id
int64 0
217
| title
stringlengths 5
52
| date
stringlengths 8
14
⌀ | text
stringlengths 404
75.5k
|
|---|---|---|---|
0 | The Age of the Essay | September 2004 | Remember the essays you had to write in high school? Topic sentence, introductory paragraph, supporting paragraphs, conclusion. The conclusion being, say, that Ahab in _Moby Dick_ was a Christ-like figure.
Oy. So I'm going to try to give the other side of the story: what an essay really is, and how you write one. Or at least, how I write one.
**Mods**
The most obvious difference between real essays and the things one has to write in school is that real essays are not exclusively about English literature. Certainly schools should teach students how to write. But due to a series of historical accidents the teaching of writing has gotten mixed together with the study of literature. And so all over the country students are writing not about how a baseball team with a small budget might compete with the Yankees, or the role of color in fashion, or what constitutes a good dessert, but about symbolism in Dickens.
With the result that writing is made to seem boring and pointless. Who cares about symbolism in Dickens? Dickens himself would be more interested in an essay about color or baseball.
How did things get this way? To answer that we have to go back almost a thousand years. Around 1100, Europe at last began to catch its breath after centuries of chaos, and once they had the luxury of curiosity they rediscovered what we call "the classics." The effect was rather as if we were visited by beings from another solar system. These earlier civilizations were so much more sophisticated that for the next several centuries the main work of European scholars, in almost every field, was to assimilate what they knew.
During this period the study of ancient texts acquired great prestige. It seemed the essence of what scholars did. As European scholarship gained momentum it became less and less important; by 1350 someone who wanted to learn about science could find better teachers than Aristotle in his own era. \[1\] But schools change slower than scholarship. In the 19th century the study of ancient texts was still the backbone of the curriculum.
The time was then ripe for the question: if the study of ancient texts is a valid field for scholarship, why not modern texts? The answer, of course, is that the original raison d'etre of classical scholarship was a kind of intellectual archaeology that does not need to be done in the case of contemporary authors. But for obvious reasons no one wanted to give that answer. The archaeological work being mostly done, it implied that those studying the classics were, if not wasting their time, at least working on problems of minor importance.
And so began the study of modern literature. There was a good deal of resistance at first. The first courses in English literature seem to have been offered by the newer colleges, particularly American ones. Dartmouth, the University of Vermont, Amherst, and University College, London taught English literature in the 1820s. But Harvard didn't have a professor of English literature until 1876, and Oxford not till 1885. (Oxford had a chair of Chinese before it had one of English.) \[2\]
What tipped the scales, at least in the US, seems to have been the idea that professors should do research as well as teach. This idea (along with the PhD, the department, and indeed the whole concept of the modern university) was imported from Germany in the late 19th century. Beginning at Johns Hopkins in 1876, the new model spread rapidly.
Writing was one of the casualties. Colleges had long taught English composition. But how do you do research on composition? The professors who taught math could be required to do original math, the professors who taught history could be required to write scholarly articles about history, but what about the professors who taught rhetoric or composition? What should they do research on? The closest thing seemed to be English literature. \[3\]
And so in the late 19th century the teaching of writing was inherited by English professors. This had two drawbacks: (a) an expert on literature need not himself be a good writer, any more than an art historian has to be a good painter, and (b) the subject of writing now tends to be literature, since that's what the professor is interested in.
High schools imitate universities. The seeds of our miserable high school experiences were sown in 1892, when the National Education Association "formally recommended that literature and composition be unified in the high school course." \[4\] The 'riting component of the 3 Rs then morphed into English, with the bizarre consequence that high school students now had to write about English literature-- to write, without even realizing it, imitations of whatever English professors had been publishing in their journals a few decades before.
It's no wonder if this seems to the student a pointless exercise, because we're now three steps removed from real work: the students are imitating English professors, who are imitating classical scholars, who are merely the inheritors of a tradition growing out of what was, 700 years ago, fascinating and urgently needed work.
**No Defense**
The other big difference between a real essay and the things they make you write in school is that a real essay doesn't take a position and then defend it. That principle, like the idea that we ought to be writing about literature, turns out to be another intellectual hangover of long forgotten origins.
