Do you work in Venture Capital? I need a reviewer for a long upcoming post. Email me for details.

Replying to Robert Wiblin on Young Rationalists

[Edit 2021/02/07] I got a couple ages wrong in the original spreadsheet. Mean/median age at founding was previously listed as 26, but is actually 27. Mean/median age in 2020 was listed as 36 and 35, but are actually both 37. This very slightly weakens my arguments. The historgrams were also off and have been corrected. Thanks to Gytis Daujotas for catching these.

In response to Where are all the Successful Rationalists, Robert Wiblin writes:
The EA community seems to have a lot of very successful people by normal social standards, pursuing earning to give, research, politics and more… Typically they aren’t yet at the top of their fields but that’s unsurprising as most are 25-35.

He’s basically right. On the SSC 2019 survey, the median reader was 30. [1]

So Wiblin’s right, but his comment begs a much more important question: why do rationalists think being young is incompatible with being at the top of your field?

Take for example, Patrick Hsu. He’s 29, an assistant professor at Berkeley, and has his name on several seminal CRISPR papers. He reads and endorses Alexey Guzey so he’s clearly into weird blogs on the internet. It’s not unreasonable to think that people like this should be part of the rationality community.

Or to take a more visible case, consider all the very young startup founders.

Brian Timar looked at the companies he’s interested in, and found that the average and median age at founding was 30. If you scan YC’s top 100 companies, many of the founders were under 35 when they started:

92% of founders were under 30 when they started their companies, mean and median age were both 27.

But okay, we might not expect to know about founders who are just getting started. You’re not successful until your company actually succeeds. So how old are those same founders now? Still pretty young.

38% are still under 35, and 85% are under 40. Mean and median are both 37.

Data for both charts here.

Granting Wiblin’s point about youth, I’ll ask again: where are all the successful rationalists?

It gets a lot worse when you consider that ideologically, rationalists should be uniquely well positioned to start a company.

Linear Returns from Wealth
For a normal person, the expected financial value of a startup may be high, but the expected returns to personal happiness are very low. A billion dollars will probably only make you a little bit happier than million, so it makes sense to be risk averse and keep your day job. But for a rationalist utilitarian, returns from wealth are perfectly linear! Every dollar you earn is another dollar you can give to prevent Malaria. So when it comes to earning, rationalists ought to be risk-neutral, and skew more heavily than normal people towards starting companies.

Willingness to be Weird and Lame
Paul Graham says “One of the biggest things holding people back from doing great work is the fear of making something lame.” Meanwhile, rationalists totally disregard this tendency to spend years working on things no one else cares about. Will MacAskill has been speaking for years on the importance of keeping EA weird, but even before that there was a decade of posts about nootropics, longevity and superintelligence, long before it was cool.

Another panel from back in 2009: “I think we can fairly say that we’re all, Peter maybe less so, not afraid to being weird”

Low Probability High Return Bets
Rationalists are also already into doing things with a tiny probability of huge impact, as described in OpenPhil’s Hits-based Giving. This is the entire justification for caring about this like AI Safety. You multiply out the probabilities, and preventing even a 1-in-a-thousand chance of extinction turns out to be a very effective use of time.

Slant Towards Tech
Many rationalists are in the UK, but today the epicenter skews more towards the Bay Area. As I wrote last time, 40% work in software engineering, so they should be relatively well poised to start companies.

So seriously, where are all the successful rationalists?

Many qualities are ascribed to startup founders. Visionary, optimist, contrarian, workaholic.

What you don’t hear is founders praised for their intellectual honesty.

That shouldn’t come as a surprise. It’s no secret that you need a kind of unreasonable self-confidence to pitch VCs. Less discussed but analogous is the process of recruiting early employees. It is not very compelling to offer: “Come work for me, if we’re incredibly lucky there’s a miniscule chance you’ll get .1% of a billion dollar company which comes out to $500 thousand after dilution, $250 thousand after tax, over 4 years, minus the strike price.”

