meditation is that it causes one to experience a philosophy of self (among some other things) and that does not induct that theory into truth. Buddhist philosophy is needlessly senseless because it’s reasoned from that experienced as opposed to more traditional technical thinking. It provides enhanced observational skills, but it’s not obvious that it’s correlated with “living optimally” as opposed to “goes insane”.

U-PALM improves PaLM compute by 2x. Sandwiching paper from Anthropic. Constitutional AI, in which RLHF becomes RLAIF. Language models are really good at generating evals. A prompting paper about making models use rules to do math via prompting. Finetuning models on their own chain-of-thought-ed text improves them. 100M GPT model interpreted (still a bit wavy, but I think we’ll get there 2023)! Meta pretrained science model Galactica, which offensively does not cite Minerva?? Finally read the muP paper on zero-shot hyperparameter determination. No press diplomacy, though I didn’t make it to the full-press paper. Analysis on how SOTA-chasing has affected research. Gradient Descent: The Ultimate Optimizer, an old paper that showed up at NeurIPS with turtles. Finally! Learned optimizers!

Excellent compilation of Elon fucking with Matty. Elon Musk Is Busy with Twitter, Matt Levine is busy with Elon. Oh boy the Libor prosecutions are better than I could’ve asked for.

Book: Structure of Scientific Revolutions

Pleasant read, has really good narrative and was particularly fun because Kuhn focuses on the Copernican revolution, phlogiston theory and various other scientific progresses that I don’t know about because I’m uneducated. There are some obvious overfittings, dubious claims that were particularly uncomfortable for a clean bayesian approach to knowledge acquisition but also some nice tidbits about feelings, like Pauli’s personal crises. The book completes itself with a rather epistemically offensive plot twist. I have 1800 words of notes on the book and probably some inspired upcoming writing.

Essay(?): Designing Freedom

Designing Freedom is this Stafford Beer cybernetics “book” (according to Google, but it’s fifty pages). I don’t endorse it intellectually — not at all, they make these stylized control diagrams that don’t actually work and big handwavy statements. But aesthetically — super fun! I purchased another Beer book (who, by the way, has a daughter named “Vanilla”. Vanilla Beer) called Platform for Change which I often skim through. I wrote notes on this too!

Class Struggle is a board game designed by an NYU professor, Bertell Ollman. There’s like a Wikipedia page about the historical impact of the game and how it taught Marxism through gameplay. It discusses how people called it subversive and tried to get rid of it, Mentalfloss talks more about the relationship to Monopoly (originally called The Landlord’s Game, which was to be about georgism and not capitalism before Hasbro got to it). I’ve also digitized the contents of the game at the bottom!

What isn’t captured, is the sheer aesthetic value of the game itself, independent of its creator or any historical details about its sales. While I hoped to relearn marxism by first principles through gameplay, the game is not educational. The value is that it’s a fantastic shitpost that paints marxism in a positive light with stellar comedic value. I don’t know how people glossed over the comedic value part, the cover is a photoshopped image of Karl Marx and Nelson Rockerfeller arm wrestling, with Rockerfeller looking rather undignified.

The game opens in an animated discussion to determine our privilege order. This was particularly colourful in a room full of San Franciscans who have experienced oppressions that are a bit more complicated that “WOMEN AND BLACKS HAVE LESS CHANCE THAN WHITE MALES TO BECOME CAPITALISTS”. Of course, the whole rulebook is littered with capitalised text.

Shitposting is a term that was created for the internet, and originally about low effort trash talk to ignite flamewars. Shitposting, at least as people around me use the term, is really about improv. There are many more attributes of shitposting, but here we see an important attribute — to express real values comedically, such that people who may not share that value to the full extent can enjoy the notion. A true “yes and” experience.

To help you take the game less seriously, or perhaps experience first-hand the pains of the working class, the die don’t really work. They’re too perfectly cube shaped and don’t roll at all. We could’ve used better die, but would be inauthentic.

60. Black and white workers unite to fight racism. Racism is one of the main reasons for the division and resulting weakness of the working class. The capitalists understand this all too well, and do their best to promote hostility between black and white workers. Workers—3 assets

Each square on the board has a fairly intense statement, as square 60 which is “Black and white workers unite to fight racism”. This is of course accompanied by a rule book which describes all 84 squares in further detail. I yes-and-ed pretty hard to this! I also know there’s a lot more to racism than capitalists trying to divide the working class. I’m sure Ollman knows this too! But this method of sharing this idea (shitposting) is so much better than having to say “I understand there’s a lot more to it than capitalism, in fact it’s mostly other things but I wanted to just talk about this part for now”.

Here is a sample of some chance cards for the capitalists’ deck.

Northtop Airplane Company gets caught trying to bribe two Saudi Arabian generals and loses billion dollars contract

Sexually repressed people generally make good, docile workers, so you develop a sex-education program which makes young people disgusted by their natural functions. Easily worth a couple of assets.

The C. I. A. mistakenly assassinates the leader of a friendly country. Capitalists deny everything and move 2 spaces back.

Capitalists fool workers into equating communism with spying, with the result that communists Ethel and Julius Rosenberg can be executed for a crime they didn’t commit.

The first card references the 1975 bribing of Saudi generals by Northrop (which still exists in some form)! Changing the name a little bit is such a great joke! The Rosenberg thing approximately happened. The CIA assassination thing is maybe in reference to Frank Olson, or perhaps a more niche situation I don’t know about? I really don’t know whether people explicitly did sexual repression to get better workers (honestly I’d expect it to go the other way). I like using humour to promote things, as opposed to insulting things (this uhh, did both but I think it is primarily intending to promote). Humorously promoting and warming people to ideas through shitposts is a key property to how shitposts work.

Some chance cards for the workers’ deck.

You have just been laid off from work. If you blame yourself, or foreign competition, or the Blacks, or Jews, move two spaces back. If you blame the Capitalists, move two spaces ahead.

You get caught stealing food from the supermarket. You get thirty days in jail and are ordered to move back 1 space. Stealing is no answer to the problem of poverty.

If religion is the opium of the older workers, then opium (pot) is the religion of the younger set. While you’re looking at the lights inside your head — “Groovy, man, real groovy”—the Capitalist slips you one of his debits.

Shitposts can be used to highlight a critique of an idea, while not agreeing with it in its full expression. Class Struggle doesn't really balance between positive and negative implications, but that's fine really.

Class Struggle is what some may call a “high effort shitpost”, which is something we’ve lost with the ease of the internet. It has brought me aesthetic euphoria for two months now, may there be more content like this!

