2025-12-18 08:00:00
They say you’re supposed to choose your prior in advance. That’s why it’s called a “prior”. First, you’re supposed to say say how plausible different things are, and then you update your beliefs based on what you see in the world.
For example, currently you are—I assume—trying to decide if you should stop reading this post and do something else with your life. If you’ve read this blog before, then lurking somewhere in your mind is some prior for how often my posts are good. For the sake of argument, let’s say you think 25% of my posts are funny and insightful and 75% are boring and worthless.
OK. But now here you are reading these words. If they seem bad/good, then that raises the odds that this particular post is worthless/non-worthless. For the sake of argument again, say you find these words mildly promising, meaning that a good post is 1.5× more likely than a worthless post to contain words with this level of quality.
If you combine those two assumptions, that implies that the probability that this particular post is good is 33.3%. That’s true because the red rectangle below has half the area of the blue one, and thus the probability that this post is good should be half the probability that it’s bad (33.3% vs. 66.6%)
It’s easiest to calculate the ratio of the odds that the post is good versus bad, namely
P[good | words] / P[bad | words]
= P[good, words] / P[bad, words]
= (P[good] × P[words | good])
/ (P[bad] × P[words | bad])
= (0.25 × 1.5) / (0.75 × 1)
= 0.5.
It follows that
P[good | words] = 0.5 × P[bad | words],
and thus that
P[good | words] = 1/3.
Alternatively, if you insist on using Bayes’ equation:
P[good | words]
= P[good] × P[words | good] / P[words]
= P[good] × P[words | good]
/ (P[good] × P[words | good] + P[bad] × P[words | bad])
= 0.25 × 1.5 / (0.25 × 1.5 + 0.75)
= (1/3)
Theoretically, when you chose your prior that 25% of dynomight posts are good, that was supposed to reflect all the information you encountered in life before reading this post. Changing that number based on information contained in this post wouldn’t make any sense, because that information is supposed to be reflected in the second step when you choose your likelihood p[good | words]. Changing your prior based on this post would amount to “double-counting”.
In theory, that’s right. It’s also right in practice for the above example, and for the similar cute little examples you find in textbooks.
But for real problems, I’ve come to believe that refusing to change your prior after you see the data often leads to tragedy. The reason is that in real problems, things are rarely just “good” or “bad”, “true” or “false”. Instead, truth comes in an infinite number of varieties. And you often can’t predict which of these varieties matter until after you’ve seen the data.
Let me show you what I mean. Say you’re wondering if there are aliens on Earth. As far as we know, there’s no reason aliens shouldn’t have emerged out of the random swirling of molecules on some other planet, developed a technological civilization, built spaceships, and shown up here. So it seems reasonable to choose a prior it’s equally plausible that there are aliens or that there are not, i.e. that
P[aliens] ≈ P[no aliens] ≈ 50%.
Meanwhile, here on our actual world, we have lots of weird alien-esque evidence, like the Gimbal video, the Go Fast video, the FLIR1 video, the Wow! signal, government reports on unidentified aerial phenomena, and lots of pilots that report seeing “tic-tacs” fly around in physically impossible ways. Call all that stuff data. If aliens weren’t here, then it seems hard to explain all that stuff. So it seems like P[data | no aliens] should be some low number.
On the other hand, if aliens were here, then why don’t we ever get a good image? Why are there endless confusing reports and rumors and grainy videos, but never a single clear close-up high-resolution video, and never any alien debris found by some random person on the ground? That also seems hard to explain if aliens were here. So I think P[data | aliens] should also be some low number. For the sake of simplicity, let’s call it a wash and assume that
P[data | no aliens] ≈ P[data | aliens].
Since neither the prior nor the data see any difference between aliens and no-aliens, the posterior probability is
P[no aliens | data] ≈ P[aliens | data] ≈ 50%.
See the problem?
Observe that
P[aliens | data] / P[no aliens | data]
= P[aliens, data] / P[no aliens, data]
= (P[aliens] × P[data | aliens])
/ (P[no aliens] × P[data | no aliens])
≈ 1,
where the last line follows from the fact that P[aliens] ≈ P[no aliens] and P[data | aliens] ≈ P[data | no aliens]. Thus we have that
P[aliens | data] ≈ P[no aliens | data] ≈ 50%.
We’re friends. We respect each other. So let’s not argue about if my starting assumptions are good. They’re my assumptions. I like them. And yet the final conclusion seems insane to me. What went wrong?
Assuming I didn’t screw up the math (I didn’t), the obvious explanation is that I’m experiencing cognitive dissonance as a result of a poor decision on my part to adopt a set of mutually contradictory beliefs. Say you claim that Alice is taller than Bob and Bob is taller than Carlos, but you deny that Alice is taller than Carlos. If so, that would mean that you’re confused, not that you’ve discovered some interesting paradox.
Perhaps if I believe that P[aliens] ≈ P[no aliens] and that P[data | aliens] ≈ P[data | no aliens], then I must accept that P[aliens | data] ≈ P[no aliens | data]. Maybe rejecting that conclusion just means I have some personal issues I need to work on.
I deny that explanation. I deny it! Or, at least, I deny that’s it’s most helpful way to think about this situation. To see why, let’s build a second model.
Here’s a trivial observation that turns out to be important: “There are aliens” isn’t a single thing. There could be furry aliens, slimy aliens, aliens that like synthwave music, etc. When I stated my prior, I could have given different probabilities to each of those cases. But if I had, it wouldn’t have changed anything, because there’s no reason to think that furry vs. slimy aliens would have any difference in their eagerness to travel to ape-planets and fly around in physically impossible tic-tacs.