Alex Danco asks Are Founders Allowed to Lie. It’s a good piece and you should read it, but the fact that he even has to ask means it’s possible the answer is “yes”, which is just not something any normal industry says about its leaders.

The easiest explanation is that founders really are just consciously manipulative, but I worry we’re underestimating how hard that would be. It takes enormous energy to maintain a lie, and tremendous sociopathy to do so consciously. From what I can tell, a lot of these founders are actually disproportionately philanthropic. I can’t rule out that this is just a PR move or whatever, but this whole idea just feels somewhat extreme and conspiratorial.

So okay, maybe it’s unconscious? Maybe founders are uniquely out of touch with reality and genuinely believe that they’re very likely to beat the overwhelming odds against them?

Again, this doesn’t feel right. Sure, you have to be optimistic, but you can’t be straight up delusional and continue to function at a very high level. Kara Swisher and Elon have a great exchange about this, maybe my favorite moment in any interview ever:

[KS:] What about things that are just critical of you that you don’t like? Do you think you’re particularly sensitive?

[EM:] No. Of course not. Count how many negative articles there are and how many I respond to. One percent, maybe. But the common rebuttal of journalists is, “Oh. My article’s fine. He’s just thin-skinned.” No, your article is false and you don’t want to admit it.

Do you take criticism to heart correctly?


Give me an example of something if you could.

How do you think rockets get to orbit?

That’s a fair point.

Not easily. Physics is very demanding. If you get it wrong, the rocket will blow up. Cars are very demanding. If you get it wrong, a car won’t work. Truth in engineering and science is extremely important.

Right. And therefore?

I have a strong interest in the truth.

If founders aren’t liars or delusional, what could explain their seemingly irrational optimism?

Rather than general dishonesty, my theory is that founders neglect one kind of reasoning very specifically. The same kind most rationalists are obsessed with: taking the outside view.

I’m using “outside view” as a kind of general term for meta-level thinking, consulting base rates, or using bayesian epistemology. Basically, it means not trusting your first-order estimates too much, looking around to see whether or not those estimates are justified, and reasoning “from behind the veil”. As Inadequate Equilibria describes it:

Modest epistemology doesn’t need to reflect a skepticism about causal models as such. It can manifest instead as a wariness about putting weight down on one’s own causal models, as opposed to others’….

If we were fully rational (and fully honest), then we would always eventually reach consensus on questions of fact. To become more rational, then, shouldn’t we set aside our claims to special knowledge or insight and modestly profess that, really, we’re all in the same boat? [2]

Here’s a more concrete example: A rationalist has a good startup idea, so they set out to calculate expected value. YC’s acceptance rate is something like 1%, and even within YC companies, only 1% of them will ever be worth $1 billion. So your odds of actually having an exit of that magnitude are 10,000 to 1, and then you’re diluted down to 10% ownership and taxed at around 50%. Of course, there are exits under and above a billion, but back-of-the-napkin, you’re looking at an expected $5,000 for 10+ years of work so grueling that even successful founders describe it as “eating glass and staring at the abyss“. [3]

This is so deeply ingrained in my head as the rational way to think, that it took me a long time to realize that other people just fundamentally don’t approach problems this way. I would venture to guess that the normal train of thought is closer to: “Most startups fail, but that’s because their ideas are bad. Since my idea is very good, I’ll neglect the base rates. I am very special.”

Does that sound mean? It shouldn’t. There’s nothing wrong with thinking that you’re special. It’s not a moral belief, or a claim to entitlement. It’s just the understanding that you are not a median member of the general population, so base rates about “anyone who has ever applied to YC” don’t apply.

The rationalist sees the 1% acceptance rate and gets intimidated. Normal people see that applying to YC explicitly does not require a business plan, incorporation, existing revenue, or an introduction, and understand that any idiot with a couple hours can will out a web form. Accordingly, they totally ignore the base rate.

That’s all the acceptance part of starting a company, but much more important is actually coming up with an idea you believe in. Rationalists tend to accept the Efficient Markets Hypothesis. They look at an industry, think “what are the odds I know more than people who have done this for a decade?” and assume any seeming inefficiencies are just a Chesterton’s Fence.

