Category: speedwrite

1605 words, 82 min

So November has come and gone — faster than expected. Recent discoveries have found that telling AI models to “be smarter” makes them smarter. Perhaps humans work the same way. Tell me “write faster,” and I will write faster. Over the course of the speedwrite, I published 18 posts for a total of ~20K words. Fewer posts, but roughly about as many words as I would have hoped. It’s much harder to measure time spent, but RescueTime claims that I spent ~40 hours on Notes in November, where I do all my writing. If you believe that, this means that I wrote at an average speed of 8 words a minute, or about a tenth of my typing speed, which feels about right, plus or minus a large confidence interval.

The first observation is that I have newfound respect for Matt Levine (and other regular columnists). Matt Levine writes a 5000 word newsletter better than anything I have ever written, and he writes it almost everyday. If you told me to rewrite Matt Levine’s newsletter on the exact same topics (so I don’t need time to do research), it would take me more than the entire workday (10 hours) and the result would be a pile of dogs*** in comparison. Once upon a time, I played with the idea of trying to become Matt Levine for AI. There is no chance of that happening in the near future, which directly implies that there is almost no chance of it happening at any point in the future. 

Quality of posts certainly went down as expected, but I think quality/time went up considerably. Past posts have taken something like 10 hours each on average (at least) and so I think it’s not an exaggeration to say that quality/time has improved by 5-10x. And a significant portion of the quality drop is just me skipping the editing passes during speedwrite, rather than some other insidious form of error. I think that’s something I should do in general. It’s my own semi-private anonymous blog after all!

A related goal is simply confidence. Before, I would often feel a certain hopelessness with writing in that I would be very excited to think/write about a topic, but my writing speed was sufficiently slow that the topic would escape me before I could pin down a draft in a reasonable form. And so my drafts folder would be littered with semi-coherent pieces of writing too incoherent to be called a draft. Each one a sign of my failure. I was happy to say that I got some of these out. Some of these “fragments” had quite literally been sitting unloved for years (keep in mind this blog is only 2.5 years old) and they finally saw the light of day! So hurrah on that goal achieved! 

The obvious critique is that I failed at my stated goal of publishing something every single day no matter what. Surprisingly, I did not fail because of writer’s block or running out of ideas, which were the two things that I would have guessed would give me the most trouble. In some sense this is not unexpected. I didn’t fail from the two problems that I foresaw because I specifically took countermeasures to prevent myself from failing those. I didn’t run out of ideas because I was referring constantly to my old scrap pile of discarded, half-finished ideas. And I didn’t fail writer’s block because I had specifically pre-committed to publishing garbage, if that’s what it came down to.

What I didn’t foresee (or at least, didn’t take enough countermeasures) was that writing time and energy would cannibalize everything else so hard. Like I didn’t get nearly as much research work done this past month as I hoped. 5-10th percentile outcome, I think, which is especially bad compared to the 80th percentile level I was at right literally days BEFORE I started speedwrite November. This cost a lot. It wasn’t a direct one-to-one cannibalization, but I think it was roughly two-to-one, in that every two hours I spent writing directly stole an hour from AI research work. Here’s hoping that AI research can cannibalize that time right back in “Research December.”

And of course, mental health (though I already knew that). The first set of failures to publish were literally me not having the mental energy to copy and paste a finished draft from Notes over to WordPress and pressing publish. This is not good. I fixed it by prescheduling finished posts, so the second set of failures was actually a lack of drafts (which occurred after several days of zero writing).

Also, a more minor complaint is that at the beginning of November, I made a list of topics that I really wanted to write and think about, and I didn’t get to most of them. My thoughts on AI safety / timelines / ethics. A big series on risk that got derailed largely because the topic soured on me after SBF/FTX. A few old incomplete drafts on various topics near and dear to my heart that just needed a little more nudging to get over the finish line. Of course, this isn’t an accident. In the past, I have found myself reluctant to work on the most important things and instead do something I know is not important.

I never quite understood why until I read a section of Richard Rhodes’s The Making of the Atomic Bomb on Niels Bohr. Apparently, Bohr, a physicist of such magnitude that even among the pantheon of 20th century physicists he is second only to Einstein himself, was afraid of publishing his own work because he feared that it would be judged. And judged, I’m sure, not merely by other physicists, but more importantly, by himself. You see, when anything is in the “in progress” stage, you can just claim (to both yourself and others) that all the flaws are just things you haven’t gotten around to doing yet. You can’t do that when you “finish.” When something is concrete, it can be criticized. It can be wrong!

If work is never finished, its quality cannot be judged. The trouble is that stalling postpones the confrontation and adds that guilt to the burden.

Richard Rhodes, The Making of the Atomic Bomb

I remember laughing after reading that section, laughing very hard, for I saw precisely why I dragged my feet for so long in finishing something. If work is never finished, its quality cannot be judged. And that’s precisely why it must be finished [1].

But maybe that’s alright. After all, the initial example that I set forth in the first speedwrite post was trying to take 100 pictures without caring about whether they were good or not. Perhaps the same thing applies to what you try to take a picture of (or write about). If the goal is to sharpen your tool, any tree will do, and perhaps it’s even wiser to leave the more important trees for after your axe is sharp. I suppose that means that now that I’m not in speedwrite November, I should either ban writing entirely (so research can cannibalize the time) or at least restrict myself to writing ONLY about the most important topics and see what new ideas I can fell with my newly sharpened axe.

That being said though, I actually think there are still more gains with even more speedwriting. Of course, the time/quality gains are not going to 10x again. At some point my typing speed is the limiter and while we aren’t there yet, we certainly don’t have too much more than another 10x to go. If I were to run it back (and I sort of want to, after I get my time management figured out) I would want to do the following.

1. Have an idea
2. Set a timer for 5/10 minutes and draft up an outline.
3. If it’s a bad idea or incomplete, put it back in the drawer for now.
4. If it’s good though, set a timer for 30/60/90 minutes and FINISH THE DRAFT IN ONE GO.

Before speedwrite November, the chance of this succeeding was ~1% — only if I had the idea fully fleshed out and felt super energized to write about it did I even have a chance. Now, I think it’s about 80/20 — I can usually finish a reasonable-ish draft as long as I feel very excited and have a decent direction after the outline.