It's often mistakenly believed that medieval universities were mostly seminaries. In fact they were more law schools. And at least in our tradition lawyers are advocates, trained to take either side of an argument and make as good a case for it as they can. Whether cause or effect, this spirit pervaded early universities. The study of rhetoric, the art of arguing persuasively, was a third of the undergraduate curriculum. \[5\] And after the lecture the most common form of discussion was the disputation. This is at least nominally preserved in our present-day thesis defense: most people treat the words thesis and dissertation as interchangeable, but originally, at least, a thesis was a position one took and the dissertation was the argument by which one defended it.
Defending a position may be a necessary evil in a legal dispute, but it's not the best way to get at the truth, as I think lawyers would be the first to admit. It's not just that you miss subtleties this way. The real problem is that you can't change the question.
And yet this principle is built into the very structure of the things they teach you to write in high school. The topic sentence is your thesis, chosen in advance, the supporting paragraphs the blows you strike in the conflict, and the conclusion-- uh, what is the conclusion? I was never sure about that in high school. It seemed as if we were just supposed to restate what we said in the first paragraph, but in different enough words that no one could tell. Why bother? But when you understand the origins of this sort of "essay," you can see where the conclusion comes from. It's the concluding remarks to the jury.
Good writing should be convincing, certainly, but it should be convincing because you got the right answers, not because you did a good job of arguing. When I give a draft of an essay to friends, there are two things I want to know: which parts bore them, and which seem unconvincing. The boring bits can usually be fixed by cutting. But I don't try to fix the unconvincing bits by arguing more cleverly. I need to talk the matter over.
At the very least I must have explained something badly. In that case, in the course of the conversation I'll be forced to come up a with a clearer explanation, which I can just incorporate in the essay. More often than not I have to change what I was saying as well. But the aim is never to be convincing per se. As the reader gets smarter, convincing and true become identical, so if I can convince smart readers I must be near the truth.
The sort of writing that attempts to persuade may be a valid (or at least inevitable) form, but it's historically inaccurate to call it an essay. An essay is something else.
**Trying**
To understand what a real essay is, we have to reach back into history again, though this time not so far. To Michel de Montaigne, who in 1580 published a book of what he called "essais." He was doing something quite different from what lawyers do, and the difference is embodied in the name. _Essayer_ is the French verb meaning "to try" and an _essai_ is an attempt. An essay is something you write to try to figure something out.
Figure out what? You don't know yet. And so you can't begin with a thesis, because you don't have one, and may never have one. An essay doesn't begin with a statement, but with a question. In a real essay, you don't take a position and defend it. You notice a door that's ajar, and you open it and walk in to see what's inside.
If all you want to do is figure things out, why do you need to write anything, though? Why not just sit and think? Well, there precisely is Montaigne's great discovery. Expressing ideas helps to form them. Indeed, helps is far too weak a word. Most of what ends up in my essays I only thought of when I sat down to write them. That's why I write them.
In the things you write in school you are, in theory, merely explaining yourself to the reader. In a real essay you're writing for yourself. You're thinking out loud.
But not quite. Just as inviting people over forces you to clean up your apartment, writing something that other people will read forces you to think well. So it does matter to have an audience. The things I've written just for myself are no good. They tend to peter out. When I run into difficulties, I find I conclude with a few vague questions and then drift off to get a cup of tea.
Many published essays peter out in the same way. Particularly the sort written by the staff writers of newsmagazines. Outside writers tend to supply editorials of the defend-a-position variety, which make a beeline toward a rousing (and foreordained) conclusion. But the staff writers feel obliged to write something "balanced." Since they're writing for a popular magazine, they start with the most radioactively controversial questions, from which-- because they're writing for a popular magazine-- they then proceed to recoil in terror. Abortion, for or against? This group says one thing. That group says another. One thing is certain: the question is a complex one. (But don't get mad at us. We didn't draw any conclusions.)
**The River**
Questions aren't enough. An essay has to come up with answers. They don't always, of course. Sometimes you start with a promising question and get nowhere. But those you don't publish. Those are like experiments that get inconclusive results. An essay you publish ought to tell the reader something he didn't already know.
