Education

We optimize for grades in school, promotion cycles, citation counts, but I keep wondering, who decided these metrics? Or more importantly, should I be optimizing for metrics I don’t believe in?

Most people never ask this question. They inherit their metrics from parents, teachers, institutions. If you go through any form of public system in your life, you’ve likely inherited the metrics rather than deriving your own. It’s easier that way. Your parents wouldn’t want you to be contrarian in a rational sense. But I think increasingly, it’s needed.

I’d like to model the world in terms of power law, where around 80% of people inherit metrics and optimize for execution while the remaining 20% invent their own metrics and develop judgment. The distinction is simple: execution means you know how to do the task, judgment means you know what task is worth doing. Most education systems train execution, almost none train judgment (by education systems I mean all institutional forms including liberal arts colleges, research universities, technical schools—despite different structures, they all face the same credentialing pressures and optimize for the same measurable outcomes).

The origin of where human efforts are headed is shaped by educational institutions. They educate and direct the masses toward efforts where necessary, and money then coordinates where the execution actually goes. Hence I want to spend this essay talking about this critical threshold because it distills everything civilizational. Any metrics measured or hacked here have downstream effects on what sort of metrics get created and hacked later.

I see the root cause as the principal-agent problem distilled across generations. When the creator delegates, taste gets lost. The person downstream doesn’t have the domain expertise and optimizes for proxy metrics instead.

This maps to Goodhart’s Law, which states that when a measure becomes a target, it ceases to be a good measure. The creator knew what mattered because they understood the territory, not just the map (Newton understood physics through first principles, his frameworks worked because of this), but downstream we’re left with lossy metrics that approximate the original intent. Physics concepts become test problems, hospital quality metrics get gamed and patients die, Wells Fargo optimized for account openings and committed fraud.

I believe this is mathematically inevitable. Skalse et al. proved that for the set of all stochastic policies, two reward functions can only be unhackable if one of them is constant, which means any sufficiently optimized agent will eventually game its metrics. It’s not a bug, it’s a certainty, and this happens everywhere. But it’s most dangerous in education because education shapes the people who become future leaders who shape society. The principal-agent problem here has civilizational downstream effects.

Universities have always been factories serving institutional needs, not individual curiosity. Medieval universities trained clergy and lawyers for church and state institutions, not for pure inquiry. The German research model introduced specialized research but still served state interests. American land-grant colleges were explicitly designed to train workers for agriculture and mechanics, the factory model made obvious. Post-WWII mass education scaled the factory through the GI Bill, and Clark Kerr literally described universities as “knowledge factories” serving the economy.

The function never changed, only the marketing did. Medieval universities advertised professional training, modern universities advertise critical thinking and innovation, but both deliver the same thing: a credentialing pipeline. It’s the same principal-agent problem built into the structure, and it compounds across generations as the metric becomes lossier with each passover.

But there’s a distinction I want to make between what institutions do and what education itself should be. Given that the inherent purpose of education is to diffuse knowledge, I have two beliefs about how this should work.

First, I believe the point of knowledge is action. You don’t know if knowledge is proper knowledge unless you’ve tested it through physics experiments, mathematical proofs, free market validation. But with the principal-agent problem stacking across generations, most people never act upon their knowledge to test it. It’s always been distilled down through layers. Some try re-deriving knowledge through self-study, which has merit, but if a class forces you to do it without curiosity, it won’t work.

Second, I believe the purpose of education has always been personalization, not standardization. Education should be tailored to your curiosity because curiosity drives learning, living, solving problems. You need foundations to know what problems are worth solving, but by the time problems reach you through institutional layers, they’ve been so distilled you should question their legitimacy from first principles (this is where the principal-agent problem becomes dangerous).

Either way it’s still a factory, and educational institutions optimize for standardization because they function as factories outputting labor. This is rational because they need funding through tuition, donations, government support. They need prestige measured by rankings and outcomes. They need measurable metrics like graduation rates, employment rates, median salaries. Check any college scorecard and these metrics reveal the actual incentive: workforce pipeline, not judgment development.

Students learn indirectly and rationalize afterward, saying “they taught me this and that,” but learning judgment indirectly through execution training is not definite. You’re still in the factory. I see the mechanism working like this: institutions claim they develop critical thinking and judgment and wisdom, but they actually deliver pre-selected problems and standardized curricula and execution training. Students conflate learning judgment with getting good grades on problems others chose, and graduates believe they have judgment because they executed well on institutional metrics.

The pathway seems straightforward according to signaling theory: good grades lead to high test scores lead to college admission lead to credentials lead to job opportunities. I doubt the original creators intended this, but the principal-agent problem distilled it into something else entirely.

You were recruited to Stanford because you took risks. You were different, you deviated from norms and started something unusual, pursued unconventional projects. But the very privilege that enabled those risks makes you risk-averse once admitted because you now have too much to lose. The Stanford credential itself becomes valuable, and you optimize to preserve it rather than take real risks.

We were admitted for being different, but once here we’re incentivized to be the same. Everyone funnels into consulting, finance, tech (the safe prestigious paths). Not because Stanford explicitly tells you to, but because recruiting cycles make it the path of least resistance, peer pressure surrounds you (everyone around you is doing it), and the structure rewards conventional outcomes for rankings and prestige.

I think Stanford wants both: the prestige of admitting risk-takers for marketing and donations, and the stability of conventional career outcomes for those same rankings and employment statistics. This creates fake leaders, people with Stanford credentials who consensus-seek rather than truth-seek. They optimized the game and aren’t reasoning from first principles.

Demis Hassabis described this during a ten-hour chess match against the Danish champion, where he realized: “Maybe this isn’t - there’s this whole room full of amazingly bright people, and they’re using their minds to basically compete with each other… maybe they should all be using their minds to solve cancer or something.”

We’re mimetic creatures, and it’s easier to inherit beliefs than examine them. Surrounded by people optimizing for GPA then internship then return offer, you absorb that frame via social osmosis. The sequence should be curiosity leading to judgment leading to taste, but most people never reach curiosity because they inherit their domains entirely.

Perhaps you enter wanting to understand the universe and leave wanting a good GPA for consulting recruiting. I call this Education as The Great Filter.

I see that the danger isn’t that institutions standardize for the masses (that’s necessary for civilization), but rather filtering out the ambitious not because they lack ability but because they couldn’t find their tribe. Judgment requires choosing your own problems and facing real consequences from markets or physics, but institutions give you problems others chose with artificial consequences like grades instead of market feedback.

It’s hard to walk alone. Hunter-gatherers evolved staying in groups through mimetic behavior and conformity, and walking the other way meant getting eaten by a lion. But in this world of rationalization and consensus, we need truth seekers. That’s the leader, and rightfully the leader reasons from first principles.

Increasingly I believe our system creates fake leaders with YC clout and credentials who consensus-seek rather than truth-seek. Most people are workers, instruments for the few who truly make decisions, and this is how civilization functions. Educational institutions standardize the masses for labor, which is correct and necessary, but the false advertising attracts people who could be prime movers and pipelines them into standardized execution.

Not everyone should be projecting their vision onto the world (only the truly sound reasoners should). That distinction is what institutions feed into the world while training for something else entirely.

If you’re a potential prime mover at Stanford, you have one question: do you want to develop judgment or optimize for execution on someone else’s metrics?

If you’re still questioning rather than optimizing, if you’re truly truth-seeking, let’s find each other before we get filtered out.

Prime movers.