But the 20% I still fail is not a random 20% — it tends to be most important 20% of topics and ideas. Perhaps it’s nerve like Bohr. Perhaps it’s skill, as those topics are certainly trickier after all. But the goal for a run-it-back speedwrite is to very explicitly get to 99/1 — a full reversal of old [memorymancer]. There, I would say: anytime I want to, anytime I have a rough inkling of a thought, I can convert it into a draft that would communicate it to my satisfaction. Right now there are two barriers: writing and thinking. At that level, there is just thinking. If I am unable to write down my thoughts, then my thoughts are unclear.

Some people say this is true in general, but I think it’s not. They underestimate just how much friction the writing process adds. The best “proof” is the last month. Did I become a substantially clearer thinker in the last thirty days? I think not. But I definitely 10x-ed by ability to get an idea on paper. Maybe after round two, I’ll clear the writing bottleneck entirely.

And wouldn’t that be a nice place to be. 

[1]: I note, with some irony, that I have not yet finished the book: “The Making of the Atomic Bomb”

2.5 hours, 849 words

So the last few days have been quite shaky, both in terms writing and in general. I’ve missed a few days of publishing, and surprisingly (at least to me), it’s not because I have ran out of things to write about. Not by a long shot.

The truth is that I haven’t been feeling very well, mainly due to the fallout from FTX. This is ironic, since the first few days of November when this started, I was not feeling well due to being sufficiently ill that I barely left my room. The last week, I’ve had basically complete drafts on the queue that I have simply failed to take a few minutes publish. And no, this isn’t an SBF “I swear I had the drafts, I just couldn’t get them out in time.” I actually had the stored up drafts but somehow not the energy to cross the finish line.

I frankly haven’t done much of anything this past week other than “float.” They say that “true artists” can’t wait for inspiration. They just write. Well, it was abundantly clear to me even before I started (and it’s still clear to me now), that I am not at that level yet.

Anyways, FTX. The SBF/FTX story was the first time where I felt a story worm inside my OODA loop. New information invalidated the story in my head before I could fully process the old information. For some people, this happened with COVID, but I escaped because I didn’t keep up with the COVID news (clearly I did not learn that lesson very well).

I think the meta-lesson is that even after recent attempts to add noise to my beliefs, I …. still need to add more noise to my beliefs. Anyways, if you aren’t heard, here is a quick summary of my beliefs. I won’t give you the usual caveat that these beliefs might be false, because I can tell you with moderate confidence that they are probably false. There is no complete summary, anywhere that I have read. And I have read quite a bit, far too much in fact. Anyways, my thoughts went from:

SBF messed up by failing to ensure enough liquidity and got crunched -> SBF messed up by leveraging too hard and being insolvent because of recent events -> SBF had been playing fast and loose for a while and it finally caught up to him -> FTX was always a pretend house of cards and the only way they did flashy things like Super Bowl ads, sports deals, and venture capital was because he was taking customer deposits and treating them as profits without any regard to paying them back -> SBF was deeply and utterly corrupted from day 1 and (some) leaders in the EA community (the ones who knew him) supported him because they felt he was their best bet for acquiring power. -> ????????

By the time you read this, it might be wrong again, at the rate things are going. I am honestly not sure where the truth is anymore and think it is neither productive nor beneficial for me to keep speculating. So here is the end for a while. 

I’m not directly in the line of fire, at least right now. I have no direct or indirect financial stake in FTX, and frankly, I’m fairly shielded from secondary effects as well. In fact, while it’s too early to say, there is some very strong chance that the downstream effects end up very, very much in my favor. But there is some impact on my mental state, and this is me trying to puzzle out why.

I think it’s mainly the weight of seeing that someone who you thought was good for the world actually turn out to be basically …. a cartoon caricature of evil. There is a sense of despair that comes with revelations like this. It’s stronger for the EA community, for those who viewed SBF as a hero. But it’s still true for me, who frankly did not. There is more, much more, that I haven’t quite dug up yet.

Perhaps an analogy will suffice. [redacted] told me a short while ago that she was shocked by hearing about a cheating scandal involving a popular comedy group known as the Try Guys. One of its members, Ned, was held up as the paragon of virtue in terms of his relationship with his wife. Like that was his whole public identity. Utterly, utterly, utterly devoted to his wife. His nickname on Youtube was apparently the “wife guy.” So when in hindsight, it turned out that he was cheating on his wife the entire time and that it was a semi-open secret among people in the know, that was, pretty devastating to people in that community. Finding out everything you have known about someone is smoke and mirrors is always rough.

At the time, I, never having heard of the Try Guys, simply said: “well, not everything is as it seems.” Ah, so it turns out that I clearly did not understand that lesson myself. 

Writer’s note: this was partly written before November.

If you want to improve at Go, there are many potential directions you can go (heh). Some people advocate learning josekis (common opening patterns). Others say you should work on more meta level skills like judging who is ahead in a given game, and using that to inform how risky of a strategy you use. Go is composed of a bunch of mini-skills that are almost all largely independent of each other. Your overall strength is a combination of them. It’s not an equal weighting though. I’m biased, but I think that the best method is doing life and death puzzles.

Go is fundamentally a war game and if your soldiers are better at fighting, you will win. It doesn’t matter if you are outnumbered. It doesn’t matter if your enemy has the high ground. It doesn’t matter if you positioning sucks. If one of your spearmen can take down ten of your opponent’s, it’s going to be pretty hard to lose, no matter how bad you are at everything else. In terms of my own personal progression, Personally, I got to shodan (roughly black belt) exclusively through fighting strength, and six-dan (one below max rank) with fighting strength + endgame.

Personally, my own journey went as follows. I learned to play around the age of five, apparently because I wanted to win some big shiny trophies (this is what my mother tells me — I have no real recollection of this). The first three years that I played, I don’t believe I ever won a single tournament game despite playing in the novice section. This is because kids tournaments have severe issues with sandbagging and I was not really playing against novices in the novice section. After some time, I started learning from [redacted], who ran this wonderful daycare / Go school hybrid program. 

[redacted] had a wonderful model of how to teach children Go. For one, it was fun. For two, he offered prizes. In hindsight, the prizes were simply Pokemon cards bought in bulk for a few pennies each, but to a kid they were everything! I wanted Pokemon cards and so I did a lot of life and death.

