I can't express strongly enough how much I wish we had something like the Civilization Conservation Corps today. We really need to let go of college as the North star for every child - the credentialing issue trickles all the way down to elementary school so we have second graders worried about getting Bs. Not everyone is meant to be an entrepreneur or an academic! But we've so devalued (literally) manufacturing, public service, farming, etc, it's going to be a tough sell. But like...we'll always need people to fix the roads and grow food and be social workers.
An underappreciated point about the Oxbridge system is that a student’s class depends mostly on end-of-year exams, and exam scripts are anonymized before being marked. Often the lecturer will not be the person doing the marking, and there are procedures in place to make sure that individual courses are not too easy or too hard (although those procedures are not perfect). So there is no sense in which different professors are competing with each other to give lower/higher grades, which removes much of the pressure for grade inflation.
I was horrified when I learned that in the US it’s normal for students to go to a TA or professor and beg for higher grades on individual pieces of work. That system seems to be open to so much abuse.
I really think they've got it figured out at Oxford. It's a great call out -- the fact that the exams more or less happen at the very end means the entire rest of the time can _only_ be used for education. In this way the higher education purpose of the university remains paramount. But I also think that a degree from Oxford, though prestigious, is probably less of a direct hiring credential than one from an Ivy League in the US. Most of the credentialist value of the Oxford degree is baked into just being accepted to the university in the first place; American companies don't know how to think through the rest of their educational structure.
(Less true outside of the US. In the UK, Europe, and India, if you mentioned you went to Oxford, people would immediately think of you differently and treat you with more respect. No one knows any of the Ivy leagues besides Harvard, but everyone knows Oxford.)
So. Gonna be a bit contrarian. I taught CS - in college - once upon a time. I taught a class - Formal Languages and Finite Automata - that the students widely thought was a waste of time. "Why do we need to learn this?" I heard constantly. As you say, arrogant little shits.
But they were also smart and aware and they would graduate and get jobs and then come back a year after graduation and drop by my office and tell me "finite state machines are everywhere!". I would just nod. That's why we taught them. They wouldn't ever see it or think it valuable if we hadn't taught them about them.
But you know, we are teaching people technical skills. Things that would make them better engineers or scientists.
Knowing about finite-state machines does not make you a better entrepreneur. Database-backed websites are really easy to build. You can do it without knowing about FSM. You can hire people to do it and know even less.
So everybody has their favorite story about a college dropout that made billions in the computer industry. I'm not sure, but I think that window has closed. Most startups go nowhere. Some get acquired and the rewards for founders are sufficient for the founders to try again. It's the bankers making the money. Maybe you have the next pet rock, I mean Facebook, and you will make billions. The odds aren't there, though.
However, and this is something that we sort of agree on, I learned and taught this stuff because I loved it. Because I thought it was fascinating in its own right. Because it was some fundamental knowledge of how the universe worked. Like math. And it would have been nice to have more students with that same attitude.
Most of my feelings about the uselessness of my college courses come from my time at Google research rather than my time in the startup world. Part of this is because I went hard into deep learning, a subject that at the time was not given any importance in college classes that were explicitly about ML (we spent a month on Gaussian models and two days on neural networks). But part of it was because I quickly found that it was much easier to learn in a focused goal driven environment than in a classroom setting.
Here's an example. I took a class on linear algebra and forgot all of it because I was studying for a very abstract test. There was no intuition building, no world modeling. What made the linear algebra eventually stick many years later was working on real world problems that actually required some understanding of linear algebra. But that understanding wasn't abstract, it was extremely applied -- "this matmul represents a camera coordinate shift in 3d space", things like that. The thing is, a bunch of those applied uses of linear algebra led to a more abstract understanding that I could then apply in a bunch of other places. It was practical AND educational.
I also took a class that taught me about FSMs, and not only have I not really encountered a time where that classwork proved useful, I have completely forgotten all of that classwork _anyway_, so even if I did find such a use case I would have to start from scratch. Maybe I'm unique in this, but I suspect I'm not. Most people learn on the job. In many ways, building expertise through usage is really efficient, because you're only ever learning stuff that is immediately useful, to the level that it's immediately useful. If you need to use something over and over again -- like linear algebra for me -- you'll build the mental models you need through the usage. Classes by contrast always lag behind, so you'll often be learning stuff that is outdated by the time you hit the real world.
--
I do agree that education is valuable on its own accord. I wish that more classes focused on inspiring wonder and awe and genuine interest in a field. But I think that as long as education systems and credentialing systems are in the same place, you cannot have that. The corporation demands legibility.
One of the things I remember from my time at an Ivy was the commentary around Brown's CS major. People said that you could graduate with a CS major from Brown with a lot of flexibility in which classes you wanted to take. But this was framed as a _bad_ thing. A common refrain was "o ya recruiters don't really like hiring from Brown because they don't really know what those students have actually learned." The same tension plagues Oxbridge. That is a learning environment that is basically as free form and self directed as you can imagine. For self starters who are legitimately curious about the world, it's a beautiful system. But the average corporation just doesn't know what to do with the mishmash of "classes" the Oxbridge student takes. It doesn't fit into a neat box, so it's not derisked as much, so the value of the credential is less.
We are in complete concord with regard to the conflict between credentialing systems and educational systems.
I have been musing this morning on how state systems are ubiquitous, not just in computing, but in, for instance, quantum mechanics and biology. Neurons embody state. To be fair, I never tried to motivate it that way or make that connection when I was teaching it.
I submit that while you may not remember the stuff from your FSM class, but the work you did for it changed you in a way that is valuable. This is "classwork as fitness exercise". Football players do pushups and run through tires, even though during a game they do neither of these things.
One final note. Apparently you went to Brown? For all the complaining about how "employers don't like Brown graduates" you got a job at Google, and quite a good one, I think.
The employers who don't like Brown, I would guess, are offering jobs that are narrow, rigid and stultifying. Not something a Brown grad would like.
I went to Columbia 😂 maybe it was just cope to make us Columbia grads feel better than the Brown grads. There was a lot of that too.
I did get a job at Google, but it was not due to the *educational* value of my degree. There was definitely a class of person who made it through to the phone screens because of the *credential* value of the degree, but even that didn't apply to me -- I ended up getting through the phone screen because I built something cool at Hack Rutgers and one of the Google folks there took notice.
> I have been musing this morning on how state systems are ubiquitous, not just in computing, but in, for instance, quantum mechanics and biology. Neurons embody state. To be fair, I never tried to motivate it that way or make that connection when I was teaching it.
When I first introduce people to a new video game -- like Civ or Smash Bros -- I recommend that they pick a leader or character and get really good with that character first. They should play all their games with that same character, and only when they feel like they really grok how to play that character should they move around and explore other ways of playing the game.