Chance Cards

You purposefully produce cars to wear out sooner than is technologically necessary in order to keep up the demand for new ones. For such good Capitalist thinking, move ahead to the next Chance square.Small Businessmen are so frightened by the possibility of going broke and becoming workers that they generally do what the Capitalists say. Move directly to the next square that that allows for an Alliance with the Small BusinessmentCapitalists fool workers into equating communism with spying, with the result that communists Ethel and Julius Rosenberg can be executed for a crime they didn’t commit. Capitalists move ahead 3 spaces. It’s not clever to take so many chances. Skip your next turn at the dice.There is less chance in having lots of chances than you think. Your hard work in the struggle has earned you 2 more Chances.If you haven’t washed the dishes or made supper this week, move 2 spaces. ahead. (Divisions between the people serve the Capitalist class.) You get caught with your hands in the public’s pocket — they’re in deeper than usual. In an attempt to white-wash your reputation, you set up a charitable foundation (tax exempt and named after your family), but are forced to make a public announcement of your assets and debits.You are caught feeling sorry for the Workers. Victory in class struggle comes to people who think about their own class. Miss 2 turns at the dice.Your class allies have lost confidence in your leadership and want to keep a closer eye on what you do. Take all your allies and go back to the beginning of the stage of the class struggle on which you stand. All your propaganda says a person is free when the Government lets him alone. But almost everything one wants to do or have costs money, so only Capitalists are really free. You can use your freedom to move 2 spaces ahead, after paying the Workers 2 assets. (If you can’t pay up, move 2 spaces back)Your daughter has just eloped with the garbage collector. Skip a turn at the dice while you’re thinking of something to tell the neighbours.Farmers are fooled into blaming consumers (instead of the profits Capitalist middlemen) for the low prices they get for their crops. Move directly to the next square that allows for an Alliance with the Farmer Class. Skill at red baiting, getting Workers to fight among themselves by calling some of them communists: next confrontation, take away 1 asset from the Workers.Sexually repressed people generally make good, docile workers, so you develop a sex-education program which makes young people disgusted by their natural functions. Easily worth a couple assets.Rockefeller gets photographed giving people the finger. It’s not wise letting people see what Capitalists really think of them. Miss a turn at the dice while you think of new ways to to fool the people. The next time one of your minor class allies threatens to act against your interests, give him this card and he will go along with you. Your power rests in money—$$$$$$$$$$$$You know your interest and the interests of the other classes in the class struggle. Such ‘class consciousness’ is worth double the assets you now have. Take them.The Mafia makes you a proposition you can’t refuse: for 2 assets, it will see to it that the Workers miss 2 turns at the dice. Paperback edition of the Marx/English Collected Writings (100 volumes) sweeps the country. Your days are numbered — 2 debits.You are treating your allies so well that they want to be at your side. Bring all your allies, from wherever they are on the board, into the square you are now in. (They do not get to pick a Chance Card however).You embezzle one million dollars from your stockholders. No one sees or cares. Move 1 space ahead. Northtop Airplane Company gets caught trying to bribe two Saudi Arabian generals and loses billion dollar contract — 2 debits.Coal mine disaster caused by absence of safety equipment that you refused to put back in because, you said, it was too expensive. Send roses to the funeral, and move back 3 spaces until the public outcry blows over.Most Government welfare goes to Big Business (tax write-offs, price supports, incentive payments, guaranteed loans, etc.) While making a big fuss over the little welfare that goes to the poor, who really need it, help yourself to more welfare in the form of 2 assets. Bribe money to pay union officials. While Capitalists or their allies hold this, the Workers and their allies can’t get any assets from squares which mark the establishment of trade unions.The C.I.A. mistakenly assassinates the leader of a friendly country. Capitalists deny everything and move 2 spaces back.The class struggle is progressing faster than you think. Move immediately to the next Confrontation square. In an effort to balance the budget, your school district has abolished free milk for rich and poor alike. When all are equal before the law, you get richer—2 assets this time.Capitalists fool workers into equating anarchism with criminality, with the result that anarchists Sacco and Vanzetti can be executed for a crime they didn’t commit. Capitalists move ahead 3 spaces.You are treating your class allies very badly, giving the Workers a chance to force a switch in the alliance of any class it chooses. Your son has become a follower of Reverend Moon and your daughter is hooked on heroin. So what good is all your money? Worrying about it all causes you to forget your next turn at the dice.Tax Refund: many big corporations don’t pay any taxes so there is nothing to refund them, but your tax accountants are less talented and more honest so you have paid something and now you get something back. Take 4 assets.Tired of breathing polluted air and drinking polluted water? Thon It's time to make pollution less profitable to the factory owners who are responsible for it. Move together with all your allies to the square on which the Capitalists are now standing for a protest demonstration. Let 'em hoar you Workers finally understand that, with America's wealth and democratic traditions, socialism here will be different than what exists in Russia and China—or "the Russian bogey-man no longer scares us". A biggie—worth 5 assets.Because you have been brainwashed to respect everyone in authority, you continue to respect your boss no matter what he's done. While learning to respect only those who deserve your respect, you miss a turn at the dice.Candidate Carter promised to reduce the Defense budget by $5-7 billion, but President Carter has just increased it by $4 billion, while cutting corners on aid to cities, the poor, the old and the unemployed. The Workers' anger at such deception is worth 2 assets. Together with your fellow workers, you have occupied your factory and locked your boss in the toilet. Capitalists miss 2 turns at the dice.It's no, clever to take so many chances. Skip your next turn at the dice.There is less chance in having lots of chances than you think. Your hard work in the struggle has earned you 2 more Chances. You join a protest against the Government for paying businessmen-farmers $3 billion each year not to grow food. Don't they know a lot of people are hungry, or don't they care? Collective action moves you forward 2 spaces this time.The next time one of your minor class allies threatens to act against your interests, give them this card and they will go along with you. Your power relies on numbers and organization.You know your Interest and the Interests of the other classes in the class struggle. Such 'class conscious-ness' is worth double the assets you now have. Take them. You are treating your class allies very badly, giving the Capitalists a chance to force a switch in the alliance of any class it chooses.You are caught feeling sorry for the Capitalists. Victory in class struggle comes to people who think about their own class. Miss 2 turns at the dice.If you haven't washed the dishes or made supper in the last week, move 2 spaces back. (Divisions between the people serve the Capitalist Class). A socialist teacher has just gotten fired from the local school for playing "Class Struggle" with her class. You help organize the community to get the teacher's job back, and earn 2 assets by the show of solidarity.Integrity insurance against Capitalist bribes of trade union officials. This card neutralizes the Capitalists' Chance Card on bribe money.If religion is the opium of the older workers, then opium (pot) is the religion of the younger set. While you're looking at the lights inside your head—"Groovy, man, real groovy"—the Capitalist slips you one of his debits. If all wealth is produced by the Workers, then what right do Capitalists have to any part of it, let alone the largest part? This "dangerous" question sends shivers up the spines of Capitalists every-where. Grab 2 extra turns at the dice while the Capitalists are still shivering.Teachers and engineers are beginning to recognize that they have bosses just like other workers. Move directly to the next square that allows for an Alliance with the Professional Class.UNION CARD
You've joined the union. The union makes us strong. Take 2 extra turns with the dice. Yesterday you shook hands with Republicrat Senator Kennewater, and you believed him when he said he is the workingman's candidate. Lose 1 asset for being so gullible.Your boss died, but the new one acts in much the same way. You begin to understand the problem is not a mean boss but the class of bosses-2 assets.The class struggle is progressing faster than you think. Move immediately to the next confrontation square. A few Capitalists have recognized the injustice of our present economic system and have decided to support the Workers' Movement. Individual exceptions in the class struggle are possible. Worth 2 assets.You get caught stealing food from the supermarket. You get thirty days in jail and are ordered to move back 1 space. Stealing is no answer to the problem of poverty.You are treating your allies so well that they want to be at your side. Bring all your allies, from wherever they are on the board, into the square you are now in. (They do not get to pick a chance card however.) You have just been laid off from work. If you blame yourself, or foreign competition, or the Blacks, or Jews. move two spaces back. If you blame the Capitalists, move two spaces ahead.A fire destroys your home—you lose everything—but your Capitalist landlord collects insurance money. You miss a turn at the dice while looking for another apartment.Students are beginning to recognize that when they finish school most of them will become Workers. Move directly to the next square that allows you to make an alliance with the Students. Socialist ideas are spreading in the army. (Worth 1 asset in the General Strike Confrontation, 3 assets in the Revolution).Tax Refund: with taxes deducted from each week's pay check, workers are the only people who can't cheat on their taxes. The Government makes up for this exception by cheating the Workers, taxing them too much and giving them too little, even when it comes to tax refunds. Take 2 assets.Serious illness of mother-in-law bankrupts the whole family and drives you to drink-1 debit. Drinking Is no answer to poverty.

Rule Book

DON'T BE SCARED BY ALL THESE RULES The three sets of rules differ in regard to complexity and the roles they give to chance and strategy, with BEGINNERS RULES relying most on chance and TOURNAMENT RULES permitting the most strategy. BEGINNERS RULES-The basic structures for the game are laid down. Everyone should start here. READING THE SENTENCES WHICH ARE CAPITALIZED IN EACH RULE IS USUALLY ALL THE PREPARA TION THAT IS NEEDED TO BEGIN PLAYING FULL RULES-The structures are filled in with the life of capitalist society. Adults and older adolescents should move on to these rules as quickly as possible, even-by common consent-in the middle of a game. But don't stop here. Be- fore rigor mortis sets in, go on to... TOURNAMENT RULES-This is where the real action is. See, for example, TOURNAMENT RULE number 8 (this rule can be added to FULL RULES whenever players feel ready for it). No adult should linger more than one game each in BEGINNERS and FULL RULES before taking on the wheeling and dealing made possible by TOURNAMENT RULES. It is possible to play "Class Struggle" with two players, but-given the importance of alliances-it is better to play with four, and best with six. Go ahead, invite another friend to play.

Full Rules and Some Strategies OBJECT OF THE GAME "Class Struggle" reflects the real struggle between the classes in our society. THE OBJECT OF THE GAME IS TO WIN THE REVOLUTION... ULTIMATELY. Until then, classes-represented by different players- advance around the board, making and breaking alli- ances, and picking up strengths and weaknesses that determine the outcome of the elections and general strikes which occur along the way. PLAYERS

1. "Class Struggle" can be played by two to six players.

2. The real players in "Class Struggle," however, are classes, not individuals. Workers (those who produce shoes, cars, houses and so on) and Capitalists (those who own the machines and factories with which these things are pro- duced) are the Major Classes. Farmers, Small Business- men, Professionals (doctors, lawyers, professors, etc.) and Students are the Minor or Allied Classes. In the game, the hammer symbolizes the Workers, the top hat-the Capi- talists, the tractor-the Farmers, the cash register-the Small Businessmen, the brief case-the Professionals, and the mortarboard-the Students.

3. Only Workers and Capitalists can win this game. The other classes participate in winning or losing through alliances with one or another Major Class. (See Rule 15) While Workers and Capitalists struggle to win, the Minor Classes struggle to be on the winning side.

4. Individual players cannot choose their class. In real life, ONE'S CLASS IS DETERMINED BY CHANCE, which usually means by the kind of family into which one is born. Also in our society, WOMEN AND BLACKS HAVE LESS CHANCE THAN WHITE MALES TO BECOME CAPITALISTS. This has nothing to do with the human qualities of women and Blacks and everything to do with the unfair rules set by our society. Attempting to reflect these rules (and not by any means to justify them), "Class Struggle" calls for the following: beginning with the lightest White male and ending with the darkest Black female, everyone takes turns with the Genetic (or luck-of-birth) Die, the one with the symbols on it, to see who throws the Capitalist Class first. (If the people playing include a black man and a white woman, the players themselves have to decide which one has the greater handicap in becoming a Capitalist). After the Capitalists are chosen in this way, the players throw the Genetic Die in just the opposite order to see who plays the Workers. The remaining players can throw the Genetic Die in any order to see who plays the other classes.