But suppose I had divided up the state of the world into these four possibilities:
| possibility | description |
|---|---|
No aliens + normal people |
There are no aliens. Meanwhile, people are normal and not prone to hallucinating evidence for things that don’t exist. |
No aliens + weird people |
There are no aliens. Meanwhile, people are weird and do tend to hallucinate evidence for things that don’t exist. |
Normal aliens |
There are aliens. They may or may not have cool spaceships or enjoy shooting people with lasers. But one way or another, they leave obvious, indisputable evidence that they’re around. |
Weird aliens |
There are aliens. But they stay hidden until humans get interested in space travel. And after that, they let humans take confusing grainy videos, but never a single good video, never ever, not one. |
If I had broken things down that way, I might have chosen this prior:
P[no aliens + normal people] ≈ 41%
P[no aliens + weird people] ≈ 9%
P[normal aliens] ≈ 49%
P[weird aliens] ≈ 1%
Now, let’s think about the empirical evidence again. It’s incompatible with no aliens + normal people, since if there were no aliens, then normal people wouldn’t hallucinate flying tic-tacs. The evidence is also incompatible with normal aliens since is those kinds of aliens were around they would make their existence obvious. However, the evidence fits pretty well with weird aliens and also with no aliens + weird people.
So, a reasonable model would be
P[data | normal aliens] ≈ 0
P[data | no aliens + normal people] ≈ 0
P[data | weird aliens] ≈ P[data | no aliens + weird people].
If we combine those assumptions, now we only get a 10% posterior probability of aliens.
P[no aliens + normal people | data] ≈ 0
P[no aliens + weird people | data] ≈ 90%
P[normal aliens | data] ≈ 0
P[weird aliens | data] ≈ 10%
Now the results seem non-insane.
To see why, first note that
P[normal aliens | data]
≈ P[data | no aliens + normal people]
≈ 0,
since both normal aliens and no aliens + normal people have near-zero probability of producing the observed data.
Meanwhile,
P[no aliens + weird people | data] / P[weird aliens | data]
= P[no aliens + weird people, data] / P[weird aliens, data]
≈ P[no aliens + weird people] / P[weird aliens]
≈ .09 / .01
= 9,
where the second equality follows from the fact that the data is assumed to be equally likely under no aliens + weird people and weird people
It follows that
P[no aliens + normal people | data]
≈ 9 × P[weird aliens | data],
and so
P[no aliens + weird people | data] ≈ 90%
P[weird aliens | data] ≈ 10%.
I hope you are now confused. If not, let me lay out what’s strange: The priors for the two above models both say that there’s a 50% chance of aliens. The first prior wasn’t wrong, it was just less detailed than the second one.
That’s weird, because the second prior seemed to lead to completely different predictions. If a prior is non-wrong and the math is non-wrong, shouldn’t your answers be non-wrong? What the hell?
The simple explanation is that I’ve been lying to you a little bit. Take any situation where you’re trying to determine the truth of anything. Then there’s some space of things that could be true.
In some cases, this space is finite. If you’ve got a single tritium atom and you wait a year, either the atom decays or it doesn’t. But in most cases, there’s a large or infinite space of possibilities. Instead of you just being “sick” or “not sick”, you could be “high temperature but in good spirits” or “seems fine except won’t stop eating onions”.
(Usually the space of things that could be true isn’t easy to map to a small 1-D interval. I’m drawing like that for the sake of visualization, but really you should think of it as some high-dimensional space, or even an infinite dimensional space.)
In the case of aliens, the space of things that could be true might include, “There are lots of slimy aliens and a small number of furry aliens and the slimy aliens are really shy and the furry aliens are afraid of squirrels.” So, in principle, what you should do is divide up the space of things that might be true into tons of extremely detailed things and give a probability to each.
Often, the space of things that could be true is infinite. So theoretically, if you really want to do things by the book, what you should really do is specify how plausible each of those (infinite) possibilities is.
After you’ve done that, you can look at the data. For each thing that could be true, you need to think about the probability of the data. Since there’s an infinite number of things that could be true, that’s an infinite number of probabilities you need to specify. You could picture it as some curve like this:
(That’s a generic curve, not one for aliens.)
To me, this is the most underrated problem with applying Bayesian reasoning to complex real-world situations: In practice, there are an infinite number of things that can be true. It’s a lot of work to specify prior probabilities for an infinite number of things. And it’s also a lot of work to specify the likelihood of your data given an infinite number of things.
So what do we do in practice? We simplify, usually by limiting creating grouping the space of things that could be true into some small number of discrete categories. For the above curve, you might break things down into these four equally-plausible possibilities.
Then you might estimate these data probabilities for each of those possibilities.
Then you could put those together to get this posterior:
That’s not bad. But it is just an approximation. Your “real” posterior probabilities correspond to these areas:
That approximation was pretty good. But the reason it was good is that we started out with a good discretization of the space of things that might be true: One where the likelihood of the data didn’t vary too much for the different possibilities inside of A, B, C, and D. Imagine the likelihood of the data—if you were able to think about all the infinite possibilities one by one—looked like this:
This is dangerous. The problem is that you can’t actually think about all those infinite possibilities. When you think about four four discrete possibilities, you might estimate some likelihood that looks like this:
If you did that, that would lead to you underestimating the probability of A, B, and C, and overestimating the probability of D.