Something that I've noticed about myself is that when I became really comfortable / good at one field, I started figuring out all sorts of analogies to how things in that field applied to other, unrelated fields. For me, that's ML -- I got really familiar with embedding spaces and manifolds and topology and linear algebra, and now I can't help but see all sorts of problems in *other* spaces through the same lens. For example, I see a lot of systems as optimization problems and use intuitions developed from how neural networks behave to peer into disparate fields like economics or politics or molecular biology.
I think this is how a lot of people learn things and become generalists -- they start by becoming an expert in one thing, and then use that one thing to get hand holds in adjacent fields, until those adjacent fields also become intuitive and then they go farther out.
So in your case, I could see that through FSMs. And fwiw I totally buy that a deep intuition about FSMs is really useful. I just think its a bit non-unique to FSMs. Being an expert in *anything* is approximately as useful as being an expert in FSMs.
> I submit that while you may not remember the stuff from your FSM class, but the work you did for it changed you in a way that is valuable. This is "classwork as fitness exercise". Football players do pushups and run through tires, even though during a game they do neither of these things.
This got me thinking a bit as I am typing out this response. I actually did end up doing a lot of state-machine-like things -- a lot of my research at Google was about graph neural networks, which of course...graphs. But I really am struggling to even remember what we discussed in my discrete math classes as they related to FSMs; certainly I can't remember a single moment when I was doing graph learning stuff and felt like I could draw a connection to my discrete math class... Unclear. You may be right but its not falsifiable without more data points
I like the "deep dive" as a model/motivator of learning.
One note about discrete math, which I also taught. It is a non-subject. There's no cohesive theme to it, it's just a bunch of little bits of lore that students need to know about for other classes in a CS major.
I also have a math degree and nobody thinks of it as a topic, or sub-branch of mathematics. But it does satisfy a specific need.
When I took it, it was mostly stuff I already knew from other sources. But it was required, so I sat through it.
Anyway, I think this is why you can't remember anything from it.
I agree much is signaling, and double agree that we collectively should be able to test out of a lot more things (I myself was so outraged in undergrad that they wanted me to waste 2 years on "gen eds" so pointless that attendance was part of the grade, that I took nothing but physics and math classes for two years, then CLEPPED out of ~36 credits worth of gen eds, and fortunately by that time had been invited to continue studying under / doing research by my PI).
A couple of thoughts:
I agree that school is probably 90% signaling. But as a founder and employer who's interviewed and hired lots of smart people over the years, the "Ivy" signal is super strong and very worth it for entry level employees. I'm not sure an IQ test can replace it, because IQ tests only test part of what makes somebody good at their job, and conscientious and discipline are both long-term traits that you can't assess in an interview. "Competent, conscientious, and good enough emotional regulation and short and long term planning" is the other important part of the educational signal.
Companies still need a way to identify the best potential workers, customers still need to identify the best service providers, and future workers themselves need some means of reference to understand how they genuinely compare to their competitors in ability and long-running accomplishment.
Even if we overturn Griggs, and allow IQ testing as part of the interview process, and even if we have a lot more industry-and-job-specific tests, it is not a good enough signal.
All FAA(N/M)Gs and finance already do IQ test equivalents, and have been doing for ages. Remember when everyone was treating MSFT or GOOG interview questions like brain teasers several years ago?
I agree that better and more wide ranging testing would help a lot of people, particularly genuinely talented minorities and middle class people who got stuck with crappy schools and so don't show well on paper.
But how do we test for conscientiousness and discipline? Many a wild hair can max on an IQ or job-specific test, but then doesn't deliver solid results over time, are lazy, are inconsistent, fail to see the big picture, fail to map their work back to business value, and more. That's such a prevalent factor, in fact, that Google basically gave up on the brain teaser style interviews! They were optimizing for the wrong thing, and it didn't correlate with actual success in their work life.
Think of this - even the smartest, best companies in the world, with the best modeling and design of experiments and analytics capablities, still bat about 50% in terms of choosing good long-term candidates via the hiring process. And that's WITH being able to skim only the crème de la crème of the entire world, including heavily oversampling Ivy grads.
There's a lot of inherent noise in the system, in other words - and imagine how much noisier it gets when you're down in the "35%" instead of the "top 0.5%" of the tail.
Now imagine that, but at the 50th or 85th percentile instead of 35th. Is there *any* real solution there?? No matter how well calibrated the tests are?
Second point - you bring up kids.
Like the typical Rationalist Stemlord, I hated school from the day it started clear through undergrad. A crushing waste of time, and a grinding Sisyphean ordeal that's more about babysitting and soul crushing than teaching or learning anything.
Obviously, I'm going to unschool / home school my kids, and try to go for an 'elite tutoring' dynamic per Hoel.
But I think founders have a great and underappreciated option here, too. I actually think of Ivy (or equivalents) as the *backup* option, rather than the primary option, because you can lead and teach by example, and any kid that has the capability of doing a startup or business themselves will be much better off doing that rather than spending 4 years in Uni.
Sure, some kids actually like school, and some kids might want to do research (I know I loved research, for all that I hated school). Uni and grad school is a great option for them. But for any with the talent and inclination to actually do things in the world, I think there's a lot of value in mentoring them into a place where they can start and grow a business instead.
As for the arrogance problem, I think it's quite different getting knowledge and funding and advice from your parents in an ongoing way from your childhood, versus being funded / ideologically recruited by a billionaire when you're 18.
I personally worry less about arrogance - I more or less feel that where unmerited, the arrogance will be knocked out of them by the slings and arrows of the world, and if they really are better / more successful on various metrics, I don't see much harm in knowing this. Like I'd be happy if one of my kids became a Thiel fellow or equivalent, because it puts them on the path of doing and having more of a positive impact on the world, and even if they're a*holes in personal life (which hopefully you steered away from in childhood, I agree this is bad), if you found and grow a business big enough to make you rich, I feel like you generally definitionally had a positive impact on the world for millions of people (with some fundamentally adversarial exceptions I've written about before, like the attention economy).
But I do think the whole thing would be sidestepped neatly by the "raising them with business intuition" approach above - then they're being shaped by you both morally and business knowledge wise throughout childhood, and will probably be better off than your 18 year old insufferables.
> But how do we test for conscientiousness and discipline? Many a wild hair can max on an IQ or job-specific test, but then doesn't deliver solid results over time, are lazy, are inconsistent, fail to see the big picture, fail to map their work back to business value, and more. That's such a prevalent factor, in fact, that Google basically gave up on the brain teaser style interviews! They were optimizing for the wrong thing, and it didn't correlate with actual success in their work life.
Strong agree that generally the Ivy signal is a strong signal, both as a founder and Xoogler. That said, I'm not sure though that I am that beat up about diminishing that signal. I think that if the market needs to figure out something else, it will -- especially since it is a *very* important signal with a lot of capital behind it. And I think that most of the Ivy school signal comes from selection effects, which in turn means you should be able to construct testing requirements for businesses that match those that universities give to highschool seniors (why DONT jobs like McKinsey ask their candidates for written essays? Legal liability? Something else?)