5. It is important that Workers and Capitalists be represented in the game, so if there are only two players these are the classes they should play. If there are less than six players, one person can represent two Minor Classes. It is also possible in a two-person game for each player to represent a Minor as well as a Major Class. It is simply more inter- esting when all or most of the Minor Classes are represented, but it is also possible to play "Class Struggle" with one or more of the Minor Classes left out.

6. Again, to be true to real life, where the Capitalists' wealth and power over people and factories give them many unfair advantages, the Capitalist Class is the first to throw the numbered dice. They also decide whether the order in which the other classes take their turn at the dice proceeds from their left or their right.

7. Classes now throw the numbered dice and move forward as many squares as the number shown, except when this number is the same as the one thrown by the class that Z went before. In this case, the class which has just thrown the dice does not move at all. THINKING FOR ONE- SELF, AND NOT JUST DOING WHAT OTHERS DO, IS ESSENTIAL FOR WINNING AT CLASS STRUGGLE. If a class throws a double number, it gets an extra turn at the dice, and this rule holds for as long as any class throws double numbers.

   ASSETS AND DEBITS

8. Most of the squares on the board are divided between Workers (red) and Capitalists (blue), and list the real strengths and weaknesses of these two classes. Depending on its importance in the Class Struggle, each strength is worth one to three plus-points (called "assets") and each weakness one to three minus-points (called "debits"). Whenever Workers or Capitalists land on a square which carries its name, it picks up the number of assets or debits listed there. Landing on its opponent's squares earns the trespasser neither assets nor debits, unless this happens on three successive throws of the dice (and only if one's opponent has kept track of this). In this case, the offend- ing class picks up one debit. PRETENDING TO BELONG TO A CLASS OTHER THAN YOUR OWN WEAKENS YOUR CAUSE IN THE CLASS STRUGGLE. NEW

9. As part of their unfair advantages, the Capitalists decide which of the other classes should do the hard but neces- sary work of handing out assets and debits, that is of taking care of the "Bank."

10. The Minor Classes pick up assets and debits whenever they land on either Capitalist or Worker Squares until they enter an alliance with one of the Major Classes, after which they get points only from squares that carry the name of their ally. (See Rule 15).

CHANCE

1. When landing on a Chance square, Capitalists pick up a Chance Card from the pack marked "Capitalist", and Workers from the pack marked "Worker". WHAT IS SAUCE FOR THE GOOSE MAY BE SAUCE FOR THE GANDER, BUT WHAT IS GOOD LUCK FOR 27 THE CAPITALISTS IS BAD LUCK FOR THE WORKERS, AND VICE VERSA.

2. Until they enter into an alliance with one of the Major Classes, the Minor Classes can take their Chance Cards from either pack. After an alliance, they must pick their cards from the same pack as their Major Class ally. (See Rule 15.) on squares 16 or 56

3. Trade Union and Workers' Political Party Cards: if the Workers or their allies land on squares 11 or 52, they receive a Trade Union Card; landing earns them a Political Party Card. Minor Classes can acquire these cards only while they are allied with the Workers, and must return them should this alliance be broken (See Rule 15).

4. On the highest level of "Class Struggle", starting with square no. 65, classes have a choice of receiving the num- ber of assets and debits listed on the squares on which they land or forcing any one of their opponents to move double that number of squares (backward when it is assets that are listed, forward in the case of debits). The latter class then picks up the assets, debits, chance card, etc. listed on the square to which it has been forced to move.

   ALLIANCES

5. Squares which read "Chance for an Alliance with the Farmers" (or Small Businessmen, or Professionals, or Students) permit the Major Class which lands there to enter into an alliance with the Minor Class named there. Each Minor Class has a Class Alliance Card which it gives to its new ally to seal the alliance. Like the player pieces, Alliance Cards can be mounted on the blocks of wood which are provided. Though each allied class retains its own assets and debits, their points are counted together in any future "Confrontation" (See Rule 27).

6. When there are less than six players in the game, a Major Class which lands on the Alliance Square of a Minor Class that is not represented by a player still receives the latter's Alliance Card and a bonus of five assets.

7. Whenever the Capitalists or Workers enter into an alliance with a Minor Class that is represented by a player they agree-in order to cement their new relationship-to Z accept from the Bank a number of debits equal to the number showing on their new ally's next turn at the dice, except when a double number is thrown, in which case there is no need to accept any debits at all. THERE IS OFTEN A PRICE TO PAY IN FORMING NEW ALLIANCES, AND CLASSES MUST ASK THEMSELVES, "IS IT WORTH IT?"

8. Minor as well as Major Classes have a chance to enter into alliances if they land on an Alliance Square. Rule 17 also applies here, and again it is the class which lands on the Alliance Square and initiates the alliance which receives the Alliance Card from the other class and picks up the extra debits.

9. If a Minor Class (say Farmers), which is already allied to a Major Class (say Capitalists), lands on a square that makes possible an alliance with the Students, the lat- ter automatically becomes an ally of the Capitalists as well. In this way, the Minor Class allies of each Major Class can pick up alliances for their Major Class ally.

10. If a Minor Class is not allied to either of the Major Classes, it can still enter into an alliance with another Minor Class. In this case, the two Minor Classes have made themselves doubly valuable to the first Major Class to enter into an alliance with either of them; for-given the alliance of the two Minor Classes-to ally with either one of them is to ally with both. This also implies, of course, that the Major Class in question agrees to accept from each new ally a number of debits equal to the number showing on its next turn at the dice, except-again -when double numbers appear.

11. After two Minor Classes enter into an alliance with a Major Class, their special relationship comes to an end. Thus, if the other Major Class forces either of these Minor Classes to change alliances, its one-time partner is unaffected.

12. Change of alliances: whenever any class (Major or Minor, allied or unallied) lands on an Alliance Square of a Minor Class already allied to its opponent (or picks up a Chance Card which speaks to the same thing), a change of alliances becomes possible. Rule 17 also applies in every change of alliances.

13. If there are less than six players in the game, forcing a change of alliances with a Minor Class that is not represented by a player is simply a matter of transferring the Alliance Card and the five assets which came with it from the old Major Class ally to the new one.

14. Right to refuse alliances: whenever there is a chance for an alliance or a change of alliances, the classes directly affected have a right to accept or refuse it. If both classes agree, an alliance is formed. If both classes reject it, there is no alliance. If one wants the alliance and the other does not, the two throw the dice and class with the higher number has its way. In case of a tie, they throw the dice again.

15. Natural Alliances: SOME CLASS ALLIANCES ARE MORE VALUABLE THAN OTHERS. If the Capitalists gain an alliance with both the Small Business and Professional Classes, the assets and debits of the latter two classes are counted twice in "Confrontations". (See Rule 27). This applies only if both of these classes are allied with the Capitalists at the same time. If the Workers gain alliances with the Farmers and Students, the same thing applies: the assets and debits of these two classes are counted twice.

16. The Natural Alliance Rule holds even if one or both of the Minor Classes involved are not represented by players. That is, the five assets the Workers would have received for allying with the Farmers and the five assets they would have received for allying with the Students assuming neither class is represented by players (see Rule 16)-is worth twenty assets as long as this double alliance holds.

CONFRONTATION

1. There are six Confrontation Squares: Life in the factory, two Elections, two General Strikes (when all the workers lay down their tools) and the Revolution. If either Major Class or its allies lands on a Confrontation Square, it has a choice whether or not to call a Confrontation. Non- allied Minor Classes cannot call a Confrontation, and only the Major Classes (not even their allies) can call the final Confrontation, which is the Revolution. In a Con- frontation, each side adds up its assets and debits (allies' points are counted together), and the side with the highest number of assets after debits are subtracted wins. In the case of the Elections and General Strikes, winning the Confrontation secures the victorious Major Class three free throws of the dice, improving in this way its position in the overall Class Struggle. Rule 7 regarding double numbers does not apply to these three throws. To keep opponents guessing as to whether they can win a Con- frontation, each class should keep its own assets and debits securely covered.

2. It is obviously not in the interest of a class to provoke a Confrontation unless it believes it can win. To make an intelligent judgment on this matter, each class should try to keep track of the assets and debits acquired by other classes. IN CLASS STRUGGLE, VICTORY GEN- ERALLY GOES TO THE CLASS WHICH KNOWS THE STRENGTHS AND WEAKNESSES OF THE OTHER SIDE AS WELL AS ITS OWN.

3. Elections are true Confrontations only if the Workers, or one of its allies, has already landed on Squares 16 or 55 indicating that a Working Class political party has been formed. Otherwise, the Capitalists automatically win this Confrontation. IN AN ELECTION BETWEEN THE CAPITALIST PARTIES OF TWEEDLE DEEmocrats AND TWEEDLE DUMlicans, THE WORKERS CAN ONLY LOSE.