That’s not what normal people do at all. Normal people look at an industry, they see a gross inefficiency staring at them in the face, and they think “wow, that’s grossly inefficient!”

And then sometimes, they even set out to solve it.

[1] If the median rationalist is now 30, and Yudkowsky started writing in 2007, was his audience mostly teenagers?

[2] I’m not exaggerating. In fact, this is a massive oversimplification of the estimates of startup success rationalists actually put together.

[3] To be fair, Yudkowsky is specifically attempting to correct against modest epistemology, concluding with an exhortation to not take the outside view so much and instead “spend most of your time thinking about the object level”. To be clear, this is not his solution for all humans, nor his model of perfect rationality. It’s targeted specifically at the kinds of people who read this book and who he believes are a) disproportionately likely to overvalue the meta level and b) disproportionately likely to have good object level beliefs.

Revising my Views on the Impact of Teachers

In No One is Even Trying, I wrote that Grant Sanderson’s Youtube channel 3B1B has 161 million views. Based on how Youtube counts “views”, that’s between 1.3 and 21.4 million hours of learning.

This is incredible to me. If you asked me last night who my hero is, I would have said Grant Sanderson. If you asked me what’s wrong with the world, I would have said that there aren’t more people like him.

But maybe there are.

A regular teacher, teaching 8am–3pm 180 days a year to a class of 20 is producing 25,000 learning hours a year! Over a 50 year career, that’s 1.3 million, same as Sanderson, though in 10x as much time.

It’s even more surprising to compare a teacher with Khan Academy. Depending on how you count “views” they’re at 15 million to 120 million learning hours. But they have 600 employees, and they’ve been around for 12 years. So that’s best case 17,000, worst case 2,000 learning hours per employee per year, compared to a regular teacher’s 25,000.

That’s a little unfair, Khan Academy hasn’t had 600 employees for the full 12 years. But if we assume linear growth, they’ve averaged 300 employees, which would increase their learning hours to 4,000 (best case 33,000).

To be clear, the point is not that Khan Academy is bad (they have way more than lectures, and they’re available to anyone with an internet connection for free). The point is that seemingly non-scalable things like teaching can still multiply out to surprisingly large impact.

Since views counts on Youtube are legible and quantified, it’s really easy to see 100,000,000 and feel really impressed. But counting a view as 30 seconds, a teacher with 20 students and a 50 year career could hit 1.3 million learning hours, the equivalent of 151 million “views”!

Until now, I didn’t have that much respect for teachers. I understood that they’re probably good people, and a lot of them teach as a labor of love, but I didn’t appreciate their impact.

In fact, teaching has been my go-to example for illustrating how effective altruism differs from lay person opinions. Teaching seems like a great job, but if you think about it through an EA lens, it’s just not very impactful. 80,000 hours has very negative reviews of teaching, and another article of reasons not to go into education.

Nothing I’ve learned recently invalidates those articles. And yet, it’s hard to reconcile my belief in the seeming inefficacy of teaching jobs with my reverence for Sanderson. As always, one man’s modus ponens is another man’s modus tollens. I can have whichever belief I want, but not both of them at once. Either teachers are heroes, or Sanderson is merely moderately praiseworthy.

To further drive this point, consider the scenario:

You’re driving through a small town, and while stopping for lunch, hear rumors of a teacher who delivers unusually engaging lectures. As someone interested in pedagogy and social impact at scale, you decide to go check it out in person, and sure enough, the teacher’s amazing!

What would your first thought be? If you’re like me, your first conscious reaction would be “we’ve got to get this guy on Youtube!” If you’re like me, you see it as a horrendous waste of potential that this brilliant educator is stuck in a small town when he could be on the internet, delivering content to the masses at scale.

So you talk to him, set up a small film crew, convince him to take time off teaching to record Youtube lectures, and sure enough it’s a hit! Within your first few years, he’s up to 10 million views.

And yet, it turns out that intuition is totally wrong. He was already clocking in “10 million views” every few years. Of course there are benefits of being online. The content is recorded, the students could be anywhere in the world, you can go back if you missed a section. But none of that feels sufficient to justify my initial reaction.