I suspect that some of the conscientiousness and discipline you are talking about comes from the network of people you build up when going to an ivy. And having that network is independently useful, especially in a startup setting where knowing smart people helps you bring in other smart people. Still, you could imagine that a free market with a bit more competition in it for credentialing may find ways to get around these problems.
> But for any with the talent and inclination to actually do things in the world, I think there's a lot of value in mentoring them into a place where they can start and grow a business instead.
I'm not convinced that starting a business is the default way to go for the average kid, or really the average person. I do think its important to teach kids how to be independent and give them the tools to think critically, but I think you have to have certain desires / opinions about where you want to be in the world to justify starting a company. I'm someone who thinks of my work as a core part of my identity and so am happy to dive into being a founder. But I have friends and family who just aren't like that -- who see their work as a 9-5 in order to pursue other, more compelling interests, whether that is writing or reading or sports or something else. And I don't think there's anything wrong with that either... tldr a well rounded education seems more important than one specifically tailored towards entrepreneurship. But this is obviously highly subjective and kids are pretty plastic.
> I more or less feel that where unmerited, the arrogance will be knocked out of them by the slings and arrows of the world
Yea, maybe. I just feel like I know a handful of folks who went out to the valley around 18 and have become extremely cynical as a result of it. Something about the experience results in maladjustment. It's not like these kids have bad relationships with their parents either; but it is possible that those parents don't have entrepreneurship experience themselves, and the kids are far from home (e.g. in SF vs in their hometown).
More generally, I think that unschooling and homeschooling works for individual cases but it will never scale up. The university system is in part so successful because it *mostly* works even when you pass hundreds of millions of people through it. My suggestions at the bottom of the essay are trying to get at systems that may work at that scale
> if the market needs to figure out something else, it will -- especially since it is a *very* important signal with a lot of capital behind it.
Yeah, but my point here was that it's probably impossible, the noise is much higher than the signal. Even the literally best companies in the world using the strongest possible signals bat 50%, and that's at the <0.5% percentile of base talent and selection on multiple correlated desirable metrics.
Like if you were just some regular company fishing at the 50th percentile, what hope do YOU have? Approximately zero? I think we'd just be back to nepotism and in-person referrals (which are the best channel for ANY business doing recruiting, IMO, I consider a referral from an employee you respect the strongest possible signal), instead of "selection at arms-length via the educational system," which I agree is slightly nepotistically captured at the top.
> I suspect that some of the conscientiousness and discipline you are talking about comes from the network of people you build up when going to an ivy.
Funnily enough I've been reading about conscientiousness interventions recently - I was skeptical that we can move it at all, and indeed most of the studies are p-hacked garbage, and posit absolutely ridiculous effects like a cohens d=0.5 from a 6-12 week once a week coaching intervention. Lol, as if.
My basic prior going in was "this is going to be like diet and exercise, which approximately nobody succeed at (80% failure to lose 10% and keep it off for a year, 98% failure for 20% over 5 years). I think it's basically around there, but there are actually heterogenous populations who respond and keep up the changes (and they indeed range from 5-20% of the pop, depending on the pop and intervention methods). First jobs are a big one for C buffs, as is military enlistment - school didn't show up in aggregate, but that's not surprising to me, given how lax most school's standards are.
Another mind-blower on this front - did you know Jim O'shaughnessy literally has an in-house AI that he's fed Cialdini's *Influence* and Nir Eyal's *Hooked* so that it optimizes the workflows and habits of his family with intermittent variable rewards to be healthier, etc? He's literally living in the "AI assistant" future we'll all be in in a year or two, already!
But on an Ivy environment's effect on C, I agree with you overall - I think the competitive forge of interacting with, learning from, and competing against other bright talents is one of those environments. And that the network is *impossibly* valuable. I've actually talked about how even in a post-scarcity future, I'd still be trying to get my kids into T10 schools (because then your social networks matter more than anything else! It's all you'll be doing with your time!).
> I'm not convinced that starting a business is the default way to go for the average kid, or really the average person.
Oh yeah, sorry - here I switched from "thinking of societal level ideas" to personal one-to-one communication without signaling a context shift. Like this is what I'm planning for my own kids, and I've recommended it to a few other friends.
I agree there's no way it works at scale, especially in an AGI future. But we're personally highly selected pops, and it's likely at least some of our kids will inherit our predispositions and talents on that front.
> Yeah, but my point here was that it's probably impossible, the noise is much higher than the signal. Even the literally best companies in the world using the strongest possible signals bat 50%, and that's at the <0.5% percentile of base talent and selection on multiple correlated desirable metrics.
It can't *actually* be impossible, right? The ivy leagues figured it out! I don't think my suggestion is "each individual company should institute their own credentialling system". My suggestion is more like "we should have dedicated institutions for credentialling that are completely different from those that do education".
I admit straight up that I have no idea how you could 'test' for conscientiousness, but also I've spent only 1 day thinking about it. I suspect an organization(s) thinking about this 24/7 could come up with something reasonable. Off the top of my head, a battery of personal essays seems like it would potentially help a lot in figuring out this signal.
Now...its also possible that we've naturally evolved towards the best possible signal already i.e. something that looks like an Ivy league school. I'm a bit skeptical that that's really the case, but also culture tends to be smarter than too-smart-for-their-own-good individuals. I'd be curious though if being at a good uni with a bunch of other smart people improves your conscientiousness, seems somewhat easy to study though potentially really hard to untangle from the effects of just, like, getting older and more mature. But I also knew a lot of assholes at my school, they were assholes when we met and assholes when we left *shrug*
Somewhat related: probably the best thing about most high end universities is that it gives a selection of poorer kids access to those same networks, which I feel like may do a lot more for lifting people out of poverty than the credential itself. But we have some kinds of things that replicate that experience. Even though I bash them, I think that VC incubator programs can be similar (though the main risk there is that its explicitly an extractive relationship, whereas the friends you make in college are peers).
> Another mind-blower on this front - did you know Jim O'shaughnessy literally has an in-house AI that he's fed Cialdini's *Influence* and Nir Eyal's *Hooked* so that it optimizes the workflows and habits of his family with intermittent variable rewards to be healthier, etc? He's literally living in the "AI assistant" future we'll all be in in a year or two, already!
Sick! I have to take a look at this
> Oh yeah, sorry - here I switched from "thinking of societal level ideas" to personal one-to-one communication without signaling a context shift.
makes sense. I think I personally am planning to have tutors for my kids up through middleschool and maybe past that. That's more or less what my parents did for me and I think it really helped immensely. I don't think I'm going so far as to homeschool though -- I know a bunch of homeschooled kids and I think the lack of socialization can hurt them.
> Now...its also possible that we've naturally evolved towards the best possible signal already i.e. something that looks like an Ivy league school.