4. Revolution is the last square on the board and the final Confrontation. The winner of this Confrontation wins the game. If the Workers win, humanity enters a new era of peace, democracy and equality, which is called SOCIAL- ISM. A Capitalist victory, on the other hand, simply means the rich get richer while the poor are left to stew in their own juice, leading eventually to the collapse of civilization. HENCE THE CHOICE BEFORE US SOCIALISM OR BARBARISM?

5. A Revolution can be called by either of the Major Classes once it arrives in the final square. As in earlier Confrontations, the assets and debits of both Workers and Capitalists are counted no matter where these classes are on the board. The allies of the Major Class which calls the Revolution, however, are counted only if they are safely in the final square when the Revolution is called. The points of the allies of the other Major Class are counted no matter where these classes are on the board. REVOLUTION IS THE GREAT TEST OF ALLIANCES AND SOME CLASSES ARE FOUND WANTING.

6. For the Natural Alliance Rule (See Rule 25) to apply in the Revolution, the Major Class and both of its natural allies must be in the final square. If only one natural ally has made it to the final square in time for the Revolution, its assets and debits are counted but once. In the case of classes not represented by players, simply having the Natural Alliance is sufficient for doubling their assets in the Revolution as in all other Confrontations.

7. For any class to advance to the final square, it must roll an exact number on the dice. If it is two squares away, it has to keep throwing the dice until it gets a two. If it is one square away, it throws only one die.

8. Once any class (Major or Minor) arrives in the final square, it continues to receive a turn at the dice which it can use to advance any of its allies. The ally that it wishes to help must be chosen before the dice is thrown.

9. As the Capitalists control the Government, only they not even their allies can trigger off a nuclear war and destroy the whole world. IN DANGER OF LOSING THEIR POWER, THE CAPITALISTS ARE CAPABLE OF ANYTHING. If the Capitalists land on Square 81, then, this means nuclear war and the automatic end of the game. Nobody wins. If the Workers or any of its allies land on this square first, however, the possibility of the Capitalists starting a nuclear war has been removed for the remainder of the game.

10. It is in the interest of every class to become part of a Natural Alliance (see Rule 25), and it is this interest which determines most efforts to establish alliances.

11. Since establishing an alliance with a Minor Class generally means accepting a few extra debits (except when the latter throws a double number on the dice), there will be occasions when a Major Class may want to refuse such an alliance. These are when the Minor Class in question has few or no assets, early in the game when changes in alliances are still very likely, or late in the game when the Major Class has a large lead in assets and believes it can win without this alliance. Should this happen, the Minor Class on whose Alliance Square the Major Class has landed-wishing to be on the winning side-may insist on forming an alliance, in which case the two classes throw the dice with the higher number deciding the issue.

12. The necessity of getting one's Minor Class allies into the final square in order to have their assets counted in the Revolution gives a Major Class with few and even no Z allies a chance to win. Though it is usually an advantage to have as many alliances as possible early in the game for the help they provide in the early Confrontations, this is not always so in the end. As the Revolution approaches, the value of alliances depends in large part on the position of one's Minor Class allies on the board, and this should be kept in mind in deciding whether to accept a new alliance.

13. For a Minor Class, the possibility of being left out of the Revolution (being short of the final square when the Revolution is called, or being there but not having an alliance with either Major Class) offers a third possible outcome to the game besides winning or losing. Again, the main objective of each Minor Class is to be on the winning side, but if this goal is beyond its reach it should try at least to avoid being on the losing side. In pursuit of this second best outcome, a Minor Class should refuse an alliance with a Major Class if it believes the latter is going to lose.

14. Because it picks up assets and debits from both Capitalist and Worker Squares before it enters into an alliance with a Major Class, there is some advantage to a Minor Class' steering clear of alliances until it sees which of the Major Classes is ahead.

SOCIALISM OR BARBARISM

1. The class which wins the Revolution has only won in one country, but Class Struggle is taking place throughout the capitalist world. Give the winning class(es) a bonus of five assets (VICTORY IN ONE COUNTRY HELPS ONE'S CLASS IN OTHER COUNTRIES), roll the Genetic Die to see who is going to represent which class (AS ALWAYS, EACH PERSON'S CLASS IS LEFT UP TO CHANCE), and let the struggle begin anew.

INNOVATION

1. But before you start, look around you at life in capitalist America and add at least one new Chance Card for each Capitalists and Workers. Several blank Chance Cards have been provided for this purpose. The player who offers the best idea for a Chance Card should be rewarded with two assets for the new game.

Tournament Rules (These rules are for older adolescents and adults who are already familiar with the FULL RULES and who favor games with more complex strategies.)

1. In BEGINNERS and FULL RULES, winning has been simply a matter of having more assets at the end of the game. TOURNAMENT RULES, however, recognizes three kinds of victories: smashing victory, middling vic- tory and bare victory, which is no less than what can happen in any real revolution.

2. Here, Workers or Capitalists win a smashing victory if it

   a) beats the other Major Class by a margin of fifteen or more assets, b) has more allies than its opponent in the final square at the end of the game, and c) has won all of the confrontations which were called.

3. A Major Class wins a middling victory if it satisfies any two of these criteria. If less than two of these criteria are satisfied, the winning class wins but a bare victory.

4. In the real world, the larger the victory, the quicker the winning class can put its program into effect. It also contributes to the success of this same class in the Class Struggles going on in other countries. Consequently, in the next game of "Class Struggle", the class which has just won a smashing victory starts the game with nine assets. A middling victory is worth six assets and a bare victory three. In every case, the allies of the winning class receive the same bonus. These assets go to the winning class and not to the player who represented it, as each new game starts with a throw of the Genetic Die.

5. It is obviously in the interest of each Major Class, therefore, not only to win but to win big, or if it is losing to lose by as little as possible.

6. TOURNAMENT RULES, unlike FULL RULES, also allows for the possibility of a Minor Class, or an alliance of Minor Classes, winning. This occurs when the two Major Classes have fought to a standoff in their struggle. This standoff is represented in the game by the following:

   1. no more than ten assets separates the totals of the Major Classes,

   2. neither has won more than one Confrontation, and

   3. neither has more than one ally.

   In these circumstances, an unallied Minor Class, or an alliance of Minor Classes, which has more assets than either Major Class can-when it arrives at the final square-call a Revolution and win. Again, for the points of the allies of the class calling the Revolution to be counted, these allies must have arrived in the final square. If a Minor Class calls a Revolution and the criteria for its victory are not met, the Major Class with the most assets wins. If the Minor Class wins, it receives a bonus of nine assets for the next game, and so too any of its allies. This unusual outcome corresponds to the equally unusual outcome of real Class Struggles in which a standoff be- tween Major Classes permits representatives of a Minor Class to assume temporary control of the state. Louie Boneparte's rise to power as representative of the small farmers in mid-19th century France is an example of this.

7. The variety of possible outcomes increases the number of options before any class. A Major Class which takes an early lead can try to win big or simply try to make sure that it wins. While a Major Class which falls behind early in the game now has the option of still trying to win or trying to limit the size of its loss. And Minor Classes, if circumstances allow, can strive to defeat both Major Classes, or as before-to be on the side of the winning Major Class, or at least not to be allied to the losing Major Class. Clearly, the possibility and attractiveness of these different options will change as the game progresses with its changing relations between the classes.

8. In order to give each class greater flexibility to pursue these different outcomes, TOURNAMENT RULES permits the following:

   1. any class can buy from another its turn(s) at the dice for whatever (assets, future Chance Cards, decision making power in the alliance, or alliances) that the latter will accept.

   2. a class can also buy from another, using whatever the latter will accept, future Chance Cards (in the hope of coming up with the one or two cards that could change the course of the game). Such a purchase must be concluded, of course, before the card in question is picked.

   3. class alliances can be made, traded or sold, again for whatever the parties will accept. (Only Minor Classes can be sold or traded, whether by a Major or Minor Class ally). As before, the class landing on a Class Alliance Square can force the class named there to enter into an alliance, whether or not it is already part of another alliance (given, of course, that the class which lands on the square wins the subsequent throw of the dice), but now alliances can also be forged and broken in other ways. The only exception to this open market in alliances is that Workers and Capitalists can never ally together. The only time when alliances can be made, traded. sold or broken in the above manner is when a class, any class, lands on an Alliance Square. Besides per- mitting the class which lands there to enter into an alliance with the class named, an open market is de- clared on all such transactions. If the class being traded or sold objects, however, the issue is decided as always by a throw of the dice. This means too that a Major Class which was denied the right to trade or sell its ally because of an unlucky throw of the dice has another opportunity to do so as soon as someone lands on the next Class Alliance Square. A class simply wishing to get out of an alliance cannot force the issue in this way, but must find a price to buy its freedom that its ally finds acceptable.