But _what_ you tell him doesn't matter, so long as it's interesting. I'm sometimes accused of meandering. In defend-a-position writing that would be a flaw. There you're not concerned with truth. You already know where you're going, and you want to go straight there, blustering through obstacles, and hand-waving your way across swampy ground. But that's not what you're trying to do in an essay. An essay is supposed to be a search for truth. It would be suspicious if it didn't meander.
The Meander (aka Menderes) is a river in Turkey. As you might expect, it winds all over the place. But it doesn't do this out of frivolity. The path it has discovered is the most economical route to the sea. \[6\]
The river's algorithm is simple. At each step, flow down. For the essayist this translates to: flow interesting. Of all the places to go next, choose the most interesting. One can't have quite as little foresight as a river. I always know generally what I want to write about. But not the specific conclusions I want to reach; from paragraph to paragraph I let the ideas take their course.
This doesn't always work. Sometimes, like a river, one runs up against a wall. Then I do the same thing the river does: backtrack. At one point in this essay I found that after following a certain thread I ran out of ideas. I had to go back seven paragraphs and start over in another direction.
Fundamentally an essay is a train of thought-- but a cleaned-up train of thought, as dialogue is cleaned-up conversation. Real thought, like real conversation, is full of false starts. It would be exhausting to read. You need to cut and fill to emphasize the central thread, like an illustrator inking over a pencil drawing. But don't change so much that you lose the spontaneity of the original.
Err on the side of the river. An essay is not a reference work. It's not something you read looking for a specific answer, and feel cheated if you don't find it. I'd much rather read an essay that went off in an unexpected but interesting direction than one that plodded dutifully along a prescribed course.
**Surprise**
So what's interesting? For me, interesting means surprise. Interfaces, as Geoffrey James has said, should follow the principle of least astonishment. A button that looks like it will make a machine stop should make it stop, not speed up. Essays should do the opposite. Essays should aim for maximum surprise.
I was afraid of flying for a long time and could only travel vicariously. When friends came back from faraway places, it wasn't just out of politeness that I asked what they saw. I really wanted to know. And I found the best way to get information out of them was to ask what surprised them. How was the place different from what they expected? This is an extremely useful question. You can ask it of the most unobservant people, and it will extract information they didn't even know they were recording.
Surprises are things that you not only didn't know, but that contradict things you thought you knew. And so they're the most valuable sort of fact you can get. They're like a food that's not merely healthy, but counteracts the unhealthy effects of things you've already eaten.
How do you find surprises? Well, therein lies half the work of essay writing. (The other half is expressing yourself well.) The trick is to use yourself as a proxy for the reader. You should only write about things you've thought about a lot. And anything you come across that surprises you, who've thought about the topic a lot, will probably surprise most readers.
For example, in a recent [essay](gh.html) I pointed out that because you can only judge computer programmers by working with them, no one knows who the best programmers are overall. I didn't realize this when I began that essay, and even now I find it kind of weird. That's what you're looking for.
So if you want to write essays, you need two ingredients: a few topics you've thought about a lot, and some ability to ferret out the unexpected.
What should you think about? My guess is that it doesn't matter-- that anything can be interesting if you get deeply enough into it. One possible exception might be things that have deliberately had all the variation sucked out of them, like working in fast food. In retrospect, was there anything interesting about working at Baskin-Robbins? Well, it was interesting how important color was to the customers. Kids a certain age would point into the case and say that they wanted yellow. Did they want French Vanilla or Lemon? They would just look at you blankly. They wanted yellow. And then there was the mystery of why the perennial favorite Pralines 'n' Cream was so appealing. (I think now it was the salt.) And the difference in the way fathers and mothers bought ice cream for their kids: the fathers like benevolent kings bestowing largesse, the mothers harried, giving in to pressure. So, yes, there does seem to be some material even in fast food.
I didn't notice those things at the time, though. At sixteen I was about as observant as a lump of rock. I can see more now in the fragments of memory I preserve of that age than I could see at the time from having it all happening live, right in front of me.
**Observation**
So the ability to ferret out the unexpected must not merely be an inborn one. It must be something you can learn. How do you learn it?