Sure, and what does that process look like?
Looking at some of the other parents I know, it apparently looks like "the competition to get into Harvard starts 6 months before birth, when you need to get on the waiting list for the right exclusive pre-school to give your precious Jansen a leg up, because if you don't get in there, and if you don't bared-teeth-grind furiously and nonstop for the next 18 years, their chances of getting into Harvard are *ruined!*"
Like it's literally a lifelong battle, or at least a decade or 5+ year battle, where both kids and parents exert maximal effort for years in a zero-sum competition and come out the winners.
It seems like that's pretty well tuned to surface genuine conscientiousness AND ambition AND academic talent, etc.
Versus any test you can do can and will be Goodharted immediately. Like there are current Big 5 tests that test for C, just use those, right? Lol, how fast would it take for everyone to 'know' the right answers?
There's some obvious technologically mediated solutions. Everyone installs an app when they're teenagers that tracks what they eat and their exercise behavior and whatever. As a former athlete, I've def been biased towards hiring other former athletes, because I know what it takes on the C and discipline fronts to be competitive. Still only gives you a small buff, doesn't correlate with business value all that much.
Or when everyone has AI assistants, the assistants will DEFINITELY know how naturally conscientious their people are, and how much of an external scaffolding and buff their people need to execute at the level they're currently executing. But who knows, the AI assistants could lie, or be induced to lie, or something.
Certainly if we were given unmediated lifelong data, we could build much better predictive models that could do pretty well. But using those would be pretty fraught, legally, societally and morally. Shades of Gattaca, you know.
I'm definitely not saying "school systems are the best method." We know they're staggeringly wasteful in time and cost, and that they've moved from something like 5 students per administrator to 1:1 students per administrator, which is so ridiculous I can't even talk about it without ranting.
I'm just saying they currently do encode something like a "years deep, unfalsifiable signal" about C, at least at the T20 level, and that it would be fairly hard to surface that signal via other methods that aren't Goodhartable.
> There's some obvious technologically mediated solutions. Everyone installs an app when they're teenagers that tracks what they eat and their exercise behavior and whatever.
That's grim.
Maybe the answer is that we just need less focus on credentials overall. Like, how many jobs *actually* require more than a secondary degree or a high degree of conscientiousness? I'm sure it would be great if the average line cook had intelligence + conscientiousness for the restaurant owner, but it doesn't really benefit the rest of society.
> I'm just saying they currently do encode something like a "years deep, unfalsifiable signal" about C, at least at the T20 level, and that it would be fairly hard to surface that signal via other methods that aren't Goodhartable.
arguably even college has a bit of goodhardt-ness to it, given how people game things like recommendations and sports and testing and so on.
I'd love to see a company take a serious crack at this, though idk what exactly that would look like. One thing that could be interesting -- albeit dystopian -- is to try and get some fine tuned AI classifiers that are really good at figuring out someone's conscientiousness from conversation. Like, you have a conversational AI agent that someone talks to for a period of time, maybe even a long period of time, and then a different agent reads the entire convo and makes a judgement. Very ethically fraught though given how blackbox it would feel.
> Like, how many jobs *actually* require more than a secondary degree or a high degree of conscientiousness? I'm sure it would be great if the average line cook had intelligence + conscientiousness for the restaurant owner, but it doesn't really benefit the rest of society.
I mean, you say that from a consumer perspective, but from the restaurant owner perspective, it's basically the only thing that matters.
I've got a lot of real estate developers in my social circles - do you know how hard it is to find construction people who will A) show up on time reliably, and B) not be high or drunk / start fights / steal tools or materials / deal drugs on the job?
It's *extremely* rare to pass this seemingly minimal qualification threshold. This is why everyone employs immigrants, legal or otherwise, and even that is hit or miss and pretty much everyone has crew shortages all the time.
I've heard similar tales from people trying to hire for manufacturing or factories in the Rust Belt.
I think the floor of "net positive to employ from a pure cost benefit perspective" is probably ABOVE the 30-50th C percentile for most geographies / jobs. And of course, that floor is continually rising as selection effects draw all the high C, high intelligence people to major metros, and as jobs get more complex, and as AGI and robotics comes on the scene.
There's a giant self-sorting effect at work in top ten metros, you know - on intelligence and C and a lot of other metrics. The great majority of the country doesn't live in those top ten metros, though.
> I'd love to see a company take a serious crack at this, though idk what exactly that would look like. One thing that could be interesting -- albeit dystopian -- is to try and get some fine tuned AI classifiers that are really good at figuring out someone's conscientiousness from conversation.
Oh yeah, I can think of tons of these, but you probably wouldn't be allowed to do any of them. Disparate impact would nuke this conversational one for sure.
But honestly, give me any unmediated long-term data source, and we can probably build un-Goodhartable proxies that do much better than anything today.
Like all of your search engine searches for the last ten years, or your geolocation data, or your email or texting or commenting history over that time, any one of those would work fine.
Heck, genomic data is probably as strong as any of those - just a little pinprick and a quick blood test, and we'd probably know enough.
Back to my "shades of Gattaca" point, think anyone's gonna be happy about any of the above? The howls of outrage would shut it all down as soon as it was even proposed, before anything ever got off the ground.
The idea that people might have naturally differing levels of human capital or ability that are discernible via pretty much any data source is anathema to society overall, with very rare exceptions (education being one of them, because in theory, anyone can be "educated," even though THAT'S not true either, and of course the value of the "education" signal has declined precipitously exactly in line with access and credentialing expanding to greater proportions of the population, just as we'd expect).
Any time I try to think about how I would approach this problem, it's far outside the Overton Window pretty much immediately - which actually sucks for lots of poor / minority / disadvantaged populations who could genuinely be sorted to and perform in much more rewarding and interesting jobs!
But the fact that it wouldn't perfectly adhere to base population distributions of various "victimhood" checkboxes pretty much kills anything along these lines from the beginning. Our current Overton Window sucks.
I couldn't find a whole post about it anywhere, apparently it's a beta product he's testing for his OSventures.
But he also mentions it in a few other places (output from o3):
Skim the OSVerse transcript for Ep 265. That’s where he outlines his data‑ingestion strategy (books → casual blog posts → personal letters) and admits the letters were the “unlock” for style fidelity.
Pull his February 12, 2025 thread on ThreadReader. It’s the only place he shares multi‑paragraph model output so you can see how the agent writes.
Watch for future OSV engineering posts. In August 2024 he hinted the internal AI stack would become a public‑facing product; given OSV’s cadence it will probably surface this year.
I can't express strongly enough how much I wish we had something like the Civilization Conservation Corps today. We really need to let go of college as the North star for every child - the credentialing issue trickles all the way down to elementary school so we have second graders worried about getting Bs. Not everyone is meant to be an entrepreneur or an academic! But we've so devalued (literally) manufacturing, public service, farming, etc, it's going to be a tough sell. But like...we'll always need people to fix the roads and grow food and be social workers.