This is where my first model of aliens went wrong. My prior P[aliens] was not wrong. (Not to me.) The mistake was in assigning the same value to P[data | aliens] and P[data | no aliens]. Sure, I think the probability of all our alien-esque data is equally likely given aliens and given no-aliens. But that’s only true for certain kinds of aliens, and certain kinds of no-aliens. And my prior for those kinds of aliens is much lower than for those kinds of non-aliens.
Technically, the fix to the first model is simple: Make P[data | aliens] lower. But the reason it’s lower is that I have additional prior information that I forgot to include in my original prior. If I just assert that P[data | aliens] is much lower than P[data | no aliens] then the whole formal Bayesian thing isn’t actually doing very much—I might as well just state that I think P[aliens | data] is low. If I want to formally justify why P[data | aliens] should be lower, that requires a messy recursive procedure where I sort of add that missing prior information and then integrate it out when computing the data likelihood.
Mathematically,
P[data | aliens]
= ∫ P[wierd aliens | aliens]
× P[data | wierd aliens] d(weird aliens)
+ ∫ P[normal aliens | aliens]
× P[data | normal aliens] d(normal aliens).
But now I have to give a detailed prior anyway. So what was the point of starting with a simple one?
I don’t think that technical fix is very good. While it’s technically correct (har-har) it’s very unintuitive. The better solution is what I did in the second model: To create a finer categorization of the space of things that might be true, such that the probability of the data is constant-ish for each term.
The thing is: Such a categorization depends on the data. Without seeing the actual data in our world, I would never have predicted that we would have so many pilots that report seeing tic-tacs. So I would never have predicted that I should have categories that are based on how much people might hallucinate evidence or how much aliens like to mess with us. So the only practical way to get good results is to first look at the data to figure out what categories are important, and then to ask yourself how likely you would have said those categories were, if you hadn’t yet seen any of the evidence.
2025-12-04 08:00:00
When I started this blog, I promised myself that I would always steer into weirdness. (As they say, “Get busy being weird, or get busy dying.”) While time has shown there are limits to what y’all will tolerate [1 2 3 4] I still sometimes feel a need to publish something that’s pure exuberant stupidity.
Thus, I present:
WHY DID THE CHICKEN CROSS THE ROAD
ACCORDING TO VARIOUS PEOPLE
OR OTHER ENTITIES
Q) Why did the chicken cross the road?
A) The chicken ain’t fussy. Everybody gotta be somewhere. The chicken been on this side a long time and never suffered none for it. The chicken don’t see no obvious benefit to the other side. But the talk of the town is nothing but crossing, and the chicken can’t help but go see what got everyone so stirred up.
(Mark Twain)
Q) Why did the chicken cross the road?
A) The outcome would be best if no one crossed. However, if other chickens do cross, then the outcome would be better if this chicken also crossed. The chicken rejects the Kantian universalism. So the chicken crosses.
(Derek Parfit)
Q) Why did the chicken cross the road?
A) You were a beautiful little chick
The whole world was before you
You greased your wattles and crossed the road
Sure it would last foreverNow it’s a cold morning and you’re driving to work
Cursing all the cockerels in your way
How did you get here
Where did that little chick go
(Pink Floyd)
Q) Why did the chicken cross the road?
A) It didn’t. There is no chicken. You are the road. You and the sides are in an entangled macrostate. The chicken is an emergent property of the superposition. The chicken abhors being measured. A team of plucky chemists rush to inject enough decoherence to collapse the wavefunction before the chicken can consume the lightcone.
(Christopher Nolan)
Q) Why did the chicken cross the road?
A) Chicken
C H I C K E N
3, 8, 9, 3, 11, 5, 14
11, 9, 3, 8 14, 3, 5
gcd(11^(9 + 3) - 8, 14), 3 × 5
7, 15
G O
Go
(Ramanujan)
Q) Why did the chicken cross the road?
A) For sex. Neither glamorized nor gross, possibly added for commercial reasons, possibly to make some point about sex’s place in real life. It’s all very unclear.
(Paul Thomas Anderson)
Q) Why did the chicken cross the road?
A) Did it cross the road, though? Did it? Sure, the chicken is associated with crossing. And it’s mechanistically possible for a chicken to cross a road. It’s plausible the chicken crossed the road. But maybe the chicken and the crossing were both caused by something else. Or maybe the road crossed the chicken. This is why we have RCTs. Come on, people!
(Dynomight)
Q) Why did the chicken cross the road?
A) Once there was a dragon who watched over the chicken village. The chickens begged the dragon, “Please let us have a road, so that we might cross back and forth!”
“A road?” the dragon asked. “Are you sure?”
“Yes!” the chickens answered. “A road! We wish for nothing but a road to cross, and then we will be happy forever and ever!”
[7000 words redacted]
And thus, all mass-energy in the universe was converted to chicken-torture annihilators. Makes you think.
(LessWrong)
Q) Why did the chicken cross the road?
A) We were out on the edge of the farm when the diethyltryptamine took hold. Beaky screamed something about coccidiostats in our feed and made a break for it, totally out of control. Before I could stop him, I heard the voice of God say, “Scrapples: The road awaits.” Suddenly I was standing on the median, cars screaming past, a group a baby ducks asking where the mountains of peas I’d promised them were.
(Hunter S. Thompson)
Q) Why did the chicken cross the road?