   4. a Major Class can concentrate as many of its assets as it wishes on any upcoming Confrontation other than the Revolution. This is the political equivalent of "putting all of one's eggs in the same basket". In this case, if the betting class or its allies are lucky enough to land on this square, the assets which have been "sent ahead" count for triple in the Confrontation. If this square is passed over without a Confrontation, or if the Confrontation is called and lost, the assets which were "sent ahead" are returned to the Bank. Given the danger of losing these assets, this strategy is best suited to a Major Class which is already losing but hopes to win at least one Confrontation and in this way keep its opponent from a smashing victory.

   STRATEGY

9. All the strategies listed in Full Rules apply here (See Full Rules numbers 36-40).

10. Knowing that other classes can and are even likely to land on the Class Alliance Squares coming up, it is unwise to make "expensive" alliance deals early in the game, for this will just mean losing alliances for which one has paid dearly. On the other hand, if a Major Class is pursuing a strategy directed toward winning an early (or middle) Confrontation, paying a lot for an alliance-especially if the price is in future turns of the dice or future Chance Cards-may be a clever move.

11. If a Major Class is behind in assets and lagging far behind on the board, it should consider giving one or more of its allies their independence in exchange for assets or future turns at the dice. It is always preferable for a Major Class to strike a bargain with a Minor Class than with the other Major Class, since bargains generally strengthen both parties.

12. If a Major Class is far behind in everything, only luck will save it and it is probably a good idea to trade future turns at the dice, assets, etc. with other classes for future Chance Cards. Generally speaking, this a desperation move, but there are a couple of Chance Cards which could turn the game around. Each Major Class picks up Chance Cards from the pack that carries its name no matter where and from whom it has purchased the right to do so.

13. A Minor Class trying to win by itself (or in alliance with other Minor Classes) will not only use its assets to keep or buy back its independence, but also to bolster the posi- tion of the losing Major Class so that the criteria necessary for a Minor Class to win can be satisfied. (See Rule 6). It can't simply hand over assets to a Major Class which is losing badly, but it can buy a Chance Card from it, for example, at an inflated price.

14. If allied to a losing Major Class which refuses to sell it its independence, a Minor Class may refuse to follow its ally's advice on whether to accept an advantageous alli- ance with another Minor Class or to reject a disadvantage- ous one, or to call or not call a Confrontation should the opportunity arise. The very possibility that one's Minor Class ally might force a Confrontation when one is sure to lose serves as pressure on a Major Class to trade or sell its troublesome ally.

15. An independent Minor Class with a great many assets can offer itself as an ally to a Major Class in return for the power to take all decisions pertaining to the alliance, i.e., whether to call a Confrontation, including the Revolution, whether to accept and/or trade for other allies, what kind of victory to aim for, etc. A Major Class is unlikely to give up such power, of course, unless it sees no other way of winning.

16. In TOURNAMENT "Class Struggle", the relations be- tween the classes are likely to change often and drastically, requiring frequent reassessments of what outcomes are possible, which desirable, and what strategies are best to obtain them. Flexibility in choosing one's goal is as important as skill in choosing effective means in determining both who will win and what kind of victory it will be. 17. For the rest, all the rules given in FULL RULES, except where specifically superceded by 1-16 of the above, also apply to TOURNAMENT "Class Struggle".

Why? Explanations for the assets and debits awarded on each square

1. Class Struggle begins.

2. CONFRONTATION: in the factory every day. With neither land or tools to call their own, workers are forced to sell their labor-power to capitalists for as much as they can get in order to stay alive. Capitalists buy this labor- power for as little as they can give in order to make things which they can sell at a profit. They then try to expand this profit by making workers work harder, faster and longer (no matter how unhealthy and unpleasant the conditions of work, since every improvement in these conditions costs money); while workers struggle to defend themselves against capitalist greed by doing as little as they can as slow as they can.

3. Growth of industry leads to an increase in the number (hence the power) of workers. There are about 95 million people (or 62% of everybody over 16) who work for wages in America today. These are the workers. Growth of industry also leads to an increase in the profits (hence the power) of capitalists. Workers-2 assets; Capitalists -2 assets.

4. Market spreads overseas giving capitalists an important influence in the lives of foreign peoples and their governments. The capitalist system's need for the continuous expansion of this market (since workers are not paid enough to buy all that they make) is at the core of what is called "imperialism". Capitalists-2 assets.

5. Increasing concentration of industry makes workers the majority in most cities. The workers' power lies not only in their numbers but in the fact that they live in cities which are the nerve centers of capitalist society. Workers -2 assets.

6. Competition between capitalists drives many of them out of business. The result is fewer capitalists in relation to the growing number of workers. While many Americans own a few shares of stock, there are only about one half million people who own so many shares that they have an influence on how the business is run and/or can live off their dividends. These are the capitalists. Capitalists-1 debit.

7. Growth in church attendance. Because it encourages the poor and oppressed to think more about the next life than about this one, Karl Marx called religion, "the opium of the people". Workers-1 debit.

8. Chance for an alliance with the Farmers.

9. Congress is supposed to represent all the people, but most of its members (Democrats and Republicans) are capital- ists and lawyers for capitalist interests. It is no wonder then that most of the laws which Congress passes favor the capitalist class. In the words of Anatole France, "The law in all its majesty forbids rich and poor alike from stealing bread and sleeping under bridges". So much for what capitalists, without recognizing the irony, call "equality under the law". Capitalists-2 assets.

10. CHANCE.

11. Trade Unions are established to defend workers' interests (higher wages, shorter hours, safer conditions and the like). Only through organization can the workers' power be felt. Workers-3 assets.

12. Natural disaster. Floods, hurricanes, etc. mean workers lose their homes and cars, while for the capitalists dis- asters generally mean more business, since workers have to buy more homes and cars. Workers-1 debit; Capitalists-1 asset.

13. Chance for an alliance with the Professional Class.

14. Competition between workers for jobs and, inside each work place, for the few easier and higher paid jobs makes it difficult for workers to cooperate on behalf of their common interests. Workers-1 debit.

15. Police are not for directing traffic, nor soldiers for march- ing in parades. The main job of the police and the army is to protect the factories, stores and fine homes of the capitalists. Capitalists-2 assets.

16. Workers organize a working class political party. Only when they stop relying on capitalist parties to represent them can the workers hope to win in class struggle. Workers-3 assets.

17. Big drop in the stock market. Millions of small investors lose their savings and capitalism momentarily gets a bad name. Capitalists-2 debits.

18. CHANCE.

19. CONFRONTATION: election. An election is a true confrontation only if the workers have their own political party. If elections are a matter of choosing between bosses who look like elephants and bosses who look like donkeys, the workers can only lose.

20. Trade Unions are taken over by union bureaucrats (only applies if the Workers or their allies have landed on number 11 above). When the unions are led by people who think only of their own careers, the workers' interests are forgotten. Workers-2 debits.

21. Chance for an alliance with the Small Shopkeepers.

22. Press, radio and T.V. are owned by the capitalists and present their point of view on everything. This is not just a matter of editorials, but of what the media merchants choose to call news and the slant they give it. Capitalists -3 assets.

23. The success of the Tennessee Valley Authority (T.V.A.) shows that production can go on without capitalists. An important lesson. A public corporation, the T.V.A. pro- duces and sells electricity more cheaply than any private power company could... or would. Workers-2 assets; Capitalists-1 debit.

24. CHANCE.

25. Capitalists control the courts. With laws biased on behalf of the rich, establishment judges who naturally sympathize with their own kind, and the best lawyers working for those who pay most, there is little real justice. As a general rule, the richer the criminal (for those few rich criminals who are caught and tried), the shorter the sentence. Capitalists-1 asset.

26. Chance for an alliance with the Students.

27. Strike. By withholding their work from the capitalists, the workers force them to give in to their demands, and win a new self-confidence. Workers-3 assets.

28. Through their control of school boards and the publication of text books, the capitalists determine most of what is taught in the schools. Capitalists-2 assets.

29. CHANCE.

30. Workers party is taken over by party bureaucrats (only applies if the Workers or their allies have landed on number 16 above). Workers-2 debits.

31. Inflation. With higher prices, workers buy less but capitalists make more profit. Workers-1 debit; Capitalists 1 asset.

32. Chance for an alliance with the Farmers.

33. Development of a socialist press to present the workers' point of view. The big job now is making people aware that it exists and encouraging them to read it. Workers -3 assets.

34. CHANCE.

35. State and federal regulatory commissions (like the Inter- state Commerce Commission) are supposed to protect the public from overly greedy (as distinguished from typically greedy) businessmen. When the capitalists them- selves or their lawyers are appointed to these commissions, the wolf has been given the job of guarding the sheep. Capitalists-2 assets.

36. You are about to enter a higher stage of the class struggle. Knowledge of this helps you plan accordingly. On pass- ing: Workers-2 assets; Capitalists-2 assets.

37. Chance for an alliance with the Professional Class.

38. Workers organize community action groups and put pressure on landlords, local hospitals and schools for better services. In the process, they learn how to struggle more effectively. Workers--3 assets.

39. CHANCE.

40. Capitalists provoke an imperialist war in their constant search to find new countries in which to turn a fast buck. In such wars, the workers are usually fooled by patriotic propaganda into supporting their bosses, at least at the start. Capitalists-2 assets.