An underappreciated point about the Oxbridge system is that a student’s class depends mostly on end-of-year exams, and exam scripts are anonymized before being marked. Often the lecturer will not be the person doing the marking, and there are procedures in place to make sure that individual courses are not too easy or too hard (although those procedures are not perfect). So there is no sense in which different professors are competing with each other to give lower/higher grades, which removes much of the pressure for grade inflation.
I was horrified when I learned that in the US it’s normal for students to go to a TA or professor and beg for higher grades on individual pieces of work. That system seems to be open to so much abuse.
I really think they've got it figured out at Oxford. It's a great call out -- the fact that the exams more or less happen at the very end means the entire rest of the time can _only_ be used for education. In this way the higher education purpose of the university remains paramount. But I also think that a degree from Oxford, though prestigious, is probably less of a direct hiring credential than one from an Ivy League in the US. Most of the credentialist value of the Oxford degree is baked into just being accepted to the university in the first place; American companies don't know how to think through the rest of their educational structure.
(Less true outside of the US. In the UK, Europe, and India, if you mentioned you went to Oxford, people would immediately think of you differently and treat you with more respect. No one knows any of the Ivy leagues besides Harvard, but everyone knows Oxford.)
So. Gonna be a bit contrarian. I taught CS - in college - once upon a time. I taught a class - Formal Languages and Finite Automata - that the students widely thought was a waste of time. "Why do we need to learn this?" I heard constantly. As you say, arrogant little shits.
But they were also smart and aware and they would graduate and get jobs and then come back a year after graduation and drop by my office and tell me "finite state machines are everywhere!". I would just nod. That's why we taught them. They wouldn't ever see it or think it valuable if we hadn't taught them about them.
But you know, we are teaching people technical skills. Things that would make them better engineers or scientists.
Knowing about finite-state machines does not make you a better entrepreneur. Database-backed websites are really easy to build. You can do it without knowing about FSM. You can hire people to do it and know even less.
So everybody has their favorite story about a college dropout that made billions in the computer industry. I'm not sure, but I think that window has closed. Most startups go nowhere. Some get acquired and the rewards for founders are sufficient for the founders to try again. It's the bankers making the money. Maybe you have the next pet rock, I mean Facebook, and you will make billions. The odds aren't there, though.
However, and this is something that we sort of agree on, I learned and taught this stuff because I loved it. Because I thought it was fascinating in its own right. Because it was some fundamental knowledge of how the universe worked. Like math. And it would have been nice to have more students with that same attitude.
Appreciate the contrarian take!
Most of my feelings about the uselessness of my college courses come from my time at Google research rather than my time in the startup world. Part of this is because I went hard into deep learning, a subject that at the time was not given any importance in college classes that were explicitly about ML (we spent a month on Gaussian models and two days on neural networks). But part of it was because I quickly found that it was much easier to learn in a focused goal driven environment than in a classroom setting.
Here's an example. I took a class on linear algebra and forgot all of it because I was studying for a very abstract test. There was no intuition building, no world modeling. What made the linear algebra eventually stick many years later was working on real world problems that actually required some understanding of linear algebra. But that understanding wasn't abstract, it was extremely applied -- "this matmul represents a camera coordinate shift in 3d space", things like that. The thing is, a bunch of those applied uses of linear algebra led to a more abstract understanding that I could then apply in a bunch of other places. It was practical AND educational.
I also took a class that taught me about FSMs, and not only have I not really encountered a time where that classwork proved useful, I have completely forgotten all of that classwork _anyway_, so even if I did find such a use case I would have to start from scratch. Maybe I'm unique in this, but I suspect I'm not. Most people learn on the job. In many ways, building expertise through usage is really efficient, because you're only ever learning stuff that is immediately useful, to the level that it's immediately useful. If you need to use something over and over again -- like linear algebra for me -- you'll build the mental models you need through the usage. Classes by contrast always lag behind, so you'll often be learning stuff that is outdated by the time you hit the real world.
--
I do agree that education is valuable on its own accord. I wish that more classes focused on inspiring wonder and awe and genuine interest in a field. But I think that as long as education systems and credentialing systems are in the same place, you cannot have that. The corporation demands legibility.
One of the things I remember from my time at an Ivy was the commentary around Brown's CS major. People said that you could graduate with a CS major from Brown with a lot of flexibility in which classes you wanted to take. But this was framed as a _bad_ thing. A common refrain was "o ya recruiters don't really like hiring from Brown because they don't really know what those students have actually learned." The same tension plagues Oxbridge. That is a learning environment that is basically as free form and self directed as you can imagine. For self starters who are legitimately curious about the world, it's a beautiful system. But the average corporation just doesn't know what to do with the mishmash of "classes" the Oxbridge student takes. It doesn't fit into a neat box, so it's not derisked as much, so the value of the credential is less.
We are in complete concord with regard to the conflict between credentialing systems and educational systems.
I have been musing this morning on how state systems are ubiquitous, not just in computing, but in, for instance, quantum mechanics and biology. Neurons embody state. To be fair, I never tried to motivate it that way or make that connection when I was teaching it.
I submit that while you may not remember the stuff from your FSM class, but the work you did for it changed you in a way that is valuable. This is "classwork as fitness exercise". Football players do pushups and run through tires, even though during a game they do neither of these things.
One final note. Apparently you went to Brown? For all the complaining about how "employers don't like Brown graduates" you got a job at Google, and quite a good one, I think.
The employers who don't like Brown, I would guess, are offering jobs that are narrow, rigid and stultifying. Not something a Brown grad would like.
I went to Columbia 😂 maybe it was just cope to make us Columbia grads feel better than the Brown grads. There was a lot of that too.
I did get a job at Google, but it was not due to the *educational* value of my degree. There was definitely a class of person who made it through to the phone screens because of the *credential* value of the degree, but even that didn't apply to me -- I ended up getting through the phone screen because I built something cool at Hack Rutgers and one of the Google folks there took notice.
> I have been musing this morning on how state systems are ubiquitous, not just in computing, but in, for instance, quantum mechanics and biology. Neurons embody state. To be fair, I never tried to motivate it that way or make that connection when I was teaching it.
When I first introduce people to a new video game -- like Civ or Smash Bros -- I recommend that they pick a leader or character and get really good with that character first. They should play all their games with that same character, and only when they feel like they really grok how to play that character should they move around and explore other ways of playing the game.
Something that I've noticed about myself is that when I became really comfortable / good at one field, I started figuring out all sorts of analogies to how things in that field applied to other, unrelated fields. For me, that's ML -- I got really familiar with embedding spaces and manifolds and topology and linear algebra, and now I can't help but see all sorts of problems in *other* spaces through the same lens. For example, I see a lot of systems as optimization problems and use intuitions developed from how neural networks behave to peer into disparate fields like economics or politics or molecular biology.