Maybe I’m just a shameless technophile retroactively justifying my beliefs, but I think there’s still a good reason to revere Sanderson.

I love 3B1B videos because they’re creative and engaging in a way that math never was in school. It genuinely feels like he’s broken through whatever obfuscation prevents kids from understanding math, and discovered better ways of explaining key concepts. And because math is so fundamental to reasoning, this is more than a good educational channel. It feels almost like a leap forward for civilization.

But still, that’s not a metric effective altruism cares about. Maybe the real modus tollens is to abbandon the value system.

No One is Even Trying

This post is about how little work people do, but it starts with stories of unusually high output.

Gory Movies
Quentin Tarantino is the legendary directory of America’s most violent mainstream movies. For a while, he’s been talking about retiring after his 10th film. If you count the Kill Bill franchise as 1, he’s at 9 so far, and now claims the next will be his last.

Meanwhile, Japanese director Takashi Miike has made literally 10 times as many movies, finishing his 100th in 2017. His movies are similarly violent, so he’s been compared to Tarantino, and actually cast the American director in his wacky Sukiyaki Western Django.

Obviously, the movies are not all good. His worst movie has a 0% audience score on Rotten Tomatoes (98 reviews). In contrast, Tarantino’s worst movie is Once Upon a Time in Hollywood which sits at a relatively illustrious 74%.

But it would be a mistake to write off Miike on that basis. Take a look at their top movies instead. Despite its lows, Miike’s filmography also contains 21 movies with audience scores above 70%, twice as many as Tarantino. Here are each director’s top ten films by Audience Score and Critic Score:

(Taranito’s movies tend to have way more reviewers. Data here.)

There are a couple ranks where each director pulls ahead, but overall it’s pretty even.

The point is, Miike’s managed to keep up with Tarantino’s top 10 while simultaneously churning out another 90 movies. When asked about this output in a recent interview, Miike says:

I never set out to make 100 movies. And I never never have a personal motto to make lots of movies, either. I just started making movies, and kept making them at my pace…. But at the same time, I’m lazy.

Online Education
3Blue1Brown is a math education Youtube channel run by Grant Sanderson. It spans interactive livestream lectures, visualizations, and approachable explanations for difficult math concepts.

He also has 161 million views, which is just insanely high for a math teacher.

For context, Khan Academy is at 1,806 million, which is 10x higher, but it also has 600 employees! (Not to mention funding from Bill Gates, Carlos Slim and Google.)

In terms of views per employee, that puts Sanderson ahead by a factor of 60. Of course, Khan Academy does more than just produce videos, they also have a gamified app, problems sets and so on.

But still, 60x is just enormous.

It’s worth noting also that KA videos tend to be short because they break each concept into bite sized pieces, around 4 minutes each. Sanderson tends more towards extended essays on a topic, averaging around 16 minutes per video. So if you care about total learning minutes, it’s more like 240x.

Khan Academy has not always had 600 employees, but it’s also been around 12 years to Sanderson’s 5. So if you assume linear-ish growth and an average of 300, and then account for age, these corrections are a wash.

I don’t even know who the point of comparison is here. There are plenty of low-output writers to go around, I’m not going to call anyone out in particular.

But at the far other end of the spectrum, take a look at Byrne Hobart. He’s published 5 days a week, every single week since he launched his newsletter. His last 5 posts are 1968, 2281, 3352, 2535 and 2586 words, so 12,722/week, times 52 weeks, is 661,544 words each year. But the really impressive thing is that he’s simultaneously writing for Coindesk, Medium and Palladium. No, there isn’t any overlap or self-plagiarization.

And then, as I wrote as an edit to my last post, there are people like Heather Cox Richardson who publish literally every day, including weekends. Her word count comes in at 1600, a bit below Byrne, but that’s still 11,200/week. She’s one of the highest paid authors on Substack, second only to The Dispatch, a 15 person team of professional journalists with decades of experience.