In hindsight, [redacted]’s Go school was too cheap (he eventually quit to become a software engineer, and who could blame him). But if I recall, [redacted] charged like 40$ for my brother and I for an entire day? My mother would just drop us off there Saturday morning and not worry about us until evening. I think he probably could have charged an order of magnitude more money than he did. Of course, had he done that, he wouldn’t have had any clients, but if [redacted] ran a Go school now and charged that price, I would pay for my sister to attend. In terms of personal benefit accrued, it’s probably even an order of magnitude more than that.

[may be continued]

1272 words, ?? min

Let’s start out with a softball. An unpopular opinion that I hold is that urban biking is obscenely dangerous. It’s possible I am a worse biker than other people. I won’t deny that honestly. But I don’t think that’s the whole story, not even close. 

Of course, how dangerous biking is various enormously based on where you are biking. For example, biking at Stanford was super safe. The few cars that drove there understood that they were second class citizens. They would rive very timidly and basically always yield to you regardless of why had the legal right of way. Also, most places didn’t have any cars anyways. In my four years there, I had no bike accidents on Stanford campus despite biking pretty much every day. In fact, I don’t even recall any close calls (on Stanford campus that is — I had close calls while at Stanford, but they occurred off campus). 

Harvard is a different f**ing story. The reason is that Harvard is essentially a bunch of fiefdoms that are geographically and bureaucratically completely separate. Bureaucratically, you can see this by noting that whether or not you have free access to Harvard’s gyms depends on which school you are in. My friend at the divinity school has to pay for his gym membership. At SEAS (my school) it’s free. The business school gets access to all the normal gyms AND it’s own gym which is exclusively open to its own students. A similar story can be said for printing. I can print for free at any SEAS building, but I have to pay to print at any other printer at Harvard. 

Geographically, this causes some issues as well. If you bike say, from Stanford Econ to Stanford CS, you will cross only “Stanford” regions along the way, and there will be very few cars. If you bike from Harvard Econ to Harvard CS (roughly the same distance as the analogous journey at Stanford), the vast majority of your journey will just be non Harvard affiliated Boston/Cambridge and you will likely pass a hundred or so cars on the way (or rather, they will pass you). 

Obviously, biking is much safer if there are no cars nearby. If cars cannot be avoided, it can be made safe via a protected bike lane with a divider and/or buffer between you and the car.

Barring a huge detour, there is only one viable bike route between Harvard Econ and Harvard CS (over the Anderson Memorial Bridge). It is not a safe route. There is a bike lane, but it is very narrow and there is no divider between the bike lane and the regular car traffic. The bike lane is lined with drainage gutters and manholes which are unavoidable because the bike lane is not wide enough. Worst of all, the road is right next to an exit ramp for a parkway, so cars will regularly zoom past you at only slightly slower than highway speeds.

Not long ago, I was making this journey when a semi zoomed past me at 45 mph, mere INCHES to my left. If I slipped even a tiny bit off the bike lane, I’m a dead man. If the truck driver loses attention for a split second, I’m a dead man. It’s a game of inches. Scared the f***ing sh*** out of me. 

Another time, I was biking through Central Square in a bike lane not too different from what is pictured below. It was a “sandwiched” bike lane where street parking is to your right and the regular road is to your left. These lanes are dangerous for fairly obvious reasons. That day, one parked car did not look for cyclists before exiting his parking space, and I only barely managed to avoid a crash by swerving around him.

A not so safe bike lane in Boston. This is an actual bike lane very close to Harvard. Luckily for me though, it is not a route I take very often. Source. 

There is something so surreal about this situation. I think it is the juxtaposition of my own perspective that I’m risking permanent injury on my everyday commute with the common perception that biking is super safe. I feel like a quack every time I talk about this, like I’m concerned that my paper cut is going to somehow contract gangrene and require amputation or something. It is the feeling of shouting into the void, with no one listening to you. So I understand why Eliezer Yudkowsky often sounds so annoyed, even if I personally disagree with him.

The practical result of these incidents is that I often feel a strong reluctance to go somewhere for no reason other than a fear of biking. And when I do go, I either take a huge detour to use a safer route. Either that or just bike on the sidewalk. The main downside is that lots of people yell at me — at least once a week. It’s quite annoying. The fact that Boston cycling law actually recommends that you “ride your bike on sidewalks when necessary in the interest of safety” is irrelevant. Strangers do not yell at you because you are breaking the law. They yell at you because their culture tells you that they can (and should). 

Perhaps worst of all is that it’s not even clear that Boston is especially bad relative to other cities! For example, the bike lane between Harvard Econ and Harvard CS I complained about is a pretty normal bike lane, the kind you would expect to see basically anywhere. 

In some sense, this is the whole crux of my unpopular opinion.  “Normal” bike lanes are unacceptably dangerous for bikers. And don’t even get me started on the abnormal ones. Since very few cities have protected bike lanes, urban biking in general is dangerous enough that I would stop immediately if I had a better alternative. I have to get around somehow though! Boston is actually better than some cities in that it has a mix of safe and unsafe bike lanes. But like any chain, you are only as strong as your weakest link. 

Perhaps you may think I’m overreacting, so here is some additional anecdotal evidence. I know only around ten people who bike regularly at Harvard and three have them have been seriously injured in the past year. Like go-to-the-hospital level injuries. The most recent biker fatality at Harvard was only two years ago. So honestly, I think I might even be on the lucky side to only have two close calls (and a minor accident, which was my fault) after one year of sporadic biking. Compare that to Stanford, where I biked everyday and had no trouble whatsoever!

If you are still not convinced, let me give an argument via transitivity. Motorcycling to work everyday is well acknowledged to be extremely dangerous. I’d argue that biking to work is even worse! This is because most of the danger is how many inattentive drivers you are exposed to and thus, going slower is more dangerous than going faster here.

Truthfully, I find this entire situation quite ironic. As a child, my father was terrified to let my brother and I bike around the neighborhood, and we always thought he was overreacting. In some sense he was: my old neighborhood is quite safe, car wise. But nigh on twenty years later, I have grown up and now I’m the one terrified of biking. I suppose we all find ourselves becoming like our parents as we age.

The finale of the trilogy began by The Catholic Church Was Right All Along.