To some extent it's like learning history. When you first read history, it's just a whirl of names and dates. Nothing seems to stick. But the more you learn, the more hooks you have for new facts to stick onto-- which means you accumulate knowledge at an exponential rate. Once you remember that Normans conquered England in 1066, it will catch your attention when you hear that other Normans conquered southern Italy at about the same time. Which will make you wonder about Normandy, and take note when a third book mentions that Normans were not, like most of what is now called France, tribes that flowed in as the Roman empire collapsed, but Vikings (norman = north man) who arrived four centuries later in 911. Which makes it easier to remember that Dublin was also established by Vikings in the 840s. Etc, etc squared.
Collecting surprises is a similar process. The more anomalies you've seen, the more easily you'll notice new ones. Which means, oddly enough, that as you grow older, life should become more and more surprising. When I was a kid, I used to think adults had it all figured out. I had it backwards. Kids are the ones who have it all figured out. They're just mistaken.
When it comes to surprises, the rich get richer. But (as with wealth) there may be habits of mind that will help the process along. It's good to have a habit of asking questions, especially questions beginning with Why. But not in the random way that three year olds ask why. There are an infinite number of questions. How do you find the fruitful ones?
I find it especially useful to ask why about things that seem wrong. For example, why should there be a connection between humor and misfortune? Why do we find it funny when a character, even one we like, slips on a banana peel? There's a whole essay's worth of surprises there for sure.
If you want to notice things that seem wrong, you'll find a degree of skepticism helpful. I take it as an axiom that we're only achieving 1% of what we could. This helps counteract the rule that gets beaten into our heads as children: that things are the way they are because that is how things have to be. For example, everyone I've talked to while writing this essay felt the same about English classes-- that the whole process seemed pointless. But none of us had the balls at the time to hypothesize that it was, in fact, all a mistake. We all thought there was just something we weren't getting.
I have a hunch you want to pay attention not just to things that seem wrong, but things that seem wrong in a humorous way. I'm always pleased when I see someone laugh as they read a draft of an essay. But why should I be? I'm aiming for good ideas. Why should good ideas be funny? The connection may be surprise. Surprises make us laugh, and surprises are what one wants to deliver.
I write down things that surprise me in notebooks. I never actually get around to reading them and using what I've written, but I do tend to reproduce the same thoughts later. So the main value of notebooks may be what writing things down leaves in your head.
People trying to be cool will find themselves at a disadvantage when collecting surprises. To be surprised is to be mistaken. And the essence of cool, as any fourteen year old could tell you, is _nil admirari._ When you're mistaken, don't dwell on it; just act like nothing's wrong and maybe no one will notice.
One of the keys to coolness is to avoid situations where inexperience may make you look foolish. If you want to find surprises you should do the opposite. Study lots of different things, because some of the most interesting surprises are unexpected connections between different fields. For example, jam, bacon, pickles, and cheese, which are among the most pleasing of foods, were all originally intended as methods of preservation. And so were books and paintings.
Whatever you study, include history-- but social and economic history, not political history. History seems to me so important that it's misleading to treat it as a mere field of study. Another way to describe it is _all the data we have so far._
Among other things, studying history gives one confidence that there are good ideas waiting to be discovered right under our noses. Swords evolved during the Bronze Age out of daggers, which (like their flint predecessors) had a hilt separate from the blade. Because swords are longer the hilts kept breaking off. But it took five hundred years before someone thought of casting hilt and blade as one piece.
**Disobedience**
Above all, make a habit of paying attention to things you're not supposed to, either because they're "[inappropriate](say.html)," or not important, or not what you're supposed to be working on. If you're curious about something, trust your instincts. Follow the threads that attract your attention. If there's something you're really interested in, you'll find they have an uncanny way of leading back to it anyway, just as the conversation of people who are especially proud of something always tends to lead back to it.
For example, I've always been fascinated by comb-overs, especially the extreme sort that make a man look as if he's wearing a beret made of his own hair. Surely this is a lowly sort of thing to be interested in-- the sort of superficial quizzing best left to teenage girls. And yet there is something underneath. The key question, I realized, is how does the comber-over not see how odd he looks? And the answer is that he got to look that way _incrementally._ What began as combing his hair a little carefully over a thin patch has gradually, over 20 years, grown into a monstrosity. Gradualness is very powerful. And that power can be used for constructive purposes too: just as you can trick yourself into looking like a freak, you can trick yourself into creating something so grand that you would never have dared to _plan_ such a thing. Indeed, this is just how most good software gets created. You start by writing a stripped-down kernel (how hard can it be?) and gradually it grows into a complete operating system. Hence the next leap: could you do the same thing in painting, or in a novel?