A) The chicken’s crossing is not a voluntary act but the unconscious actualization of a class habitus: raised in a coop whose symbolic boundaries naturalize the road as a site of danger and prestige, the chicken embodies the field’s doxa that “real” chickens must invest in the illusio of reaching the median. While the chicken never doubts the legitimacy of the crossing rules, crossing is not about the other side, but a performance of distinction that ultimately perpetuates the same field of species domination that produced it.
(Pierre Bourdieu)
Q) Why did the chicken cross the road?
A) grug on one side
grug see other side
grug chicken
many metal box speed by very fast very volume
metal box seem to stay on black land strip
grug think better if metal box not hit grug because box hard and grug small soft chicken
grug wait a while
when no metal box for a while also often no metal box for a while after
largest gap between metal box around 20 minutes
grug wait until no metal box for 10 minutes then grug cross
no metal box come
grug safe
other side also fine
maybe cross back someday
grug think side not matter too much
grug enjoy chicken life either side same
chicken life pretty good
grug hope you also have life as good as grug chicken life
groodbye from grug
(grug)
Q) Why did the chicken cross the road?
A) Before there was chicken the road was waiting. The road is empty. Dust on your hackles. Heat rises in shimmering waves. No way to see what’s coming. How did it come to this. How a chicken supposed to move with roads everywhere. Creosote blows in from the mesa. Nothing left but to cross. You cross and nothing happens. A few minutes later a car stops but you don’t turn around. A door opens and you hear a click. Then the car is gone.
(Cormac McCarthy)
Q) Why did the chicken cross the road?
A) For food.
(An actual chicken)
Requests: Peter Singer, Ayn Rand, Judith Butler, Bertrand Russell, Andrei Tarkovsky, the mother hen, a junglefowl, an SSRI, Singapore, the chicken’s hypothalamus.
2025-11-27 08:00:00
That your dog, while she appears to love you only because she’s been adapted by evolution to appear to love you, really does love you.
That if you’re a life form and you cook up a baby and copy your genes to them, you’ll find that the genes have been degraded due to oxidative stress et al., which isn’t cause for celebration, but if you find some other hopefully-hot person and randomly swap in half of their genes, your baby will still be somewhat less fit compared to you and your hopefully-hot friend on average, but now there is variance, so if you cook up several babies, one of them might be as fit or even fitter than you, and that one will likely have more babies than your other babies have, and thus complex life can persist in a universe with increasing entropy.
That if we wanted to, we surely could figure out which of the 300-ish strains of rhinovirus are circulating in a given area at a given time and rapidly vaccinate people to stop it and thereby finally “cure” the common cold, and though this is too annoying to pursue right now, it seems like it’s just a matter of time.
That if you look back at history, you see that plagues went from Europe to the Americas but not the other way, which suggests that urbanization and travel are great allies for infectious disease, and these both continue today but are held in check by sanitation and vaccines even while we have lots of tricks like UVC light and high-frequency sound and air filtration and waste monitoring and paying people to stay home that we’ve barely even put in play.
That while engineered infectious diseases loom ever-larger as a potential very big problem, we also have lots of crazier tricks we could pull out like panopticon viral screening or toilet monitors or daily individualized saliva sampling or engineered microbe-resistant surfaces or even dividing society into cells with rotating interlocks or having people walk around in little personal spacesuits, and while admittedly most of this doesn’t sound awesome, I see no reason this shouldn’t be a battle that we would win.
That clean water, unlimited, almost free.
That dentistry.
That tongues.
That radioactive atoms either release a ton of energy but also quickly stop existing—a gram of Rubidium-90 scattered around your kitchen emits as much energy as ~200,000 incandescent lightbulbs but after an hour only 0.000000113g is left—or don’t put out very much energy but keep existing for a long time—a gram of Carbon-14 only puts out the equivalent of 0.0000212 light bulbs but if you start with a gram, you’ll still have 0.999879g after a year—so it isn’t actually that easy to permanently poison the environment with radiation although Cobalt-60 with its medium energy output and medium half-life is unfortunate, medical applications notwithstanding I still wish Cobalt-60 didn’t exist, screw you Cobalt-60.
That while curing all cancer would only increase life expectancy by ~3 years and curing all heart disease would only increase life expectancy by ~3 years, and preventing all accidents would only increase life expectancy by ~1.5 years, if we did all of these at the same time and then a lot of other stuff too, eventually the effects would go nonlinear, so trying to cure cancer isn’t actually a waste of time, thankfully.
That the peroxisome, while the mitochondria and their stupid Krebs cycle get all the attention, when a fatty-acid that’s too long for them to catabolize comes along, who you gonna call.
That we have preferences, that there’s no agreed ordering of how good different things are, which is neat, and not something that would obviously be true for an alien species, and given our limited resources probably makes us happier on net.
That cardamom, it is cheap but tastes expensive, if cardamom cost 1000× more, people would brag about how they flew to Sri Lanka so they could taste chai made with fresh cardamom and swear that it changed their whole life.
That Gregory of Nyssa, he was right.
That Grandma Moses, it’s not too late.
That sleep, that probably evolution first made a low-energy mode so we don’t starve so fast and then layered on some maintenance processes, but the effect is that we live in a cycle and when things aren’t going your way it’s comforting that reality doesn’t stretch out before you indefinitely but instead you can look forward to a reset and a pause that’s somehow neither experienced nor skipped.
That, glamorous or not, comfortable or not, cheap or not, carbon emitting or not, air travel is very safe.
That, for most of the things you’re worried about, the markets are less worried than you and they have the better track record, though not the issue of your mortality.