My freshman year at university, I took a delightful class on the Western canon. One of our readings was the poetess Sappho, who in her day, was said to rival even Homer himself in fame. Sadly, Sappho is no longer as popular largely because the vast majority of her poetry has been lost to time. Of all her work, only one complete poem survives. For the rest of work, we have mere scraps, sometimes just a single word. In the modern day, Sappho is known for something else entirely. Wikipedia describes Sappho as a “symbol of love and desire between women, with the English words sapphic and lesbian deriving from her name and that of her home island [the Greek island Lesbos] respectively.”

In office hours one day, my section leader told me a story of a lecture he had once attended on historical perceptions of Sappho. It turns out that the view of Sappho as the iconic lesbian is actually a very modern view. For example, the Greeks thought that Sappho was a lover of men, not woman. Legends of her death say that she was in love with a ferryman who did not return her love. Distraught, she threw herself from a cliff in despair. When the Romans came to power, a different view of Sappho emerged. Her sexuality was derided and she was called “a courtesan, not a poetess.” Still later, in the Middle Ages, she regained her reputation as a scholar and extraordinary poet. And finally in the Romantic area and onward, her fame has grown to the point where her home island Lesbos is perhaps the most famous island in Greece, in some sense at least. 

The theme of the lecture was that Sappho’s reputation, instead of being based on her actual life, was actually every culture imbued their own beliefs into her as a symbol. What they said about Sappho said more about themselves than it did about Sappho. And as the lecturer came to the present day thought on Sappho, my section leader grew ever more excited. He was ready to hear the scathing critique that our view of Sappho as a symbol for sexuality and love was just a reflection of modern values, no closer to reality than any of the previous views of the long dead, long lost poetess. But he did not. The lecturer claimed, with almost complete irony, that the modern day interpretation was the objective, correct answer.

Jesus preached a similar idea, much more concisely: “Why do you look at the speck of sawdust in your brother’s eye and pay no attention to the plank in your own eye?” It’s a lot easier to see someone else’s blind spots. So, turning that knife inward instead of outward, here are three changes that I will try to implement in my life.

1. Try not to make fun of people for their blind spots. This is hard, since I actually really like cracking these kinds of jokes. And honestly, sometimes it’s think it is actually alright. But I’ll try to reduce the amount. And even better yet, try to guide them to the light, as much as a mostly blind man can.
2. Acknowledge that I have blind spots I cannot see. I have eliminated a large number of blind spots over the last few months. Nevertheless, I am certain many more remain, and I do not know where they are. Unknown unknowns. A gap in an area that you think well guarded. These are the most dangerous kinds of mistakes. Of the seven sins, pride is the one I struggle with most — by far. So admitting this on a deep level feels hard for me, even while another part of me knows that I must do it.
3. Notice my own confusion. I recently spent some time implementing several famous reinforcement learning algorithms from scratch. I’ll probably write about that experience at some point. One important lesson is that when debugging RL models, you need to be extra vigilant towards anything that causes confusion. This is because RL models tend to fail silently. If anything, and I mean anything, smells funky, your first priority is to resolve it. If you instead follow a “if it ain’t broke, don’t fix it” mindset, you may find out a few hours before submitting to NeurIPS that your new trick for making RL agents learn 100x faster was to accidentally turn on a flag that changed the environment name from Pong to Cartpole.

A word of warning: it is easy to fall into the trap of getting “addicted” to finding the truth, so to speak. The goal (or at least my goal) in life is not to collect the largest set of true statements. There are more important things to do. Knowing the truth is often very helpful for these goals, but to me, not an end into itself. One must not lose sight of the true goal.

Recent changes in my life have triggered some serious reflection on my fundamental beliefs and have led me to discover than many of my adjacent-to-core beliefs were actually totally false. At first, I was extremely distraught, but now I feel better about it. After all Taleb tells us that: 

A loser is someone who, after making a mistake, doesn’t introspect, doesn’t exploit it, feels embarrassed and defensive rather than enriched with a new piece of information, and tries to explain why he made the mistake rather than moving on.

Antifragile

In one of my favorite childhood stories, a farmer is carrying rice home from the market. Halfway there, he discovers there is a hole in his bag and that he has been leaking rice the whole walk back. Instead of being enraged, he is quite thankful that there is still some rice left in the bag. Similarly, one can bemoan the fact that I have held incorrect beliefs (that have led to incorrect decisions) or be thankful that I have discovered them with (hopefully) time still left to do something about it.

I will try to write about some of these blindspots as part of speedwrite November. Depending on your own levels of calibration, it may very well be that most or even all of these ideas are already obvious to you. That is good! On the other hand, if you read these posts and think: “AHA, [memorymancer] has stumbled upon the one and only Truth. I must follow his vision for the future!” then you’ve got it all wrong. But I hope that some of these things will add a few error bars to your calculations. 

Four unimportant facts to serve as an apéritif. Traditional Italian cuisine does not use tomatoes. Sugar does not (chemically) make kids hyperactive. Cats cannot digest milk. And of course, never let anyone convince you that the Earth revolves around the Sun.

Part 2 of the trilogy began by The Catholic Church Was Right All Along

Perhaps you think that the observation that the reference frames are arbitrary is a cute fun fact from physics. I claim instead that this seemingly unimportant fact actually hides something very important.

Think again about what had to happen for this to occur. At face value, the Earth is obviously not moving (from our frame of reference). Somehow, the education system managed to convince everyone of a fact that seems very, very, very, very clearly wrong. And not only that. It managed to brand everyone saying the obviously correct thing as an idiot.

Crucially, it did NOT do so by explaining the underlying mechanism behind the mistake. Rather, the indoctrination worked by convincing us that anyone who holds a certain set of ideas is a stupid, religious bigots that deserve to be made fun of. And because no one wants to be made fun of, everyone dutifully recites the party line. Someone who believes that Earth revolves around the Sun? They are OK to make fun of. Believe that the Earth is flat? OK to make fun of. The universe is only 10K years old? You get the point.

This is important enough that I need to repeat it. In some sense, modern education makes us even DUMBER in many ways than before. We are taught to distrust obvious logic and observation and defer to what other “smarter” people have decreed to be fact. We don’t teach the truth in schools so much as we teach the idea that modern humans are smarter than the ancients — and that it’s OK to ridicule the ancients for their idiotic beliefs. And sadly, it can’t really be otherwise, at least not without substantial change in the education system. Would most high school physics teachers even understand the description in the previous post [1]? 