See what you can extract from a frivolous question? If there's one piece of advice I would give about writing essays, it would be: don't do as you're told. Don't believe what you're supposed to. Don't write the essay readers expect; one learns nothing from what one expects. And don't write the way they taught you to in school.
The most important sort of disobedience is to write essays at all. Fortunately, this sort of disobedience shows signs of becoming [rampant](http://www.ojr.org/ojr/glaser/1056050270.php). It used to be that only a tiny number of officially approved writers were allowed to write essays. Magazines published few of them, and judged them less by what they said than who wrote them; a magazine might publish a story by an unknown writer if it was good enough, but if they published an essay on x it had to be by someone who was at least forty and whose job title had x in it. Which is a problem, because there are a lot of things insiders can't say precisely because they're insiders.
The Internet is changing that. Anyone can publish an essay on the Web, and it gets judged, as any writing should, by what it says, not who wrote it. Who are you to write about x? You are whatever you wrote.
Popular magazines made the period between the spread of literacy and the arrival of TV the golden age of the short story. The Web may well make this the golden age of the essay. And that's certainly not something I realized when I started writing this.
**Notes**
\[1\] I'm thinking of Oresme (c. 1323-82). But it's hard to pick a date, because there was a sudden drop-off in scholarship just as Europeans finished assimilating classical science. The cause may have been the plague of 1347; the trend in scientific progress matches the population curve.
\[2\] Parker, William R. "Where Do College English Departments Come From?" _College English_ 28 (1966-67), pp. 339-351. Reprinted in Gray, Donald J. (ed). _The Department of English at Indiana University Bloomington 1868-1970._ Indiana University Publications.
Daniels, Robert V. _The University of Vermont: The First Two Hundred Years._ University of Vermont, 1991.
Mueller, Friedrich M. Letter to the _Pall Mall Gazette._ 1886/87. Reprinted in Bacon, Alan (ed). _The Nineteenth-Century History of English Studies._ Ashgate, 1998.
\[3\] I'm compressing the story a bit. At first literature took a back seat to philology, which (a) seemed more serious and (b) was popular in Germany, where many of the leading scholars of that generation had been trained.
In some cases the writing teachers were transformed _in situ_ into English professors. Francis James Child, who had been Boylston Professor of Rhetoric at Harvard since 1851, became in 1876 the university's first professor of English.
\[4\] Parker, _op. cit._, p. 25.
\[5\] The undergraduate curriculum or _trivium_ (whence "trivial") consisted of Latin grammar, rhetoric, and logic. Candidates for masters' degrees went on to study the _quadrivium_ of arithmetic, geometry, music, and astronomy. Together these were the seven liberal arts.
The study of rhetoric was inherited directly from Rome, where it was considered the most important subject. It would not be far from the truth to say that education in the classical world meant training landowners' sons to speak well enough to defend their interests in political and legal disputes.
\[6\] Trevor Blackwell points out that this isn't strictly true, because the outside edges of curves erode faster.
**Thanks** to Ken Anderson, Trevor Blackwell, Sarah Harlin, Jessica Livingston, Jackie McDonough, and Robert Morris for reading drafts of this.
If you liked this, you may also like [**_Hackers & Painters_**](hackpaint.html). |
1 | A Plan for Spam | August 2002 | _(This article describes the spam-filtering techniques used in the spamproof web-based mail reader we built to exercise [Arc](arc.html). An improved algorithm is described in [Better Bayesian Filtering](better.html).)_
I think it's possible to stop spam, and that content-based filters are the way to do it. The Achilles heel of the spammers is their message. They can circumvent any other barrier you set up. They have so far, at least. But they have to deliver their message, whatever it is. If we can write software that recognizes their messages, there is no way they can get around that.
\_ \_ \_
To the recipient, spam is easily recognizable. If you hired someone to read your mail and discard the spam, they would have little trouble doing it. How much do we have to do, short of AI, to automate this process?