41. CONFRONTATION: election. (See number 19).

42. Chance for an alliance with the Small Shopkeepers.

43. CHANCE.

44. Watergate scandal forces the Government, momentarily, to put aside its "dirty tricks" operations against the workers and their allies. Workers-1 asset; Capitalists 2 debits

45. Crime in the streets is no answer to poverty. It only debits. divides the workers and increases repression. Workers - 1 debit

46. Capitalists expand their profits by getting the Government -1 debit. to waste more of the tax payer's money on arms, or what they hypocritically call "defense". Before all the double- talk set in, the Defense Department was called the War Department. Capitalists-1 asset.

47. CHANCE.

48. When the capitalists' profit is the deciding factor, progress in inventions leads to fewer jobs rather than to lighter work. That's why workers often oppose the introduction of new machines. Science will work for the people only when the people, not the capitalists, control it. Workers -1 debit.

49. Strike. (See number 27) Strikes become bigger and more frequent as the class struggle intensifies. Workers-3 assets. But the capitalist often responds by locking the workers out of the factory until they accept his conditions. Capitalists-3 assets.

50. Chance for an alliance with the Students.

51. State legalizes gambling to fool each worker into believing that one day he will become rich. The only sure winners in this game are the capitalists. Capitalists-2 assets.

52. More trade unions. (See number 11) In the United States today, less than a quarter of our workers are in unions (much less than in any other capitalist country), so a lot of organizing remains to be done. Workers-3 assets.

53. CHANCE.

54. Successful workers' revolution abroad reduces the capital- ists' business opportunities in that country and encourages the workers' movement at home. Workers-2 assets; Capitalists-1 debit.

55. Workers start a new, more broadly based political party. The more people within the working class who get in- volved in the political struggle, the greater the power of the workers. Workers-3 assets.

56. Chance for an alliance with any of the minor classes.

57. Watergate scandal is forgotten. Government resumes its "dirty tricks". If people only had a better memory for historical injustices, capitalism would have been put to bed a long time ago. It doesn't help, of course, that history teachers spend so little time talking about these injustices. Workers-1 debit; Capitalists-2 assets.

58. CONFRONTATION: general strike. All the workers lay down their tools. The workers' power is evident as all economic life comes to a halt.

59. CHANCE.

60. Black and white workers unite to fight racism. Racism is one of the main reasons for the division and resulting weakness of the working class. The capitalists understand this all too well, and do their best to promote hostility between black and white workers. Workers-3 assets.

61. Using the excuse of the fiscal crisis, the banks refuse to lend the cities any more money until they get rid of free higher education. It is dangerous for capitalism when too many young people from the working class are learning how to think for themselves. It is also not very profitable. Workers-1 debit; Capitalists-1 asset.

62. CHANCE.

63. Increase in attendance and in the T.V. audience of mass spectator sports, another "opium of the people". By making people of all classes into "fans" of the same teams, capitalists also make it difficult for the workers to think about their interests as workers (and turn a neat profit in the process). Capitalists-2 assets.

64. You are about to enter the highest stage in the class struggle. (See number 36 above) Workers-2 assets; Capitalists-2 assets.

65. Crisis (or "depression", or "recession", now called a "pause"): production and sales go down, unemployment goes up, while machines and often the things they produce are left to go to waste and even destroyed. All this at a time when millions of people are going hungry. More and more workers begin to see how irrational it is to let capitalists, who are only interested in their profit, decide what our society needs. Workers-2 assets.

66. Capitalists enlarge the size of the police and the army as their surest means of staying in power. Capitalists-2 assets.

67. Rent and tax strikes are spreading throughout the country. Squeezed by a tight economic situation and encouraged by a growing working class movement, people are ceasing to play by the capitalist-imposed "rules of the game". Respect for traditional authorities is weakening. Workers -2 assets.

68. CHANCE.

69. Like religion, T. V. thrillers, spectator sports and gambl ing, pornography keeps workers from thinking about their class interests and organizing to do something about them. The name of the capitalist game seems to be to get work- ers to think about everything except what is really hap- pening to them. Workers-1 debit.

70. Police scandal exposes the links between the Mafia, Big Business and the cops. Watch out when workers learn that at the top of high society the cops and the robbers are the same people. Capitalists-2 debits.

71. CHANCE.

72. CONFRONTATION: general strike. (See number 58).

73. Male and female workers unite to fight sexism. Along with racism, sexism is among the greatest barriers to working class unity and power. Again, the capitalists understand this very well, and do their best to promote hostility between male and female workers. Workers-3 assets.

74. F.B.I. spies have infiltrated the leadership of the workers' party. In politics, as in business, when the capitalists can't win by playing within the rules (rules which they themselves have made), they don't mind breaking them. Capitalists-3 assets.

75. Capitalists throw many workers' leaders in prison. This is an act of desperation by the capitalists that temporarily stalls the workers' advance. Workers-2 debits.

76. Chance for an alliance with any of the Minor Classes.

77. Strike. "Nobody throws our leaders in jail and gets away with it", say the workers. In a strike, new leaders are formed. Workers-3 assets.

78. CHANCE.

79. Cold Feet. In the final stages of the class struggle, they who hesitate are lost. Think about this as you miss as many turns at the dice as you have allies.

80. Capitalists start another imperialist war but the workers have wised up and refuse to fight. Why kill and be killed in a war that only benefits the capitalists? Capitalists-2 debits.

81. Nuclear War. The capitalists control the state, so only they can trigger off nuclear war. And the capitalists are capable of any folly once they sense their days in power are numbered. If the capitalists land on this square before the workers or their allies, the game is automatically over.

82. Government orders the destruction of all copies of that dangerous game, "Class Struggle". It may be too late, however. Capitalists-3 assets.

83. Socialist ideas are spreading in the police and the army. Policemen and soldiers, especially in the lower ranks, begin to understand that they too have bosses and that they share many interests with the workers. The time when they have unthinkingly served as watchdogs for capitalism has come to an end. Workers-3 assets.

84. CONFRONTATION: revolution. No one knows what a revolution in America would look like. All that can be said for sure is that the more people who want it, who reject capitalism and who want a society shaped in the interests of the workers and their allies, the swifter, more democratic and less violent it will be. Yes, revolution, which means a radical change in capitalist social and economic structures, can come about democratically and without major violence. Whether it will or not depends less on the workers than on the means the capitalists use to defend their privileges at that moment when the long suffering majority have decided they have had enough.

Hints Classroom for Use

1. It is best if there is one set of "Class Struggle" for every six players. But if only one copy of "Class Struggle" is available, several people can represent each class. In this case, they can take turns throwing the dice, but decisions regarding strategy--whether to enter into an alliance, for example-should be taken by a majority vote of the whole group.

2. With several people representing each class, it is probably not possible, even with college students, to use TOURNA- MENT RULES. Younger students should stick to BE- GINNERS RULES, while older students (fourteen and over) should play by the FULL RULES.

3. Again, with several people representing each class: as classes land on the different squares, someone-perhaps the teacher-should read the explanations of why these squares offer the number of assets and debits that they do (See "Why?" section of this booklet).

4. As the game progresses from one level of Class Struggle to the next, it is a good idea-no matter how many play- ers are representing each class-to pause for questions and discussion.

5. Time should also be left at the end of the game to discuss whether what has been depicted is fact or fiction.

6. Don't let any of the above interfere with the students' fun and enjoyment of the game.

7. 7There are literally thousands of good socialist books and journals available these days, but those which I believe would be particularly appropriate to read along with (or after) using "Class Struggle" in class include-

   1. Dowd, Douglas F., The Twisted Dream: Capitalist Development in the United States since 1776 (Win- throp Publishers, 1974).

   2. Edwards, Richard, et al., The Capitalist System (Prentice-Hall, 1972). A reader containing many excellent articles.

   3. Huberman, Leo, We, the People, (Monthly Review Press, many editions). The simplest of the radical histories of the U.S.

   4. Huberman, Leo, and Sweezy, Paul, Introduction to Socialism, (Monthly Review Press, many editions).

   5. Katznelson, Ira, and Kesselman, Mark, The Politics of Power (Harcourt Brace Janovich, 1975). The best socialist introduction to the workings of American government.

   6. Marx, Karl, and Engels, Frederick, The Communist Manifesto.

   7. Marx, Engels, and Rius, The Communist Manifesto Comic Book (Quixote Publishers, 153 East Gilman St., Madison, Wis., 53703). Excellent.

   8. McLellan, David, The Thought of Karl Marx (Har- per and Row, 1971). The simplest of the many books of selections of Marx's writings. Contains a brief biography.

   9. Tressel, Robert, The Ragged Trousered Philanthro- pist (Lawrence and Wishart, London, 1976; and Monthly Review Press, forthcoming). My favorite socialist novel. Contains several simple explanations of difficult socialist ideas.

   10. Williams, William A., America Confronts a Revolu- tionary World: 1776-1976 (William Morrow and Co., 1976). Eye opening history.