I think this is how a lot of people learn things and become generalists -- they start by becoming an expert in one thing, and then use that one thing to get hand holds in adjacent fields, until those adjacent fields also become intuitive and then they go farther out.
So in your case, I could see that through FSMs. And fwiw I totally buy that a deep intuition about FSMs is really useful. I just think its a bit non-unique to FSMs. Being an expert in *anything* is approximately as useful as being an expert in FSMs.
> I submit that while you may not remember the stuff from your FSM class, but the work you did for it changed you in a way that is valuable. This is "classwork as fitness exercise". Football players do pushups and run through tires, even though during a game they do neither of these things.
This got me thinking a bit as I am typing out this response. I actually did end up doing a lot of state-machine-like things -- a lot of my research at Google was about graph neural networks, which of course...graphs. But I really am struggling to even remember what we discussed in my discrete math classes as they related to FSMs; certainly I can't remember a single moment when I was doing graph learning stuff and felt like I could draw a connection to my discrete math class... Unclear. You may be right but its not falsifiable without more data points
I like the "deep dive" as a model/motivator of learning.
One note about discrete math, which I also taught. It is a non-subject. There's no cohesive theme to it, it's just a bunch of little bits of lore that students need to know about for other classes in a CS major.
I also have a math degree and nobody thinks of it as a topic, or sub-branch of mathematics. But it does satisfy a specific need.
When I took it, it was mostly stuff I already knew from other sources. But it was required, so I sat through it.
Anyway, I think this is why you can't remember anything from it.
Yea that's more or less what I felt about discrete math -- was basically totally not cohesive at all 😂
Thank you for writing this.
I went to university and have mixed feelings about how useful it really was for my financial, mental, and physical well-being.
Alternatives to 4-year degrees need to be more seriously considered by teachers and parents.
Also, university is not the only place to learn. Books and life experience exist.
Love the article.
I agree much is signaling, and double agree that we collectively should be able to test out of a lot more things (I myself was so outraged in undergrad that they wanted me to waste 2 years on "gen eds" so pointless that attendance was part of the grade, that I took nothing but physics and math classes for two years, then CLEPPED out of ~36 credits worth of gen eds, and fortunately by that time had been invited to continue studying under / doing research by my PI).
A couple of thoughts:
I agree that school is probably 90% signaling. But as a founder and employer who's interviewed and hired lots of smart people over the years, the "Ivy" signal is super strong and very worth it for entry level employees. I'm not sure an IQ test can replace it, because IQ tests only test part of what makes somebody good at their job, and conscientious and discipline are both long-term traits that you can't assess in an interview. "Competent, conscientious, and good enough emotional regulation and short and long term planning" is the other important part of the educational signal.
Companies still need a way to identify the best potential workers, customers still need to identify the best service providers, and future workers themselves need some means of reference to understand how they genuinely compare to their competitors in ability and long-running accomplishment.
Even if we overturn Griggs, and allow IQ testing as part of the interview process, and even if we have a lot more industry-and-job-specific tests, it is not a good enough signal.
All FAA(N/M)Gs and finance already do IQ test equivalents, and have been doing for ages. Remember when everyone was treating MSFT or GOOG interview questions like brain teasers several years ago?
I agree that better and more wide ranging testing would help a lot of people, particularly genuinely talented minorities and middle class people who got stuck with crappy schools and so don't show well on paper.
But how do we test for conscientiousness and discipline? Many a wild hair can max on an IQ or job-specific test, but then doesn't deliver solid results over time, are lazy, are inconsistent, fail to see the big picture, fail to map their work back to business value, and more. That's such a prevalent factor, in fact, that Google basically gave up on the brain teaser style interviews! They were optimizing for the wrong thing, and it didn't correlate with actual success in their work life.
Think of this - even the smartest, best companies in the world, with the best modeling and design of experiments and analytics capablities, still bat about 50% in terms of choosing good long-term candidates via the hiring process. And that's WITH being able to skim only the crème de la crème of the entire world, including heavily oversampling Ivy grads.
There's a lot of inherent noise in the system, in other words - and imagine how much noisier it gets when you're down in the "35%" instead of the "top 0.5%" of the tail.
Now imagine that, but at the 50th or 85th percentile instead of 35th. Is there *any* real solution there?? No matter how well calibrated the tests are?
Second point - you bring up kids.
Like the typical Rationalist Stemlord, I hated school from the day it started clear through undergrad. A crushing waste of time, and a grinding Sisyphean ordeal that's more about babysitting and soul crushing than teaching or learning anything.
Obviously, I'm going to unschool / home school my kids, and try to go for an 'elite tutoring' dynamic per Hoel.
But I think founders have a great and underappreciated option here, too. I actually think of Ivy (or equivalents) as the *backup* option, rather than the primary option, because you can lead and teach by example, and any kid that has the capability of doing a startup or business themselves will be much better off doing that rather than spending 4 years in Uni.
Sure, some kids actually like school, and some kids might want to do research (I know I loved research, for all that I hated school). Uni and grad school is a great option for them. But for any with the talent and inclination to actually do things in the world, I think there's a lot of value in mentoring them into a place where they can start and grow a business instead.
Michael Strong had a decent article about this recently: https://michaelstrong.substack.com/p/advice-to-entrepreneurial-parents
As for the arrogance problem, I think it's quite different getting knowledge and funding and advice from your parents in an ongoing way from your childhood, versus being funded / ideologically recruited by a billionaire when you're 18.
I personally worry less about arrogance - I more or less feel that where unmerited, the arrogance will be knocked out of them by the slings and arrows of the world, and if they really are better / more successful on various metrics, I don't see much harm in knowing this. Like I'd be happy if one of my kids became a Thiel fellow or equivalent, because it puts them on the path of doing and having more of a positive impact on the world, and even if they're a*holes in personal life (which hopefully you steered away from in childhood, I agree this is bad), if you found and grow a business big enough to make you rich, I feel like you generally definitionally had a positive impact on the world for millions of people (with some fundamentally adversarial exceptions I've written about before, like the attention economy).
But I do think the whole thing would be sidestepped neatly by the "raising them with business intuition" approach above - then they're being shaped by you both morally and business knowledge wise throughout childhood, and will probably be better off than your 18 year old insufferables.
> But how do we test for conscientiousness and discipline? Many a wild hair can max on an IQ or job-specific test, but then doesn't deliver solid results over time, are lazy, are inconsistent, fail to see the big picture, fail to map their work back to business value, and more. That's such a prevalent factor, in fact, that Google basically gave up on the brain teaser style interviews! They were optimizing for the wrong thing, and it didn't correlate with actual success in their work life.