Oh right, and she’s simultaneously a Professor at Boston College, authored How the South Won the Civil War, and co-authored Voter Suppression in U.S. Elections, both of which were published this year. I can’t find any mention of her being on sabbatical, and according to some informal estimates, professors work somewhere in the range of 40-60 hours a week.

What’s the deal? How is Richardson able to compete with 15 journalists while running her newsletter as a side project?

It’s worth briefly playing devil’s advocate about each case’s particulars:

  • Miike makes a ton of movies, but a lot of them are garbage. Maybe as long as you try enough times, you occasionally do exceptionally well. Or maybe “director” doesn’t entail the same workload in Japan, and Miike’s delegating to producers, assistants, cinematographers and so on.
  • Sanderson is unusually popular, but that’s not evidence that he’s actually more productive. Maybe he’s discovered a particularly compelling video format, and he’s milking it for all its worth. That’s still laudable, but it’s a different feat that the ones I’m describing.
  • Richardson’s books published this year were probably written in past years, and her Substack is relatively new so it’s not like she was doing both at once. If you can type 80 words per minute and write stream of consciousness, 11,200 words is just 20 minutes a day.

But still, none of that helps explain why there aren’t way more people like this, or why the contrasts are so stark.

The easy explanation is that they just work harder, but in some cases that just doesn’t add up. Danielle Steel has written 179 books, and she reportedly works 20 hours a day. So she’s basically at the limit of what’s humanly possible. But even then, she’s working maybe 2.5x harder than a normal author, and publishing 10 times as many books. By total sales, she’s now the world’s best selling author.

Why does it feel like no one else is even close?

One explanation is that we’re stuck in a double bind. I wrote in Bus Factor 1 that large companies are designed specifically to ensure that no one has personal responsibility, and in Quitting Won’t Save You, I explained some of the surprising difficulties with setting out on your own. Pretty much everyone I know is in one of those two camps.

Escaping the bind takes more than hard work. You have to actually go out and generate your own sense of meaning. You have to build, often without a surrounding social structure, a way of creating value on your terms.

Here’s Danielle Steel from a recent interview:

I grew up in Europe, where it was not considered polite for a woman to be working, and I was married to two different men who did not like that I worked.

Miike’s movies have been banned, not by repressive regimes, but by Finland and Germany. Responding to initial success he said he had

“no idea what goes on in the minds of people in the West and I don’t pretend to know what their tastes are. And I don’t want to start thinking about that. It’s nice that they liked my movie, but I’m not going to start deliberately worrying about why or what I can do to make it happen again”

What I’m getting at, is neither of them seems to be powered by social validation, but they’ve found a way to work pretty much constantly anyway.

But laboring in obscurity is only half the battle. The other half is continuing to do good things once you’re popular. A normal person’s natural instinct is to do everything in their power to maintain their popularity, cater to their audience, and thus become boring. At some point, you are not even a creator, just a fleshy vessel for an idea that already exists independently, and acts through you only as long as you do not exert too much personal will.

It’s hard to express that sentiment in a way that doesn’t sound trite, so let me end with personal testimony. I think a lot about giving up writing. Every day since I started this blog, I have woken up, and wanted to quit. I find myself scanning job boards, or polishing my resume, even though I’ve told myself a million times that I don’t want to go back.

And then a few days ago, something changed. I was featured in Marginal Revolution, then on Hacker News, then on Marginal Revolution a second time, and then Byrne Hobart’s newsletter twice. Practically overnight, there was an influx of subscribers, and more supportive emails than I’ve had time to reply to. From close to 0, I suddenly found myself with a powerful source of external validation.

And it’s horrible. I check the subscriber count 10 times a day. When I sit down to write, I imagine you all judging my work. I’ve read every comment, including all the totally banal ones.

So in one sense, I’m free from the double mind. I have a sense of personal responsibility for my success, and a sense of social validation to go with it. I just don’t know where I go from here to actually produce great work.

I don’t have a neat ending, but in the spirit of Learning in Public in Real Time, I’m recording these thoughts now, and trust that I’ll figure it out later.