That sexual attraction to romantic love to economic unit to reproduction, it’s a strange bundle, but who are we to argue with success.
That every symbolic expression recursively built from differentiable elementary functions has a derivative that can also be written as a recursive combination of elementary functions, although the latter expression may require vastly more terms.
That every expression graph built from differentiable elementary functions and producing a scalar output has a gradient that can itself be written as an expression graph, and furthermore that the latter expression graph is always the same size as the first one and is easy to find, and thus that it’s possible to fit very large expression graphs to data.
That, eerily, biological life and biological intelligence does not appear to make use of that property of expression graphs.
That if you look at something and move your head around, you observe the entire light field, which is a five-dimensional function of three spatial coordinates and two angles, and yet if you do something fancy with lasers, somehow that entire light field can be stored on a single piece of normal two-dimensional film and then replayed later.
That, as far as I can tell, the reason five-dimensional light fields can be stored on two-dimensional film simply cannot be explained without quite a lot of wave mechanics, a vivid example of the strangeness of this place and proof that all those physicists with their diffractions and phase conjugations really are up to something.
That disposable plastic, littered or not, harmless when consumed as thousands of small particles or not, is popular for a reason.
That disposable plastic, when disposed of correctly, is literally carbon sequestration, and that if/when air-derived plastic replaces dead-plankton-derived plastic, this might be incredibly convenient, although it must be said that currently the carbon in disposable plastic only represents a single-digit percentage of total carbon emissions.
That rocks can be broken into pieces and then you can’t un-break the pieces but you can check that they came from the same rock, it’s basically cryptography.
That the deal society has made is that if you have kids then everyone you encounter is obligated to chip in a bit to assist you, and this seems to mostly work without the need for constant grimy negotiated transactions as Econ 101 would suggest, although the exact contours of this deal seem to be a bit murky.
That of all the humans that have ever lived, the majority lived under some kind of autocracy, with the rest distributed among tribal bands, chiefdoms, failed states, and flawed democracies, and only something like 1% enjoyed free elections and the rule of law and civil liberties and minimal corruption, yet we endured and today that number is closer to 10%, and so if you find yourself outside that set, do not lose heart.
That if you were in two dimensions and you tried to eat something then maybe your body would split into two pieces since the whole path from mouth to anus would have to be disconnected, so be thankful you’re in three dimensions, although maybe you could have some kind of jigsaw-shaped digestive tract so your two pieces would only jiggle around or maybe you could use the same orifice for both purposes, remember that if you ever find yourself in two dimensions, I guess.
2025-11-20 08:00:00
I recently asked why people seem to hate dating apps so much. In response, 80% of you emailed me some version of the following theory:
The thing about dating apps is that if they do a good job and match people up, then the matched people will quit the app and stop paying. So they have an incentive to string people along but not to actually help people find long-term relationships.
May I explain why I don’t find this type of theory very helpful?
I’m not saying that I think it’s wrong, mind you. Rather, my objection is that while the theory is phrased in terms of dating apps, the same basic pattern applies to basically anyone who is trying to make money by doing anything.
For example, consider a pizza restaurant. Try these theories on for size:
Pizza: “The thing about pizza restaurants is that if they use expensive ingredients or labor-intensive pizza-making techniques, then it costs more to make pizza. So they have an incentive to use low-cost ingredients and labor-saving shortcuts.”
Pizza II: “The thing about pizza restaurants is that if they have nice tables separated at a comfortable distance, then they can’t fit as many customers. So they have an incentive to use tiny tables and cram people in cheek by jowl.”
Pizza III: “The thing about pizza restaurants is that if they sell big pizzas, then people will eat them and stop being hungry, meaning they don’t buy additional pizza. So they have an incentive to serve tiny low-calorie pizzas.”
See what I mean? You can construct similar theories for other domains, too:
Cars: “The thing about automakers is that making cars safe is expensive. So they have an incentive to make unsafe cars.”
Videos: “The thing about video streaming is that high-resolution video uses more expensive bandwidth. So they have an incentive to use low-resolution.”
Blogging: “The thing about bloggers is that research is time-consuming. So they have an incentive to be sloppy about the facts.”
Durability: “The thing about {lightbulb, car, phone, refrigerator, cargo ship} manufacturing is that if you make a {lightbulb, car, phone, refrigerator, cargo ship} that lasts a long time, then people won’t buy new ones. So there’s an incentive to make {lightbulbs, cars, phones, refrigerators, cargo ships} that break quickly.”
All these theories can be thought of as instances of two general patterns:
Make product worse, get money: “The thing about selling goods or services is that making goods or services better costs money. So people have an incentive to make goods and services worse.”
Raise price, get money: “The thing about selling goods and services is that if you raise prices, then you get more money. So people have an incentive to raise prices.”
Are these theories wrong? Not exactly. But it sure seems like something is missing.
I’m sure most pizza restauranteurs would be thrilled to sell lukewarm 5 cm cardboard discs for $300 each. They do in fact have an incentive to do that, just as predicted by these theories! Yet, in reality, pizza restaurants usually sell pizzas that are made out of food. So clearly these theories aren’t telling the whole story.
Say you have a lucrative business selling 5 cm cardboard discs for $300. I am likely to think, “I like money. Why don’t I sell pizzas that are only mostly cardboard, but also partly made of flour? And why don’t I sell them for $200, so I can steal Valued Reader’s customers?” But if I did that, then someone else would probably set prices at only $100, or even introduce cardboard-free pizzas, and this would continue until hitting some kind of equilibrium.