Strong agree that generally the Ivy signal is a strong signal, both as a founder and Xoogler. That said, I'm not sure though that I am that beat up about diminishing that signal. I think that if the market needs to figure out something else, it will -- especially since it is a *very* important signal with a lot of capital behind it. And I think that most of the Ivy school signal comes from selection effects, which in turn means you should be able to construct testing requirements for businesses that match those that universities give to highschool seniors (why DONT jobs like McKinsey ask their candidates for written essays? Legal liability? Something else?)
I suspect that some of the conscientiousness and discipline you are talking about comes from the network of people you build up when going to an ivy. And having that network is independently useful, especially in a startup setting where knowing smart people helps you bring in other smart people. Still, you could imagine that a free market with a bit more competition in it for credentialing may find ways to get around these problems.
> But for any with the talent and inclination to actually do things in the world, I think there's a lot of value in mentoring them into a place where they can start and grow a business instead.
I'm not convinced that starting a business is the default way to go for the average kid, or really the average person. I do think its important to teach kids how to be independent and give them the tools to think critically, but I think you have to have certain desires / opinions about where you want to be in the world to justify starting a company. I'm someone who thinks of my work as a core part of my identity and so am happy to dive into being a founder. But I have friends and family who just aren't like that -- who see their work as a 9-5 in order to pursue other, more compelling interests, whether that is writing or reading or sports or something else. And I don't think there's anything wrong with that either... tldr a well rounded education seems more important than one specifically tailored towards entrepreneurship. But this is obviously highly subjective and kids are pretty plastic.
> I more or less feel that where unmerited, the arrogance will be knocked out of them by the slings and arrows of the world
Yea, maybe. I just feel like I know a handful of folks who went out to the valley around 18 and have become extremely cynical as a result of it. Something about the experience results in maladjustment. It's not like these kids have bad relationships with their parents either; but it is possible that those parents don't have entrepreneurship experience themselves, and the kids are far from home (e.g. in SF vs in their hometown).
More generally, I think that unschooling and homeschooling works for individual cases but it will never scale up. The university system is in part so successful because it *mostly* works even when you pass hundreds of millions of people through it. My suggestions at the bottom of the essay are trying to get at systems that may work at that scale
> if the market needs to figure out something else, it will -- especially since it is a *very* important signal with a lot of capital behind it.
Yeah, but my point here was that it's probably impossible, the noise is much higher than the signal. Even the literally best companies in the world using the strongest possible signals bat 50%, and that's at the <0.5% percentile of base talent and selection on multiple correlated desirable metrics.
Like if you were just some regular company fishing at the 50th percentile, what hope do YOU have? Approximately zero? I think we'd just be back to nepotism and in-person referrals (which are the best channel for ANY business doing recruiting, IMO, I consider a referral from an employee you respect the strongest possible signal), instead of "selection at arms-length via the educational system," which I agree is slightly nepotistically captured at the top.
> I suspect that some of the conscientiousness and discipline you are talking about comes from the network of people you build up when going to an ivy.
Funnily enough I've been reading about conscientiousness interventions recently - I was skeptical that we can move it at all, and indeed most of the studies are p-hacked garbage, and posit absolutely ridiculous effects like a cohens d=0.5 from a 6-12 week once a week coaching intervention. Lol, as if.
My basic prior going in was "this is going to be like diet and exercise, which approximately nobody succeed at (80% failure to lose 10% and keep it off for a year, 98% failure for 20% over 5 years). I think it's basically around there, but there are actually heterogenous populations who respond and keep up the changes (and they indeed range from 5-20% of the pop, depending on the pop and intervention methods). First jobs are a big one for C buffs, as is military enlistment - school didn't show up in aggregate, but that's not surprising to me, given how lax most school's standards are.
Another mind-blower on this front - did you know Jim O'shaughnessy literally has an in-house AI that he's fed Cialdini's *Influence* and Nir Eyal's *Hooked* so that it optimizes the workflows and habits of his family with intermittent variable rewards to be healthier, etc? He's literally living in the "AI assistant" future we'll all be in in a year or two, already!
But on an Ivy environment's effect on C, I agree with you overall - I think the competitive forge of interacting with, learning from, and competing against other bright talents is one of those environments. And that the network is *impossibly* valuable. I've actually talked about how even in a post-scarcity future, I'd still be trying to get my kids into T10 schools (because then your social networks matter more than anything else! It's all you'll be doing with your time!).
> I'm not convinced that starting a business is the default way to go for the average kid, or really the average person.
Oh yeah, sorry - here I switched from "thinking of societal level ideas" to personal one-to-one communication without signaling a context shift. Like this is what I'm planning for my own kids, and I've recommended it to a few other friends.
I agree there's no way it works at scale, especially in an AGI future. But we're personally highly selected pops, and it's likely at least some of our kids will inherit our predispositions and talents on that front.
> Yeah, but my point here was that it's probably impossible, the noise is much higher than the signal. Even the literally best companies in the world using the strongest possible signals bat 50%, and that's at the <0.5% percentile of base talent and selection on multiple correlated desirable metrics.
It can't *actually* be impossible, right? The ivy leagues figured it out! I don't think my suggestion is "each individual company should institute their own credentialling system". My suggestion is more like "we should have dedicated institutions for credentialling that are completely different from those that do education".
I admit straight up that I have no idea how you could 'test' for conscientiousness, but also I've spent only 1 day thinking about it. I suspect an organization(s) thinking about this 24/7 could come up with something reasonable. Off the top of my head, a battery of personal essays seems like it would potentially help a lot in figuring out this signal.
Now...its also possible that we've naturally evolved towards the best possible signal already i.e. something that looks like an Ivy league school. I'm a bit skeptical that that's really the case, but also culture tends to be smarter than too-smart-for-their-own-good individuals. I'd be curious though if being at a good uni with a bunch of other smart people improves your conscientiousness, seems somewhat easy to study though potentially really hard to untangle from the effects of just, like, getting older and more mature. But I also knew a lot of assholes at my school, they were assholes when we met and assholes when we left *shrug*
Somewhat related: probably the best thing about most high end universities is that it gives a selection of poorer kids access to those same networks, which I feel like may do a lot more for lifting people out of poverty than the credential itself. But we have some kinds of things that replicate that experience. Even though I bash them, I think that VC incubator programs can be similar (though the main risk there is that its explicitly an extractive relationship, whereas the friends you make in college are peers).
> Another mind-blower on this front - did you know Jim O'shaughnessy literally has an in-house AI that he's fed Cialdini's *Influence* and Nir Eyal's *Hooked* so that it optimizes the workflows and habits of his family with intermittent variable rewards to be healthier, etc? He's literally living in the "AI assistant" future we'll all be in in a year or two, already!
Sick! I have to take a look at this
> Oh yeah, sorry - here I switched from "thinking of societal level ideas" to personal one-to-one communication without signaling a context shift.
makes sense. I think I personally am planning to have tutors for my kids up through middleschool and maybe past that. That's more or less what my parents did for me and I think it really helped immensely. I don't think I'm going so far as to homeschool though -- I know a bunch of homeschooled kids and I think the lack of socialization can hurt them.
