The Lesson to Unlearn

The Lesson to Unlearn. An essay by Paul Graham, from December 2019.

The most damaging lesson

What's the most harmful thing school teaches? Not anything from one particular class, Graham argues. It's how to get good grades.

In college, Graham met a philosophy grad student who cared only what he learned, never his grade. Nobody else Graham met ever said that.

Graham himself was a keen student. Yet he worked hardest before tests. Measuring the learning had crowded out the learning.

In theory, and in practice

In theory, a test just checks what you've learned in class. You shouldn't need to prepare for it, any more than for a blood test.

In practice, it's reversed. Studying means studying for a test. Good students just cram harder. And nobody pulls an all-nighter in week two.

Call a diligent student grade-driven, and most people hear no contrast at all. That's how completely grades and learning have fused.

Tests that miss

Is confusing grades with learning really so bad? Yes, says Graham. He only saw how bad decades later, running Y Combinator.

Crammed facts don't stick, of course. But the deeper problem: most tests measure something other than what they claim to.

If tests truly measured learning, grades would catch up with it, just a bit late. But nearly every test can be hacked, and good students know it.

Hacking a history final

Say your medieval history final is two days away. Surely the smart move is to read the best books on the period?

Seasoned students shake their heads. As Graham puts it in a footnote: “Learning is the naive algorithm for getting good grades.”

The expert's way: study the lecture notes, and only crisp facts that could become questions. Three causes of the Schism of 1378? Learn exactly three.

Were they really the causes? Doesn't matter. Old exams narrow things further, since professors reuse questions. Some classes even want their politics echoed.

Grades or learning

A good grade and real knowledge are so far apart that you must pick one. Can you blame students for picking grades? Everyone judges them by grades.

Graham enjoyed some papers he wrote in college. But did he ever write another, just for fun? Never: other classes had deadlines. He chose grades.

Anyone who cares about grades must play along, or be overtaken. So students compete to widen the gap between their grades and what they learn.

Why tests leak

Why are tests so easy to hack? Ask any programmer. Software built with no thought for security leaks like a colander.

Any test set by an authority is hackable by default. But that's hardly the teachers' fault. Their job is teaching, not building leak-proof tests.

The real culprit: overloaded grades. As pure feedback, like a coach's advice, nobody would hack them. But past a certain age, every lesson comes with a verdict.

The admissions test

College exams are actually the least hackable. The worst is admissions. If it weighed minds like scientists weigh mass, we'd just tell teens: learn a lot.

Instead, teens get rote classes, box-ticking activities, standardized tests, and an essay with a hidden target. Whole industries exist just to hack it.

Why so hackable? Officers want applicants who seem admirable in some broad sense, and judge that by feel. In effect, they admit who they like.

A test of fitting someone's taste is bound to be hackable. With stakes this high, it's gamed relentlessly, bending teenage lives into an artificial shape.

Founders looking for the trick

But wasted time isn't the worst of it. School teaches you that winning means finding the trick to hack the test. Graham first saw this in young founders.

At Y Combinator they'd ask: how do we raise money? What's the trick? His answer: be a good investment. Fool investors, and you fool yourselves.

What makes a good investment? Growth, in revenue or at least users. How do you get users? Not a splashy Tuesday launch. Make the product great.

Obvious, surely. Yet founders reacted like physicists meeting relativity: amazed, but suspicious. Then: can you introduce us to someone famous?

Trained to hack the test

Founders could take years to absorb lessons this simple. Graham kept wondering why they made everything so complicated, until he saw the question wasn't rhetorical.

They'd been trained to. Recent graduates had never faced a real test. And fundraising looked like one: it came last, had numbers, and bigger seemed better.

Plenty of the world does reward hacking tests. Startups mostly don't. You win by winning users, and users only care whether it does what they need.

Graham's own training

Graham had been trained too. He'd avoided big companies without knowing why: there, you win by hacking bad tests. Startups drew him: their tests can't be hacked.

He'd unlearned it slowly, without knowing. Could graduates banish this demon just by naming it? Worth a try: it draws its power from going unnoticed.

Nobody planned this. Mix learning with grades, competition, and a naive faith in tests, and you get it. It solved two of his oldest puzzles at once.

Spotting a bad test

So most ambitious graduates have something to unlearn. And a new question for any work: how much do you win here by hacking bad tests?

A quick rule: tests no authority imposes can't be hacked. A football match tests only who wins. That's why a pundit can say the better team won.

Tests set by authorities are proxies: an exam stands in for what you learned. Proxies must be sealed, and rarely are. So bad tests are, roughly, the ones authorities impose.

Good work wins, more and more

Many assume bad tests are the price of getting rich. Not anymore, says Graham. That was true mid-century, when oligopolies ran the economy.

When he was a kid, engineers built cool things and executives made money. Now building cool things can make you rich.

As work comes loose from authority, hacking bad tests matters less. Startups show it most. And writers can now skip publishers and go straight to readers.

So Graham is optimistic. As people refuse to win by hacking, that work will starve of talent, and schools will follow. Society, too, has this lesson to unlearn.

In short

School's hidden lesson: win by hacking the test.

Tests leak, because grades judge you, not just guide you.

Startups, and more and more work, reward good work.

Ask of any work: do you win by hacking bad tests?

The original says it better than this. Read it: paulgraham.com/lesson.html