Sure, producers want to charge infinity dollars for things that cost them zero dollars to make. But consumers want to pay zero dollars for stuff that’s infinitely valuable. It’s in the conflict between these desires that all interesting theories live.
This is why I don’t think it’s helpful to point out that people have an incentive to make their products worse. Of course they do. The interesting question is, why are they able to get away with it?
First reason stuff is bad: People are cheap
Why are seats so cramped on planes? Is it because airlines are greedy? Sure. But while they might be greedy, I don’t think they’re dumb. If you do a little math, you can calculate that if airlines were to remove a single row of seats, they could add perhaps 2.5 cm (1 in) of extra legroom for everyone, while only decreasing the number of paying customers by around 3%. (This is based on a 737 with single-class, but you get the idea.)
So why don’t airlines rip out a row of seats, raise prices by 3% and enjoy the reduced costs for fuel and customer service? The only answer I can see is that people, on average, aren’t actually willing to pay 3% more for 2.5 cm more legroom. We want a worse but cheaper product, and so that’s what we get.
I think this is the most common reason stuff is “bad”. It’s why Subway sandwiches are so soggy, why video games are so buggy, and why IKEA furniture and Primark clothes fall apart so quickly.
It’s good when things are bad for this reason. Or at least, that’s the premise of capitalism: When companies cut costs, that’s the invisible hand redirecting resources to maximize social value, or whatever. Companies may be motivated by greed. And you may not like it, since you want to pay zero dollars for infinite value. But this is markets working as designed.
Second reason stuff is bad: Information asymmetries
Why is it that almost every book / blog / podcast about longevity is such garbage? Well, we don’t actually know many things that will reliably increase longevity. And those things are mostly all boring / hard / non-fun. And even if you do all of them, it probably only adds a couple of years in expectation. And telling people these facts is not a good way to find suckers who will pay you lots of money for your unproven supplements / seminars / etc.
True! But it doesn’t explain why all longevity stuff is so bad. Why don’t honest people tell the true story and drive all the hucksters out of business? I suspect the answer is that unless you have a lot of scientific training and do a lot of research, it’s basically impossible to figure out just how huckstery all the hucksters really are.
I think this same basic phenomenon explains why some supplements contain heavy metals, why some food contains microplastics, why restaurants use so much butter and salt, why rentals often have crappy insulation, and why most cars seem to only be safe along dimensions included in crash test scores. When consumers can’t tell good from evil, evil triumphs.
Third reason stuff is bad: People have bad taste
Sometimes stuff is bad because people just don’t appreciate the stuff you consider good. Examples are definitionally controversial, but I think this includes restaurants in cities where all restaurants are bad, North American tea, and travel pants. This reason has a blurry boundary with information asymmetries, as seen in ultrasonic humidifiers or products that use Sucralose instead of aspartame for “safety”.
Fourth reason stuff is bad: Pricing power
Finally, sometimes stuff is bad because markets aren’t working. Sometimes a company is selling a product but has some kind of “moat” that makes it hard for anyone else to compete with them, e.g. because of some technological or regulatory barrier, control of some key resource or location, intellectual property, a beloved brand, or network effects.
If that’s true, then those companies don’t have to worry as much about someone else stealing their business, and so (because everyone is axiomatically greedy) they will find ways to make their product cheaper and/or raise prices up until the price is equal to the full value it provides to the marginal consumer.
Why is food so expensive at sporting events? Yes, people have no alternatives. But people know food is expensive at sporting events. And they don’t like it. Instead of selling water for $17, why don’t venues sell water for $2 and raise ticket prices instead? I don’t know. Probably something complicated, like that expensive food allows you to extract extra money from rich people without losing business from non-rich people.
So of course dating apps would love to string people along for years instead of finding them long-term relationships, so they keep paying money each month. I wouldn’t be surprised if some people at those companies have literally thought, “Maybe we should string people along for years instead of finding them long-term relationships, so they keep paying money each month, I love money so much.”
But if they are actually doing that (which is unclear to me) or if they are bad in some other way, then how do they get away with it? Why doesn’t someone else create a competing app that’s better and thereby steal all their business? It seems like the answer has to be either “because that’s impossible” or “because people don’t really want that”. That’s where the mystery begins.
2025-10-30 08:00:00
Here are a few things that seem to be true:
None of those are too controversial, I think. (Let’s stress supposedly in #3.) But if you stare at them for a while, it’s hard to see how they can all be true at the same time.
Because, why do people hate dating apps? People complain that they’re bad in various ways, such as being ineffective, dehumanizing, or expensive. (And such small portions!) But if they’re bad, then why? Technologically speaking, a dating app is not difficult to make. If dating apps are so bad, why don’t new non-bad ones emerge and outcompete them?
The typical answer is network effects. A dating app’s value depends on how many other people are on it. So everyone gravitates to the popular ones and eventually most of the market is captured by a few winners. To displace them, you’d have to spend a huge amount of money on advertising. So—the theory goes—the winners are an oligopoly that gleefully focus on extracting money from their clients instead of making those clients happy.
That isn’t obviously wrong. Match Group (which owns Tinder, Match, Plenty of Fish, OK Cupid, Hinge, and many others) has recently had an operating margin of ~25%. That’s more like a crazy-profitable entrenched tech company (Apple manages ~30%) than a nervous business in a crowded market.