> Now...its also possible that we've naturally evolved towards the best possible signal already i.e. something that looks like an Ivy league school.
Sure, and what does that process look like?
Looking at some of the other parents I know, it apparently looks like "the competition to get into Harvard starts 6 months before birth, when you need to get on the waiting list for the right exclusive pre-school to give your precious Jansen a leg up, because if you don't get in there, and if you don't bared-teeth-grind furiously and nonstop for the next 18 years, their chances of getting into Harvard are *ruined!*"
Like it's literally a lifelong battle, or at least a decade or 5+ year battle, where both kids and parents exert maximal effort for years in a zero-sum competition and come out the winners.
It seems like that's pretty well tuned to surface genuine conscientiousness AND ambition AND academic talent, etc.
Versus any test you can do can and will be Goodharted immediately. Like there are current Big 5 tests that test for C, just use those, right? Lol, how fast would it take for everyone to 'know' the right answers?
There's some obvious technologically mediated solutions. Everyone installs an app when they're teenagers that tracks what they eat and their exercise behavior and whatever. As a former athlete, I've def been biased towards hiring other former athletes, because I know what it takes on the C and discipline fronts to be competitive. Still only gives you a small buff, doesn't correlate with business value all that much.
Or when everyone has AI assistants, the assistants will DEFINITELY know how naturally conscientious their people are, and how much of an external scaffolding and buff their people need to execute at the level they're currently executing. But who knows, the AI assistants could lie, or be induced to lie, or something.
Certainly if we were given unmediated lifelong data, we could build much better predictive models that could do pretty well. But using those would be pretty fraught, legally, societally and morally. Shades of Gattaca, you know.
I'm definitely not saying "school systems are the best method." We know they're staggeringly wasteful in time and cost, and that they've moved from something like 5 students per administrator to 1:1 students per administrator, which is so ridiculous I can't even talk about it without ranting.
I'm just saying they currently do encode something like a "years deep, unfalsifiable signal" about C, at least at the T20 level, and that it would be fairly hard to surface that signal via other methods that aren't Goodhartable.
> There's some obvious technologically mediated solutions. Everyone installs an app when they're teenagers that tracks what they eat and their exercise behavior and whatever.
That's grim.
Maybe the answer is that we just need less focus on credentials overall. Like, how many jobs *actually* require more than a secondary degree or a high degree of conscientiousness? I'm sure it would be great if the average line cook had intelligence + conscientiousness for the restaurant owner, but it doesn't really benefit the rest of society.
> I'm just saying they currently do encode something like a "years deep, unfalsifiable signal" about C, at least at the T20 level, and that it would be fairly hard to surface that signal via other methods that aren't Goodhartable.
arguably even college has a bit of goodhardt-ness to it, given how people game things like recommendations and sports and testing and so on.
I'd love to see a company take a serious crack at this, though idk what exactly that would look like. One thing that could be interesting -- albeit dystopian -- is to try and get some fine tuned AI classifiers that are really good at figuring out someone's conscientiousness from conversation. Like, you have a conversational AI agent that someone talks to for a period of time, maybe even a long period of time, and then a different agent reads the entire convo and makes a judgement. Very ethically fraught though given how blackbox it would feel.
> Like, how many jobs *actually* require more than a secondary degree or a high degree of conscientiousness? I'm sure it would be great if the average line cook had intelligence + conscientiousness for the restaurant owner, but it doesn't really benefit the rest of society.
I mean, you say that from a consumer perspective, but from the restaurant owner perspective, it's basically the only thing that matters.
I've got a lot of real estate developers in my social circles - do you know how hard it is to find construction people who will A) show up on time reliably, and B) not be high or drunk / start fights / steal tools or materials / deal drugs on the job?
It's *extremely* rare to pass this seemingly minimal qualification threshold. This is why everyone employs immigrants, legal or otherwise, and even that is hit or miss and pretty much everyone has crew shortages all the time.
I've heard similar tales from people trying to hire for manufacturing or factories in the Rust Belt.
I think the floor of "net positive to employ from a pure cost benefit perspective" is probably ABOVE the 30-50th C percentile for most geographies / jobs. And of course, that floor is continually rising as selection effects draw all the high C, high intelligence people to major metros, and as jobs get more complex, and as AGI and robotics comes on the scene.
There's a giant self-sorting effect at work in top ten metros, you know - on intelligence and C and a lot of other metrics. The great majority of the country doesn't live in those top ten metros, though.
> I'd love to see a company take a serious crack at this, though idk what exactly that would look like. One thing that could be interesting -- albeit dystopian -- is to try and get some fine tuned AI classifiers that are really good at figuring out someone's conscientiousness from conversation.
Oh yeah, I can think of tons of these, but you probably wouldn't be allowed to do any of them. Disparate impact would nuke this conversational one for sure.
But honestly, give me any unmediated long-term data source, and we can probably build un-Goodhartable proxies that do much better than anything today.
Like all of your search engine searches for the last ten years, or your geolocation data, or your email or texting or commenting history over that time, any one of those would work fine.
Heck, genomic data is probably as strong as any of those - just a little pinprick and a quick blood test, and we'd probably know enough.
Back to my "shades of Gattaca" point, think anyone's gonna be happy about any of the above? The howls of outrage would shut it all down as soon as it was even proposed, before anything ever got off the ground.
The idea that people might have naturally differing levels of human capital or ability that are discernible via pretty much any data source is anathema to society overall, with very rare exceptions (education being one of them, because in theory, anyone can be "educated," even though THAT'S not true either, and of course the value of the "education" signal has declined precipitously exactly in line with access and credentialing expanding to greater proportions of the population, just as we'd expect).
Any time I try to think about how I would approach this problem, it's far outside the Overton Window pretty much immediately - which actually sucks for lots of poor / minority / disadvantaged populations who could genuinely be sorted to and perform in much more rewarding and interesting jobs!
But the fact that it wouldn't perfectly adhere to base population distributions of various "victimhood" checkboxes pretty much kills anything along these lines from the beginning. Our current Overton Window sucks.
On in-house AI, mentioned here:
https://substack.com/inbox/post/162590046
I couldn't find a whole post about it anywhere, apparently it's a beta product he's testing for his OSventures.
But he also mentions it in a few other places (output from o3):
Skim the OSVerse transcript for Ep 265. That’s where he outlines his data‑ingestion strategy (books → casual blog posts → personal letters) and admits the letters were the “unlock” for style fidelity.
Pull his February 12, 2025 thread on ThreadReader. It’s the only place he shares multi‑paragraph model output so you can see how the agent writes.
Watch for future OSV engineering posts. In August 2024 he hinted the internal AI stack would become a public‑facing product; given OSV’s cadence it will probably surface this year.