I am quite glad that, more often than not, society as a whole gets the facts right. But we certainly don’t get them right all the time. And perhaps more importantly, we most certainly do not get them right because individuals in modern society are that much “smarter” that the ones who came before us. The culture is far “smarter” (at least I believe so). The individual humans only marginally so (and maybe not even).

This “OK to make fun of group” is really quite dangerous. For example, I think it is one of the (deliberately engineered) reasons that it’s often hard for people of different political beliefs to get along. Let me give you some concrete examples.

I recently joined the Harvard Standup Comedy Club. At one of the first shows of the year, there was a huge scandal where someone told some very socially unacceptable jokes. This resulted in some new rules being laid down about what was acceptable and not acceptable to joke about. A new member then asked if it was OK to make Catholic jokes. The club president looked at her quizzically and said: Obviously Catholic jokes are OK. Why wouldn’t they be OK? Of course, the red lines would be different at a comedy night at your local Christian youth group. What is perfectly reasonable in one circle is fair game for ridicule in the other.

Or take cross-field collaboration in academia. In computer science, adding unnecessary math to make your paper look more technical is, while sadly a relatively common occurrence, sufficiently uncool that no one would ever dream of bragging about doing so. It is definitely in the OK-to-make-fun-of category for CS people. On the other hand, I recently overheard a conversation between several economists who were telling me that in their job market paper (Economics PhD students usually write only one real paper in their entire PhD). According to this economist, it was not only standard practice, it was openly encouraged to shove as much advanced mathematics into your job market paper regardless of its utility to the research question at hand. This is because the job market paper is primarily used as a resume to “show off your technical skills” and only secondarily as a contribution to the research literature. After all, if you get the job, you have your entire career ahead to do real research. Again, this was only an N=1 example (well N=3 if you count them individually), but was one of the first times where I finally put my finger on why cross-field collaboration is hard. It is very common for one group’s OK-to-make-fun-of category to overlap with another group’s you-should-try-your-best-to-imitate-this-example category. It’s essentially a difference of (arbitrary) norms. And this leads to friction that you often don’t even notice before the disdain sets in.

Most conspiracy theories are moronic. But you can see what happens right? Modern society tells us all sorts of lies. Some innocuous — others severe. A clever, but ill intentioned person can go around revealing a small subset of the truth to lure people in before using their newfound credibility to implant ridiculous ideas into their follower’s minds. At first, people will be on guard. They will investigate any claims made. But then, they find out that their Prophet is right! They start paying more attention to what he says. They go to group gatherings where they meet people who angrily announce that the world has been lying to them all the time. They make fun of people who used to think what they think. Slowly but surely the exact same effect as the geo/helio-centric model takes place, except with something more dangerous. Before you know it, they are going full on QAnon, microchips-in-the-vaccines, or AGI-is-coming-in-18 months (15 now, that was three months ago).

The vast majority of people who believe any of the above do not do so because they have thought things through from first principles. Because thinking things through from first principles is hard! Instead, they believe it because, for one reason or another, they fell into a group of people who told them that people who didn’t believe in the Truth were sheeple and seriously uncool compared to the people that did. Thus goes the story of humanity.

[1] A physics degree is not required to become a high school physics teacher in the United States and the majority of physics teachers (as of 2009) did not study physics (even as a minor) in university. I was personally lucky that my own high school physics teacher had a PhD in physics. Sadly, he left the year afterwards and the rumors say that his replacement didn’t know a thing.

The first of a three part series. The initial fodder for this piece was written before November.

When I’m in a whimsical mood, one of my favorite activities is telling people that the Catholic Church was actually right all along. The Sun really does revolve around the Earth. Imagine me starting a new Youtube channel. “Harvard STEM PhD Offers Strangers 50K Cash to Prove That Earth Actually Revolves Around Sun.” 

No, I’m not trolling. Well, I’m obviously trolling, but I am also 100% serious. Here is a coherent, more-or-less correct model of the solar system (keep reading and I’ll explain): the Sun rotates around the Earth and all the other planets rotate around the sun. The trolling is on a meta-level, as an adversarial attack on a certain way of thinking. But the model I believe 100%.

The crux is that frame of reference is arbitrary. For simplicity, take two planets on a 2D plane and assume one is orbiting the other in a circular fashion. If you are standing at the center, the planet at the other end is rotating to your right. Now imagine you are standing on the other planet. Notice that distance between the two planets (irrespective of frame of reference) remains constant. Furthermore, with respect to any fixed direction, the angle between the two planets is also changing at a constant rate. As a result, the original “center” planet is also rotating around you!

Perhaps you still don’t believe me, so here is a more mechanistic explanation that applies for the general case. Blue is following an arbitrary elliptical orbit around Red and we are rendering this in a Cartesian XY plane as per the figure below. Let’s arbitrarily say that Red is at the origin and that at a given timestep t, B is at the location (x_t, y_t). This is the left figure above. Now, imagine a second rendering where we “recalculate” the coordinate plane at every step such that B is instead always at origin. If you play League of Legends, this is equivalent to turning on camera lock on the Blue planet. Note that this is merely a cosmetic change which does not change the underlying orbit mechanics. Under this new rendering, Red will be at the point (-x_t, -y_t) at each timestep t …. which is also obviously results in the shape of an ellipse. If you STILL don’t believe me, look at the GIF above and verify that the relative locations of Blue and Red are always the same on both the left and right sides of the animation.

Of course, moving the frame of reference from the Sun to Earth does not change the relative relationship between the other planets and Sun, which is why the original claims that only the Sun (and not the other planets) rotates around the Earth. Nevertheless, this line of reasoning allows us to say even crazier things too. The Sun actually revolves around Pluto, despite the deep state insisting that it’s not a planet! The Moon doesn’t rotate around the Earth — how geocentric of you to have such an egotistical thought. It’s actually the other way around! The other planets revolve around the Sun which in turn revolves around the Earth which in turn revolves around the moon, which is obviously the one and only center of the universe. All hail Princess Diana!

This model is a little bit incorrect for the exact same reason that the standard heliocentric model is incorrect. Elementary physics tells us that a system rotates around its center of mass (assuming no reference frame shenanigans). The Sun is large, but not quite large enough to contain the center of mass for the solar system. Nevertheless, the statement that the planets and the Sun all orbit around a very particular point In space around 1.4 million km from the center of the Sun is a statement with far too much nuance than they would ever teach in the American schooling system.