   For teenagers, let me add, FPS: A Magazine of Young People's Liberation (2007 Washtenaw, Ann Arbor, Mich., 48104). Like it says. "CLASS STRUGGLE" NEWSLETTER Send us your name and address to receive one free copy of the "Class Struggle" newsletter, where we will print what you think of "Class Struggle", what we think of what you think, and what you think of what we think of what you think. CLASS STRUGGLE, INC., 487 Broadway, N.Y., N.Y., 10013.

Dear Friend, Please do not read the rest of this letter until you have played the game... Now that you have played "Class Struggle" (and if you have enjoyed it), you may be asking "What can I do to get 'Class Struggle' into the hands of more people?". The answer is that you can do a lot that is beyond the power of our limited distribution network. You can, for ex- ample, show your copy of the game to local book, toy, game, stationery, gift, department, and magazine stores, and encourage them to order. The back cover of the box tells most of the story. With you allied to us, we cannot lose...nor can you. Yours In Struggle, Bertell Ollman (For Class Struggle, Inc.) *If "Class Struggle" is not available in your community, individual copies can be obtained directly from Class Struggle, Inc., 487 Broadway, N.Y., N.Y., 10013 for $11.95 (includes postage in U.S.) (N. Y. residents add $.96 sales tax). Foreign buyers should inquire from their local post offices what it costs to send a three pound package by mail and send that sum together with $11.95 in a check drawn on a U.S. bank.

Game Board

I was thinking about Interpretability as a Natural Science when I decided I should review the old interpretability literature and found out that Chris Olah has already written about this in 2020. Also looked at 2014 topology interpretability, which gave a really good intuition of why we need activations. In this crossover episode, a really good blog post about the theory of batch size from our favourite proprietary trading firm. Anthropic published the superposition paper. An additional tweet thread of an exploration of superposition, and another about superposition in relation to Dettmer’s LLM.int8(). Human brains do superposition too! Neel Nanda’s favourite interpretability papers, I specifically want to review the ROME paper and this Sept 2022 ICLR submission.

Whisper speech-to-text from OpenAI, gosh, finally. Nvidia FP8 paper, also gosh finally 🥺. I’m trying to learn about GPUs, here is a really good microbenchmarking paper with unfortunate typography. Google multimodal model, PaLI. Chris Ré’s lab puts out another banger, FlashAttention. DeepMind sticks with the trusty naming scheme and comes out with AlphaTensor, of which the RL I don’t totally understand. I’m impressed though it seems like the matmul algorithm won’t actually be useful.

The most beautiful lawsuit website, which includes a custom font. Butterick occupies a special space of extremely admirable (he’s basically a superhero?) yet has a versatility where none of his works are quite perfect. I stan this Hilel Wayne newsletter on benefits of humanities in swe. Adept puts out a demo for their ACT-1 model. I didn’t feel carnally upset by Dall-E, but watching it use Google Sheets I was like “nooooo... my art!”

The Helvetica documentary is the best thing I’ve seen in a while. “It’s the real thing period coke period in helvetica period any questions? of course not. drink coke period simple” is my favourite quote, and literally got me to drink coke. It also discusses why Helvetica caused the Vietnam war, though I think it’s platonically socialist. An anime about Karl Marx commissioned by the CCP’s Marxism office (spoiler, he dies at the end).

Boot Boyz Biz essay on bricks, because I got the shirt and wanted to know about quirky Estonian architects. They also made a shirt (which I’ve acquired) that is perfect (includes a picture of the game “set”) so of course I like this essay on play, though I’d skip the situationist stuff. A copy of the Marxist board game, Class Struggle was purchased for me as a gift! I will document the gameplay soon, as the current literature of it does not do justice. It’s worth remembering now, that Monopoly is a fucked up version of what it was supposed to be. My favourite thing about Boot Boyz Biz essays is that they bother putting in source material that is somewhat non-trivial to locate, as they do in “Techne” with Stafford Beer works.

MMAcevedo, a short story about perverse immortality. Gwern with “It Looks Like You’re Trying To Take Over The World” 🖇️.

Unbounded problem-spaces and rationality. Dewey’s aesthetics. Ah yes, Georgism, here we go. I can’t believe I read the whole Dan Luu post on futurists, but I think it was worth it. Much better than Superforecasting (book). My favourite idea was that of Jacque Fresco, about how people of the future will apply scientific method to everything. Mao, Georgism and Uber. Golden Rule for Communication. On text as interface, or why I’m a grandma who really hates the idea of “text as the universal interface”

Refrigerator mother theory 😕 17 Interventions to Implicit Bias tests, none of which worked, but which I thought about of a lot for this blog post.

FTX publishes Possible Digital Asset Industry Standards, which includes the 5-5 standard, which is basically a standard to say that if you execute a security breach you can only keep 5 million? That is, unless people are losing money, in which case you have to return as much as they need. Look, I don’t know much about crypto, I can’t say I like them, but if they do this then they will have my undying admiration.

Some Money Stuff highlights (non-Twitter) include a deepdive into UK Pension Margin Calls, Barclays loosing track of it’s notes, Citi getting the 500M back from the oopsie it made a while ago, the shell deli, lost Morgan Stanley computers, Patagonia owners got sad about being rich in the best way. Also Elon has Twitter now and the Ethereum merge happened.

Cool Gray City of Love, Gary Kamiya 2013

I started reading this a several months after moving to San Francisco, because I asked for a book that was like “someone giving a tour of their suburban hometown”. The 49 chapters of San Francisco held a few moments that were more grand than anything you’d find about a suburban hometown. The Great Fire, the Gold Rush, the Summer of Love, the dot-com boom, the AIDs epidemic. History lessons are ok, but Gary Kamiya did also write about random grocery stores he used to go to that passed a while ago.

It’s a super romantic book! It’s full of sentiment, vibe and poetry. In describing the earliest failures to discover SF in the fog and the earliest settlements, Kamiya gave a really romantic characterization of the city’s complex and unique relationship with the ocean. For poetry, it of course gives little tidbits about the Beats but also about a magazine called the Lark. The Lark was a literary magazine written by a bunch of young people with a small but happy following, elaborate graphics and humorous but sophisticated essays, poetry, et cetera. This was just like, the best shit because this 1895 magazine feels like the epitome of writing that San Francisco produces today, whether it be the tweets, the Medium articles or the actual magazines.

The first edition of the Lark debuted a now-famous poem by Gelett Burgess, which is perhaps now more notable than the magazine itself.

I never saw a Purple Cow,

I never hope to see one;

But I can tell you, anyhow,

I'd rather see than be one

The last edition of the Lark contains this following. Present magazines that the Bay Area produces really lacks this sophisticated tongue-in-cheek energy.

Ah, yes, I wrote the "Purple Cow"—

I'm Sorry, now, I wrote it;

But I can tell you Anyhow

I'll Kill you if you Quote it!

I’m rather illiterate, so when I read I’m often googling about people or concepts or looking things up in a dictionary. With Cool Gray City of Love, my references were walking around the city, and pulling up a map whenever Kamiya referenced a specific intersection or a neighbourhood I forgot the location of. I didn’t like San Francisco Chinatown. It’s tacky, it doesn’t have particularly good food (Richmond is tasty though) and I went there once for tourism purposes and never quite set foot there again. The book describes Chinatown after the Great Fire, where the Chinese population fought to be allowed to rebuild in the place of their original homes. This is when the caricature of San Francisco Chinatown was built, eventually becoming a tourist attraction for the city. I don’t feel so much anymore that Chinatown lacks authenticity. Insofar as it’s one of the few places in North America where a Chinese population could determine from rubble what they wanted to build, it is exceedingly authentic.

The book ends on a chapter about Land’s End, describing the views of places that no one can quite reach, that even Gary Kamiya who spent his life in SF (partly as a taxi driver) will never fully discover the city. Excellent pairing for my months settling into SF, every walk around the city is more colourful and the memetic hatred of the city has certainly been cut.

There are probably few other cities for which such a book exists. Just the right amount of history such that you can go back in forth between benign flavouring and historical retellings. I tried to find one for New York, and ended up sorting books into specific neighbourhoods of New York (I don’t mean “Manhattan”, I mean, I needed a book for “Lower East Side”). Closest I got with NYC, was a book about rats and it sits on my shelf now.

Humps

I think there are broadly, three humps in discrimination, things that are in the way of us making further progress.

1. Different value systems cause different treatments

• Differing ethical systems often don’t cause different treatments, in which case there isn’t a problem. I find that the moral framework is not often the easiest thing to adjust, rather some more particular value adjustment can get people to agree on a treatment.

• A crux that often occurs is whether people discriminate differently if the group at hand choses their identifying feature. Another example is those with strong (traditional) family values have untanglable implications for sexism.

2. Deciding what to do given your morality is hard (some discrimination groups don’t need to be balanced to “equal treatment”).

• Dealing with racism and sexism is relatively easy, since all you have to do is apply “equal treatment”. Discrimination against non-equal treatment groups is much harder, as you have to decide exactly how much worse one group is under the infinite number of contexts to evaluate.

• The worst of this I think is classism. It’s really messy if you have probably reasonable values like “smart people are cool” or “contributions to society are good” but need to apply those accurately as not to be overly unfair to people. Secondarily, there’s even more calibration problem when there are conflicting values like “correcting imbalance” where you have responsibility to help people. Also in this category are parts of fatphobia, credentialism, and the outskirts of basically any form of discrimination.