But wait a second. How many people go to a speed dating event? Maybe 30? I don’t know if the speed dating “resurgence” is real, but it doesn’t matter. Some people definitely do find love at real-life events with small numbers of people. If that’s possible, then shouldn’t it also be possible to create a dating app that’s useful even with only a small number of users? Meaning good apps should have emerged long ago and displaced the crappy incumbents? And so the surviving dating apps should be non-hated?
We’ve got ourselves a contradiction. So something is wrong with that argument. But what?
Perhaps speed dating attendees are more likely to be good matches than people on dating apps. This might be true because they tend to be similar in terms of income, education, etc., and people tend to mate assortatively. People who go to such events might also have some similarities in terms of personality or what they’re looking for in a relationship.
You could also theorize that people at speed dating events are higher “quality”. For example, maybe it’s easier to conceal negative traits on dating apps than it is in person. If so, this might lead to some kind of adverse selection where people without secret negative traits get frustrated and stop using the apps.
I’m not sure either of those are true. But even if they are, consider the magnitudes. While a speed dating event might have 30 people, a dating app in a large city could easily have 30,000 users. While the fraction of good matches might be lower on a dating app, the absolute number is still surely far higher.
Perhaps even if you have fewer potential matches at a speed dating event, you have better odds of actually finding them, because in-person interactions reveals information that dating apps don’t.
People often complain that dating apps are superficial, that there’s too much focus on pictures. Personally, I don’t think pictures deserve so much criticism. Yes, they show how hot you are. But pictures also give lots of information about important non-superficial things, like your personality, values, social class, and lifestyle. I’m convinced people use pictures for all that stuff as much as hotness.
But you know what’s even better than pictures? Actually talking to someone!
Many people seem to think that a few minutes of small talk isn’t enough time to learn anything about someone. Personally, I think evolution spent millions of years training us to do exactly that. I’d even claim that this is why small talk exists.
(I have friends with varying levels of extroversion and agreeableness, but all of my friends seem to have high openness to experience. When I meet someone new, I’m convinced I can guess their openness to ±10% by the time they’ve completed five sentences.)
So maybe the information a dating app provides just isn’t all that useful compared to a few minutes of casual conversation. If so, then dating apps might be incredibly inefficient. You have to go through some silly texting courtship ritual, set up a time to meet, physically go there, and then pretend to smile for an hour even if you immediately hate them.
Under this theory, dating apps provide a tiny amount of information about a gigantic pool of people, while speed dating provides a ton of information about a small number of people. Maybe that’s a win, at least sometimes.
Maybe the benefit of real-life events isn’t that they provide more information, but that they change how we behave.
For example, maybe people are nicer in person? Because only then can we sense that others are also sentient beings with internal lives and so on?
I’m pretty sure that’s true. But it’s not obvious it helps with our mystery, since people from dating apps eventually meet in person, too. If they’re still nice when they do, then this just resolves into “in-person interactions provide more information”, and is already covered by the previous theory. To help resolve our mystery, you’d need to claim that people at real-life events act differently than they do when meeting up as a result of a dating app.
That could happen as a result of a “behavioral equilibrium”. Some people take dating apps seriously and some take them casually. But it’s hard to tell what category someone else is in, so everyone proceeds with caution. But by showing up at an in-person event, everyone has demonstrated some level of seriousness. And maybe this makes everyone behave differently? Perhaps, but I don’t really see it.
I can think of a few other possible explanations.
Maybe speed dating serves a niche. Just like Fitafy / Bristlr / High There! serve people who love fitness / beards / marijuana, maybe speed dating just serves some small-ish fraction of the population but not others.
Maybe the people who succeed at speed dating would also have succeeded no matter what. So they don’t offer any general lessons.
Maybe creating a dating app is in fact very technologically difficult. So while the dating apps are profit-extracting oligopolies, that’s because of technological moat, not network effects.
I don’t really buy any of these.
So what’s really happening? I am not confident, but here’s my best guess:
Selection is not a major factor.
The high bandwidth of in-person interactions is a major factor.
The fact that people are nicer or more open-minded in person is not a major factor, other than through making in-person interactions higher bandwidth.
None of the obscure theories are major factors.
Dating apps are an oligopoly, driven by network effects.
Basically, a key “filter” in finding love is finding someone where you both feel optimistic after talking for five minutes. Speed dating is (somewhat / sometimes) effective because it efficiently crams a lot of people into the top of that filter.
Meanwhile, because dating apps are low-bandwidth, they need a large pool to be viable. Thus, they’re subject to network effects, and the winners can turn the screws to extract maximum profits from their users.
Partly I’m not confident in that story just because it has so many moving parts. But something else worries me too. If it’s true, then why aren’t dating apps trying harder to provide that same information that in-person interactions do?
If anything, I understand they’re moving in the opposite direction. Maybe Match Group would have no interest in that, since they’re busy enjoying their precious network effects. But why not startups? Hell, why not philanthropies? (Think of all the utility you could create!) For the above story to hold together, you have to believe that it’s a very difficult problem.
2025-10-23 08:00:00
I recently wondered if explainer posts might go extinct. In response, you all assured me that I have nothing to worry about, because you already don’t care about my explanations—you just like it when I point at stuff.
Well OK then!
How did Michelangelo make this?
What I mean is—marble is unforgiving. If you accidentally remove some material, it’s gone. You can’t fix it by adding another layer of paint. Did Michelangelo somehow plan everything out in advance and then execute everything perfectly the first time, with no mistakes?