Let’s bring it all together. Imagine you are a peasant, circa 1550 AD. Some crackpot scientist tells you that actually the stars above are stationary, but you are moving really fast around them. Well actually he says one star, but you aren’t listening. You tell him to get lost. Anyone can see that the stars are moving — you literally see them moving every night. He insists on his “theory”. Some of your friends are kinder than you. They patiently listen to his theories and ask him to predict the future locations of the stars and promise to compare it with their observations. You tell them it’s a fool’s task. Only God can guide the paths of the heavens above. True to your expectations, the “scientist” predicts them wrong [1]. Some other people in expensive white robes say that we should burn him at the stake for blasphemy. You don’t mind. You and your family are literally starving, and this man is clearly not worth feeding. 

Fast forward a few years to 2022 AD. Every single child has it drilled into them from Day 1 that the Earth revolves around the Sun, despite the fact that the Earth is very clearly not moving from our current reference frame. This is critically NOT because we instilled every child with the necessary reasoning skills and understanding of the natural laws of our universe such that they could deduce this fact. In fact, most people would lack the ability to recognize a mathematically equivalent statement when directly presented it [2].

Animation code is available here.

[1]: Astronomers were unable to get accurate predictions of the stars/planets until Kepler came around with the ellipse almost a century later. Circular orbits DO NOT give accurate predictions of planetary motion, regardless of what you use as the center of orbit.

[2]: I tried this prank on two Harvard physics PhDs. One understood it immediately. He worked in relativistic physics. The other, however, didn’t believe me for ten whole minutes until he finally realized what was going on. That’s how deep the conditioning goes. Even physics PhDs have trouble undoing it.

774 words, 37 min

I have a radical plan for solving college debt. Hard caps on university tuition. To give a concrete number, let’s say that no US university can charge more than 10k/year in tuition.

The major benefit is obvious. Most fixes for college debt (e.g. loan forgiveness, better loan terms) are just temporary painkillers that don’t fix the underlying root. Hard capping tuition goes right to the heart of the problem. If this is implement, college debt will essentially be a thing of the past.

You might protest about the legality. I’m not a lawyer, so take this with a little bit of skepticism. While the government has a hard time literally enforcing literal price caps on private institutions (perky Bill of Rights), the remedy is quite simple. Most universities receive an enormous amount of research funding from the federal government, which is how Title IX restrictions are enforced. In America, you can’t force private businesses to do much of anything. However, the government is very much allowed to give grants of the form: Satisfy XYZ conditions or you won’t get the money.

The iron law of the universe is that you take someone’s money, one way or another, you may find yourself beholden to their whims. Today, we’ll use this force for good. The offer will be: no university who charges more than 10K a year in tuition will receive any federal research funding.

Logistically, there are a few minor things to sort out, but they are all doable. You will want to prevent the universities from circumventing this restriction by doing things like charging enormous mandatory fees, forcing students to buy overpriced food and housing, or making it impossible to graduate in four years. You probably also want to gradually anneal this restriction into effect over the course of say 10 years, where the hard cap falls by a certain amount every year. You probably want it to increase automatically with inflation. And so on. Lawyers are pretty good at this though and I bet they will have a blast sorting it out. 

The obvious drawback is obvious. You will get extremely, extremely, extremely, extremely strict austerity measures. Apparently top universities like Stanford do not actually make enough from tuition to cover their costs and rely on investment / donations to make a profit (it is unclear though if this is clever accounting or an actual reflection of the facts). So if you cut their tuition levels down to 10K, they will feel the pain and they will feel it hard. Lots of things are getting cut. Lots and lots and lots of things. Some of these things will be things that I would personally not cut, but universities will due to budgeting. An incomplete list of things I expect to be cut at Stanford (some good, some bad):

• Financial aid will be basically be obliterated. On the flip side, the need for financial aid will also be largely obliterated. Some people will end up worse off. Most people will end up better. There are no Pareto gains in a rigged system.
• Admins. Lots and lots of admins will lose their jobs.
• Humanities funding will be cut and not proportionally to how mcc they have. 
• Community centers are gone.
• Minor sports are gone.
• Mental health services are gone (frankly, they barely exist as it stands).
• Obscene grants, such as sending the entire dorm of 100 people on an all expenses paid trip to Disneyland or paying for your summer vacation in Europe for your “photography” project will be removed. Very reasonable grants, like subsidizing your internship in government instead of your consulting one will sadly also go. Admins don’t seem to be able to differentiate between the two.
• CS course assistants will take a pay cut from around $250/hr to around $25/hr (this is not a typo).
• Club funding will be cut from mid five figures a year per club to probably only a few hundred or so, or a thousand on a rare occasion.
• Stanford might stop literally buying out the police department. 
• Hopefully not the palm trees though. I kind of like the palm trees.

Sadly this won’t pass. The government does not have the ability in 2022 to do much at all, let alone an extremely unpopular but necessary surgery like this. Leftists simultaneously complain that tuition is too expensive and then call for instituting yet another bureaucratic institution that will drive up the price. Right wing conservatives are so unhappy with the modern university that they half want to see them fail.  

841 words, 70 min

Yesterday, I claimed that Chinese-American culture essentially uses clipped rewards, but not why. Here’s why.

One of the defining features of Chinese-American culture (at least in the Bay Area, though I suspect in any place with a critical mass of Asians) is the dinner party. These dinner parties are strange beasts. For one, they are really two worlds mashed together, with almost no overlap between.

The kids would mostly play video games, while being almost completely ignored by the parents. Growing up, the game was usually Super Smash Brother’s Melee in the free-for-all mode, whereupon the bottom two places rotated out and the top two stayed. [redacted] and I had several restricted gaming time as kids (e.g. Chinese Lessons), and so we were always in the bottom two. Dinner parties were a constant cycle of rotate in, get sh** on, rotate out, wait 15 minutes, then get sh** on again, and so on.

On the other hand, the parents would sit at the dinner table for hours upon hours, spending almost all that time flexing on each other, with the occasional complaints about Chinese (not Chinese-American, mainland Chinese) politics. Far from dreading it, most parents seem to treat this as the one thing in their life worth living for. And of course, the main thing that they would flex about was whatever their kids were up to these days. I know partly from my occasional drop ins that catch a fragment of conversation, partly from the summaries that I would get from my parents after the dinner party. But mainly because when the kids grew older and went off to college, the dinner parties stopped.