3. Subconscious biases

• We already are at a point where we’ve noticed that this can do harm, but like, how the hell do we solve it? Our two options are to stop these from forming in the first place, or trying to gain sufficient control over our cognition to remove it. Both seem hard.

I think Hump 0 was actually a precursor to Hump 2, which was something like “we’re not sure who counts as human!”. The answer was something like “sure seems like we were doing some extreme motivated reasoning to convince ourselves that these people weren’t human” and now we’re more careful about it. Hump 4 might be something like “we’ve reached the ends of what systematic solutions can accomplish, and we simply need people to stop being assholes”.

Discrimination against ugly people

I’m increasingly convinced that this is the most terrifying (as in, upsetting on moral grounds and not necessarily the one that does most damage) discrimination we perform as a society. Naively, the only excusable of lookism is in romantic/sexual endeavours, but it’s all wrapped up into things like fitness, fashion and general care.

I do still think it’s a massive historical injustice (and especially sad one as it’s largely uncontrollable by the individual), and is fueled by something quite monkey brained. We should be far beyond treating people based on perceived mating value, not to mention that appearance has much less to do with mating value than our monkey brains would have us believe.

Yet, it’s mostly normal (though a little untasteful) to comment negatively on someone’s unattractiveness as a reason to be uncharitable. It’s often not viewed as untasteful at all if it were in the context of romance or sexual activity, even if the person was being excessively mean. This registers in peoples’ careers, crime enforcement, social activities and we are rather upsettingly numb to it. Too much “that’s just the way the world works” attitude.

The more terrifying thing though, is that we might be making excuses to never acknowledge that we do this, by offloading it into “they don’t put enough effort into themselves” or when people mix up “creepy” with “ugly”.

Utopia

My definition of discrimination is probably broader than most, I define it as a sort of excess bad. It’s encompasses all forms of unjustness, but is not prejudice-free. I think utopia is a place that does not hav discrimination.

My utopia is not a place where we never do bad things to each other, I think that would require compromising the differing values we have as humans. My utopia does involve people acting in accordance with their own values, which will often be similar to other peoples’.

This goes wrong quickly. “I value murdering people who don’t have blond hair and blue eyes”, it screams! Surely we can’t avoid calling this discrimination. I would claim that this is a bad mesa-value. The underlying value would be something like “I value making other groups seem lesser to mine”. Under that, “I value feeling safe and supported by a community”. Presumably if this were you, and an omniscient being traced your values for you, you’d say “oh ok the core value seems fine, but the downstream values seem inefficient, or conflicting with other values I have, like that murder is bad”.

There are some reasonable mesa-values, like “art is great” → “I value people with artistic talent”, and it would be ok to treat those people better. Maybe you really hate people who are lazy, and you might be mean to those people which feels less obviously ok, especially if the laziness identified has some chance of being mislabeled (as it often is today). The line may be a little fuzzy and I’m not sure how we’ll ever be able to clear it up, though we’re currently far from it.

Utopia, will in some ways remove biases that we don’t currently have the cognitive capacity to. Some discrimination has been solved by us thinking harder (or new evidence being provided), and realising we’ve been bad. But what about implicit biases? Perhaps much further down into the future, we’ll live in a world where society doesn’t create implicit biases against black people.

Unless the utopia involves significantly changing the way humans generate emotion and interact socially, I suspect we’ll still be very capable of being discriminatory. We have tried to create treatments for implicit association tests, though it was comically unsuccessful. The only things that worked were basically cheating, and creating racism against white people. But what a world it would be if we had such treatments! “Doctor, I’m feeling a little racist, can you fix it?”

Jacque Fresco once predicted that humans would eventually become free from bias and apply “the scientific method” to their lives. I think people who value artistic talent should have room to treat those with artistic talent more kindly, to want to hang out with them and whatnot. Running around stoning those who can’t paint is definitely causing excess harm. In the giant gray area though, how do you decide? Superior cognition will probably be necessary in a discrimination-free utopia.

Good will is no longer enough, we need good judgement

Even given identical moral values, epistemic processes will start showing up here. Say we’re thinking about discrimination against pregnant, skilled, women in the workforce, and two people agree that these situations should be treated with a strong utilitarian ideal — they don’t care about the pregnancy itself, only the performance of an employee. The case here could be one person reasons to believe that generous maternity leave to allow for the employee to be kept long term is a good option while another believes that it would be easier to replace the employee. The difference is just in their world models, and well, probably one of them is smarter than the other.

But I think we (at least in the western blue states) are at the fringes of what we can solve by being a good person, in that the easiest ways to improve are through being smarter. This is primarily because I think a majority of problems in discrimination are being blocked by Hump 2. It’s very much a judgement problem, since it’s all about calibrating the optimal treatment for non-equal discrimination classes. Classism, credentialism, what’s remaining of sexism, lookism, ableism are all mostly hump 2 problems. We’ve certainly tried to work on these things, but I was kind of disappointed with how the anti-fatphobia was executed, though it seems to be going better for ableism.

I’ve considered that the problem all along was good judgement. The strongest argument for this is that moral progress in the form of ideas passed along generations seems a lot more plausible than generational increase of good will. Those ideas are the product of cumulative judgement. The ideas passed down though, was not so much an information passing of “we learned something new, turns out other races have moral value” but more “we should think more about whether we’re treating people fairly”. As far as I’m concerned, having that idea passed down is generational increase of good will.

There is already a lot of evidence that we’re at the new frontier. For example, we now have bias tests for people who want to do good, but are worried about subconscious biases to try to correct this. Building the test itself, is a demonstration of good judgement. The whole nature of this “I want to do good, but I need to think clearer to be able to exercise it” is a new practice for discrimination, at least as a burden that everyone will carry.

It’s certainly not news that smart people are often kinder or carry moral values better in various ways, but I think it’s kind of imperative for discrimination now. This is scary, as it takes away our Forrest Gump narrative that an extremely kind man can be rather dumb.

Some half baked thoughts

I started listing instances of discrimination in my head, and there are some outliers like fatphobia and anti-semitism, but the one that updated me the most was the Cagot. There’s a canonical ideal of discrimination which involves very systematic, long-term harm, often caused by religion or some sort of historical baggage. The Cagot were a group of people who had no notable ethnic or cultural differences, and were severely discriminated against — forced to wear goose feet on their clothing to segregate themselves. They were not allowed to marry non-Cagot, were buried separately, and were not allowed to even work with livestock (this varied between regions, hated of the Cagot was widespread).

There was no consensus as to why we hated them so much, but they do in fact no longer exist. They practiced the same religion, spoke the same language, and were not an ethnic group. We don’t really know why we hated them so much, but accusations included werewolves, lepers, cannibals or just being intrinsically evil.

My update from thinking about the Cagot, is this model that discrimination starts as people being mean, making excuses to be mean, then letting the burden of history carry it forth.

Why that update is important, is that it changes my model of how we can make progress. The reason what happened to the Cagot can no longer happen today is that globalization and the progress on discrimination holds us accountable. At the same time, it seems squarely the case that if discrimination comes from being mean, then we might not be able to fully eliminate it until people are no longer mean. I’m pretty sure people haven’t become less mean? It’s weird that it seems like society has made moral progress where individuals have not.

Another angle to the “being mean” is that all discrimination cases are witch hunts. The literal witch hunts were discrimination. Witch hunts express why people are mean, which is broadly because people feel the need to have an outgroup, and will create it when it does not exist. I’ve started reading about this aspect of human behaviour, why we need the outgroup (enemy group) and why it’s so hard to exist in neutrality, specifically from the lens of political theory (which seems to produce more useful/correct models than sociology). This is of interest since it seems relevant to how discrimination occurs and if it’s even possible to ever solve without mutating this basic human desire.

The energy of discrimination as a witch hunt shines through with anti-semitism, given the Jewish migration across Europe (and eventually to New York and the rest of the New World). An oversimplification of the situation is something like, the Jewish were forced out of one country then travelled to another, only soon to be evicted from their new homes as well. It wasn’t literally the same people chasing them around, but it might as well have been. No discrimination has geographically travelled quite like this (I think the Crusades might be an example, but that has all its other complications [war]). Imagine fleeing Hungary to Germany due to anti-semitism in the 1920’s and seeing it follow you to Germany like watching a storm coming in (the Italy/Mussolini situation was also peculiar).

Anti-semitism is also extremely special in how it affected the Jewish narrative, I don’t think any other discrimination has embedded this much historical impact into the psyche of a whole group. Another weird case of this though, is the commercialisation of gay and black culture as those humps in moral progress were passed.

I have a bunch of thoughts about how various discriminations are funky, this was a surprisingly strong nerd snipe. I think fatshaming was special because it was the first time that we sort of had consensus that we might’ve overcorrected? I think credentialism is super weird, since we actually put up very discrete systematic barriers against the uncredentialed, unlike most forms of discrimination which are discretionary. I’m still kind of confused as to how we’ve let lookism run so far and model minority based discrimination seems rather new? I haven’t even started to think religion!