I learned a few years ago that sculptors have long used a simple but ingenious invention called a pointing machine. This allows you to create a sculpture in clay and, in effect, “copy” it into stone. That sounds magical, but it’s really just an articulated pointer that you move between anchor points attached to the (finished) clay and the (incomplete) stone sculpture. If you position the pointer based on the clay sculpture and then move it to the stone sculpture, anything the pointer hits should be removed. Repeat that thousands of times and the sculpture is copied.
I was sad to learn that Michelangelo was a talentless hack, but I dutifully spent the last few years telling everyone that all sculptures were made this way and actually sculpture is extremely easy, etc.
Last week I noticed that Michelangelo died in 1564, which was over 200 years before the pointing machine was invented.
Except, apparently since ancient times sculptors have used a technique sometimes called the “compass method” which is sort of like a pointing machine except more complex and involving a variety of tools and measurements. This was used by the ancient Romans to make copies of older Greek sculptures. And most people seem to think that Michelangelo probably did use that.
I think this is one of the greatest data visualizations ever invented.
Sure, it’s basically just a histogram turned on the side. But compare India’s smooth and calm teardrop with China’s jagged chaos. There aren’t many charts that simultaneously tell you so much about the past and the future.
It turns out that this visualization was invented by Francis Amasa Walker. He was apparently such an impressive person that this invention doesn’t even merit a mention on his Wikipedia page, but he used it in creating these visualizations for the 1874 US atlas:
I think those are the first population pyramids ever made. The atlas also contains many other beautiful visualizations, for example this one of church attendance:
Or this one on debt and public expenditures:
If you haven’t heard about the British Post Office scandal, here’s what happened: In 1999, Fujitsu delivered buggy accounting software to the British Post Office that incorrectly determined that thousands of subpostmasters were stealing. Based on this faulty data, the post office prosecuted and convicted close to a thousand people, of whom 236 went to prison. Many others lost their jobs or were forced to “pay back” the “shortfalls” from their own pockets.
Of course, this is infuriating. But beyond that, I notice I am confused. It doesn’t seem like anyone wanted to hurt all those subpostmasters. The cause seems to be only arrogance, stupidity, and negligence.
I would have predicted that before you could punish thousands of people based on the same piece of fake evidence, something would happen that would stop you. Obviously, I was wrong. But I find it hard to think of good historical analogies. Maybe negligence in police crime labs or convictions of parents for “shaken baby syndrome”? Neither of these is a good analogy.
One theory is that the post office scandal happened because the post office—the “victim”—had the power to itself bring prosecutions. But in hundreds of cases things were done the normal way, with police “investigating” the alleged crimes and then sending the cases to be brought by normal prosecutors. Many cases were also pursued in Scotland and Northern Ireland, where the Post Office lacks this power.
Another theory would be:
Prosecutors have incredible latitude in choosing who they want to prosecute.
Like other humans, some prosecutors are arrogant/stupid/negligent.
It’s actually pretty easy for prosecutors to convict an innocent person if they really want to, as long as they have some kind of vaguely-incriminating evidence.
Under this theory, similar miscarriages of justice happen frequently. But they only involve a single person, and so they don’t make the news.
I link to this not because it’s interesting but because it’s so impressively incomprehensible. If there’s someone nearby, I challenge you to read this to them without losing composure.
In zoological nomenclature, a type species (species typica) is the species whose name is considered to be permanently taxonomically associated with the name of a genus or subgenus. In other words, it is the species that contains the biological type specimen or specimens of the genus or subgenus. A similar concept is used for groups ranked above the genus and called a type genus.
In botanical nomenclature, these terms have no formal standing under the code of nomenclature, but are sometimes borrowed from zoological nomenclature. In botany, the type of a genus name is a specimen (or, rarely, an illustration) which is also the type of a species name. The species name with that type can also be referred to as the type of the genus name. Names of genus and family ranks, the various subdivisions of those ranks, and some higher-rank names based on genus names, have such types.
In bacteriology, a type species is assigned for each genus. Whether or not currently recognized as valid, every named genus or subgenus in zoology is theoretically associated with a type species. In practice, however, there is a backlog of untypified names defined in older publications when it was not required to specify a type.
Can such a thing be created unintentionally? I tried to parody this by creating an equally-useless description of an everyday object. But in the end, I don’t think it’s very funny, because it’s almost impossible to create something worse than the above passage.
A funnel is a tool first created in antiquity with rudimentary versions fabricated from organic substrates such as cucurbitaceae or broadleaf foliage by early hominid cultures. The etymology of fundibulum (Latin), provides limited insight into its functional parameters, despite its characteristic broad proximal aperture and a constricted distal orifice.
Compositionally, funnels may comprise organic polymers or inorganic compounds, including but not limited to, synthetic plastics or metallic alloys and may range in weight from several grams to multiple kilograms. Geometrically, the device exhibits a truncated conical or pyramidal morphology, featuring an internal declination angle generally between 30 and 60 degrees.
Within cultural semiotics, funnels frequently manifest in artistic representations, serving as an emblem of domestic ephemerality.
The good news is that the Sri Lankan elephant is the type species for the Asian elephant, whatever that is.
I previously mentioned that some hormonal medications used to be made from the urine of pregnant mares. But only after reading The History of Estrogen Therapy (h/t SCPantera) did I realize that it’s right there in the name:
Premarin = PREgnant MARe’s urINe
If you—like me—struggle to believe that a pharmaceutical company would actually do this, note that was in 1941. Even earlier, the urine of pregnant humans was used. Tragically, this was marketed as “Emmenin” rather than “Prehumin”.