Over the years, I ended up witnessing enough awkward behavior that I (at the time) resolved to never host a dinner party of my own after growing up. I was cured from this curse by none other than the Office episode: Dinner Party, which was infinitely more funny due to the hundred awkward dinner parties I have personally experienced. Because as bad as anything I’ve seen was, no dinner party of mine has been even a tenth as bad as Michael and Jan’s dinner party.

What do these dinner parties have to do with clipped reward functions? In the Asian parent gossip circles, where all rewards are artificial, you can imagine your reward function goes up every time you see an opportunity to one-up another parent and down every time you get one-uped yourself. Importantly, what matters is that you can get one-uped, not how much you get one-uped by! The implied utility function is precisely clipped reward.

Imagine your best achieving kids get offered the following opportunity: 1% chance of becoming a billionaire and 99% chance of making only 50K / year (to be clear I don’t think this trade exists). Does Chinese culture recommend you take this opportunity? No, it’s too “risky”. Let me reframe it for you in the Asian-American mentality. You have a 99% chance of dropping to near the bottom of your social circle and only 1% chance of being at the top. If you are sitting near the top, will you take it? No, it’s an obvious loss.

Ironically, I’m not even sure that this applies to mainland Chinese culture? I don’t exactly have anywhere close to a random sample, but the mainland Chinese people I know have substantially lower. The one summer I was in Beijing everyone was talking about starting startups.

More generally, this applies whenever your incentives are to retain a high rank in a smaller pool and then switch to a larger pool.  Done wrong, you training in the small pool may teach you extreme risk aversion, which is often the exact opposite of what you want to do. If you optimize for your “rank” in a subgroup and not your actual “result”, it is very possible to end up in an equilibrium where people become extremely risk averse.

Here’s one concrete example. I’ve been watching Worlds 2022, which has been nothing short of spectacular may I say. One recurring theme in Worlds is that teams from North America never do well. Many reasons have been given, which are probably true. Here is another, based on the ideas in this piece. If you are a superteam in NA LCS, the optimal strategy to maximize your regional success is to reduce your variance. The narrative that TSM constantly wins first seed before choking at worlds is basically exactly this. TSM is not a fiesta style team like JDG that goes for high variance plays (they are a fiesta team in the sense that they go for low EV plays though). Against teams that you are better than, this low variance will result in clear-cut victories. Against teams that are better than you though, you will get smashed. It will not look close. And they will look like they rolled over without a fight. Which is kind of what happened. But not because they “chose” to not show up internationally. It’s a direct consequence of culture and incentives!

669 words, 120 min

I was not completely satisfied with the description yesterday. Here is another try to get at the same idea. First, some observations. 

• It doesn’t matter how elite you are at basically any sport. Very few Asian-American households will support going pro. Jeremy Lin is the exception not the rule.
• This is even stronger in e-sports. You could be second coming of Faker himself. This is not really hypothetical. Doublelift, arguably the greatest player in NA LCS history, became pro despite the best efforts of his parents. This conflict was so deep that it ended up in absolute tragedy.
• You often hear stories of Asian-Americans running a successful side business who are very hesitant to pursue it full time. On the extreme other end, some subsets of American culture heavily encourage you to quit with no traction whatsoever, just a desire. Meanwhile, Asian-American culture sees that you are bringing in six figures a month and being like … hmmm, are you sure you want to quit your safe and secure job that makes literally 10x less? It’s not money, it’s safety.
• Despite Asian-Americans being extremely well represented in Silicon Valley, very few start their own startups. Indian culture is much, much, much, better at this than East Asian culture, and still not that great. Chinese-American culture is absolutely abysmal. And this is not an accident. My Chinese-American friends who went to elite universities seem to have systematically lower outcomes that people who are otherwise seemingly equivalent. 

So what’s going on here? There is a common theme to all the above examples. With large-ish probability you fail, and with some small probability you achieve wild success. Chinese culture avoids these like the plague. Crucially, it seems like the only thing that matters are the probabilities, and not the magnitude of the difference between success and failure.

Fitting the implied value function to the actions would suggest that Chinese-American culture has a clipped reward function. That is, after a certain point, you don’t really care about further improvements.

It’s not that the culture doesn’t want to optimize! Everyone knows about how hard Asian-Americans strive to go to a good university. You optimize, ruthlessly optimize, but only up to a certain point, after which you are almost completely indifferent — sometimes actively hostile — to further improvement, for fear that you shake the boat too much and slip past the inflection point. When you transform the utility function to be clipped, then suddenly all the behavior becomes rational!

The expected value calculation goes as follows. Most of the time you fail, and that outcome isn’t great. In the rare cases you succeed, I don’t care what the rewards are because my reward functions are clipped. How much more does Asian culture value a successful politician (insert whatever career you want here) over a run-of-the-mill software engineer who pulls in half a million a year? It’s not 100x I can tell you. I doubt it’s even 10x. How much less does it value a failed politician over the same run-of-the-mill software engineer? Now we can pull out the 100x. With such value functions, it is therefore obviously rational to pursue the safe option. Always. Irrespective of your affinity for an arbitrary path or your dis-affinity for software fingering or medicine. 

Go to an elite university and study CS or medicine, but I don’t really care what you do after that. Find someone reasonable to marry, but I don’t really care if you have a great relationship or not. Chinese-American culture is hedgehog to the extreme. Everything is predicated on minimizing the left tail with almost no attention paid to the right tail at all.

If stereotypical white American culture tells you to chase your dreams regardless of reality, stereotypical Chinese-American culture is the exact opposite: always take the safe route out, no matter what. And if your culture’s attitude is that no one should take these risky plays … your culture will have almost no people who end up with right tail success outcomes. 

691 words, 131 minutes

Malcolm Gladwell talks at length about the fox and the hedgehog. Supposedly, the fox knows many things while the hedgehog knows one big thing. Today, we will focus on an entirely different difference between these two species of animals. One that is actually grounded in reality.

Foxes are hunters by nature. Each day involves risk: if they fail to kill they do not eat. On the other hand, hedgehogs are not quite herbivores, but their diet consists largely of insects like beetles, caterpillars, and so on, that do not exactly require hunting. Rare is the day a hedgehog wishes to eat but can find no worms to catch. 

As a hasty generalization, the fox and the hedgehog represent two broad philosophies that people take towards risk. Hedgehogs spend their efforts minimizing their left-tail outcomes. Foxes place their efforts into maximizing right-tail outcomes. This idea appears in many different forms, often in disguise. Risk-averse / risk-seeking behavior in the classic economics. Pessimists vs optimists. Convex vs concave dispositions by Vitalik Buterin. Kanye West and Jay Z’s Murder to Excellence.

Chinese culture is ruthlessly risk-averse, to the point that it would sell the entire right tail for pennies on the dollar, just to minimize our left tail a tiny bit more. Where are the Chinese-American CEOs, the actors, the athletes, the pop stars, the comedians, we ask? We complain about racism (and no doubt there is racism, to be clear), but perhaps the bigger answer is that we ourselves have smothered them in the crib.

We tell everyone to be software engineers or doctors, regardless of desire or talent [1]. It’s not money. It’s safety. For perhaps the most egregious example, take Jeremy Lin. Jeremy Lin was Northern California player of the year as a senior. Over a decade before Faker carried 4 dead bodies to the 2017 World Championship, Jeremy Lin carried Palo Alto High School to a state championship. Palo Alto High School. Palo Alto High School is about as likely to win a basketball championship as Mater Dei is likely to win Mathcounts. In literally any other culture, Jeremy Lin would be strapped to a rocket jet by his community. There would be no question that he would shoot for the NBA. In Asian culture, we get everyone other than his parents making fun of his family for even trying to play basketball. In fact, I bet that in f***ing 2022, even after Jeremy Lin is in some sense the iconic Asian-American, the majority of Asian parents would still prohibit their son in a similar situation to shoot for the NBA. Actually that’s a lie. They would smother the basketball dream before he ever had a chance to show potential. 

I can think of no scathing criticism sharper for our own culture than that we almost let Jeremy Lin slip away and become a doctor or software engineer. You may retort that Jeremy Lin actually made it to the NBA, but I ask you, how many others did not? And it’s not just basketball. It’s true in business, in comedy, in acting, in gaming, in basically every field.

In the end, if our culture sells the right tail for pennies on the dollar, it will have no right tail. Much has changed in my life in the past year, but if I had to summarize the biggest change in one line it is this: I wish to escape this trap. Personally, of course, but also for my brothers and sisters who could not.

[1]: I hear lawyer too, but that doesn’t match up with my experience. I don’t know many Asian-American lawyers, either growing up or, more surprisingly, at Harvard/Stanford. I suspect that this meme was either never true, or at least not true in the present day. Evidence: 5% of lawyers are Asian American …. which is exactly what you would expect from the population distribution. In fact, it’s worse than you expect conditioned on the fact that Asian Americans have higher than average university attendance. My made-up explanation. Asian parents say lawyer or doctor, but in practice the kids (or the parents) choose doctor if given the option.

842 words, 38 minutes

I arrived in London this morning and so I’ve been reflecting on what exactly I want to get done while here. One of the few things I did right during the pandemic was starting this blog. The main dividend it paid is that I now write ~10x faster than I did before (I suppose this tells you just how slowly I wrote in the past). Nevertheless, I still suspect there is another 10x to go, and I think that I should go get it. So that will be the first goal of London. 10x my writing speed (slightly lower quality is acceptable for this goal).

I once read an old story of a photographer who ran an experiment on two classes of novice photographers he was teaching. For one class, he told them that they would be graded exclusively on quantity. If they submitted 100 unique photographs by the end of the semester, they would get full marks. Quality was irrelevant. The second class was the opposite. He told them that they could take as many or as little photos as they liked, but that at the end of the semester, they would submit one photo and the quality of that one photo would determine the entire grade for the class. Which class ended up producing better photos?

At the time, I was extremely confident that the answer would be the second class. After all, the first class had ZERO pressure to get good photographs. But I was wrong. The reasoning is clear when you examine the differences in behavior. The first class spent their time experimenting with a large number of different settings and shots. Most of them were bad, but some of them ended up turning out quite good! The second class spent their time only in a few settings agonizing over “perfection.” Except, since they were novice photographers, agonizing over “perfection” is absolutely useless. Imagine a 25 kyu in Go thinking for 15 minutes on their next move. “I just want to get it perfect you know — I won’t settle for anything less than the perfect move.” The next move they play is a self-atari.

An actual story from my past. When I was a child (slightly better than 25 kyu, but not by much), I asked why my teacher was only commenting on some of the moves I played and not others. He told me that he was only pointing out my egregious, immediately game-losing errors. I told him that I wanted to play perfectly and demanded that he tell me everything I did wrong. He stared at me and said, “[memorymancer], if I’m going to be frank with you, every move you play is wrong” (北辰， 说实话, 你每一步都是错的). 

My new belief is that quantity trumps quality until you get to a very high level. If you are just starting to play Go, just play Go. Don’t worry about anything. It’s the same for running, basketball, poetry. Just do it. At the novice level, time spent doing a “good” rep is less useful than getting another rep in. Of course, at some point, your improvement will certainly stall. This, hilariously, might be take a lot longer than you think though. At that point, you can start working on quality — when you are trying to chase elite performance. But before then, it is just quantity. 

And so, I’m going to try this experiment myself. I’m going to spend November writing. The full NaNoWriMo is certainly too much at my current level. Instead, I’ll try to write a post a day for the next month. No length requirements. Just a semi-complete thought. Larger thoughts may take several days to complete, but will be published in sections. Posts can be written in advance, though I am quite nervous to say that I have zero backlog at the moment. So all posts will be at least mostly written in November. I considered the alternative of simply to dedicate an hour to writing everyday. My main with that was simply that there was too high a risk of appearing to succeed but actually failing. What if after an hour I have nothing? Is that actually a success? I don’t think so. 

Frankly, I’m terrified. The obvious failure mode is having nothing to publish. In those cases, I pre-commit to publishing absolute garbage. I’m also terrified of publishing without extensive review. Previous posts on this blog I would usually hold for a month or so simply to check that I still held that belief. In times of great strife, much can change in even a few day’s time. We will not be doing that here. These posts will be off the cuff things that I might say at a dinner conversation after having thought about it for only a few minutes. It is very possible that I will write a post strongly arguing one side and then a few days later argue the exact opposite. So apologize in advance for the deluge of potentially very low quality posts. But here we go.