I think we will be able to solve the problem with fairly simple algorithms. In fact, I've found that you can filter present-day spam acceptably well using nothing more than a Bayesian combination of the spam probabilities of individual words. Using a slightly tweaked (as described below) Bayesian filter, we now miss less than 5 per 1000 spams, with 0 false positives.
The statistical approach is not usually the first one people try when they write spam filters. Most hackers' first instinct is to try to write software that recognizes individual properties of spam. You look at spams and you think, the gall of these guys to try sending me mail that begins "Dear Friend" or has a subject line that's all uppercase and ends in eight exclamation points. I can filter out that stuff with about one line of code.
And so you do, and in the beginning it works. A few simple rules will take a big bite out of your incoming spam. Merely looking for the word "click" will catch 79.7% of the emails in my spam corpus, with only 1.2% false positives.
I spent about six months writing software that looked for individual spam features before I tried the statistical approach. What I found was that recognizing that last few percent of spams got very hard, and that as I made the filters stricter I got more false positives.
False positives are innocent emails that get mistakenly identified as spams. For most users, missing legitimate email is an order of magnitude worse than receiving spam, so a filter that yields false positives is like an acne cure that carries a risk of death to the patient.
The more spam a user gets, the less likely he'll be to notice one innocent mail sitting in his spam folder. And strangely enough, the better your spam filters get, the more dangerous false positives become, because when the filters are really good, users will be more likely to ignore everything they catch.
I don't know why I avoided trying the statistical approach for so long. I think it was because I got addicted to trying to identify spam features myself, as if I were playing some kind of competitive game with the spammers. (Nonhackers don't often realize this, but most hackers are very competitive.) When I did try statistical analysis, I found immediately that it was much cleverer than I had been. It discovered, of course, that terms like "virtumundo" and "teens" were good indicators of spam. But it also discovered that "per" and "FL" and "ff0000" are good indicators of spam. In fact, "ff0000" (html for bright red) turns out to be as good an indicator of spam as any pornographic term.
\_ \_ \_
Here's a sketch of how I do statistical filtering. I start with one corpus of spam and one of nonspam mail. At the moment each one has about 4000 messages in it. I scan the entire text, including headers and embedded html and javascript, of each message in each corpus. I currently consider alphanumeric characters, dashes, apostrophes, and dollar signs to be part of tokens, and everything else to be a token separator. (There is probably room for improvement here.) I ignore tokens that are all digits, and I also ignore html comments, not even considering them as token separators.
I count the number of times each token (ignoring case, currently) occurs in each corpus. At this stage I end up with two large hash tables, one for each corpus, mapping tokens to number of occurrences.
Next I create a third hash table, this time mapping each token to the probability that an email containing it is a spam, which I calculate as follows \[1\]: (let ((g (\* 2 (or (gethash word good) 0))) (b (or (gethash word bad) 0))) (unless (< (+ g b) 5) (max .01 (min .99 (float (/ (min 1 (/ b nbad)) (+ (min 1 (/ g ngood)) (min 1 (/ b nbad))))))))) where word is the token whose probability we're calculating, good and bad are the hash tables I created in the first step, and ngood and nbad are the number of nonspam and spam messages respectively.
I explained this as code to show a couple of important details. I want to bias the probabilities slightly to avoid false positives, and by trial and error I've found that a good way to do it is to double all the numbers in good. This helps to distinguish between words that occasionally do occur in legitimate email and words that almost never do. I only consider words that occur more than five times in total (actually, because of the doubling, occurring three times in nonspam mail would be enough). And then there is the question of what probability to assign to words that occur in one corpus but not the other. Again by trial and error I chose .01 and .99. There may be room for tuning here, but as the corpus grows such tuning will happen automatically anyway.
The especially observant will notice that while I consider each corpus to be a single long stream of text for purposes of counting occurrences, I use the number of emails in each, rather than their combined length, as the divisor in calculating spam probabilities. This adds another slight bias to protect against false positives.
When new mail arrives, it is scanned into tokens, and the most interesting fifteen tokens, where interesting is measured by how far their spam probability is from a neutral .5, are used to calculate the probability that the mail is spam. If probs is a list of the fifteen individual probabilities, you calculate the [combined](naivebayes.html) probability thus: (let ((prod (apply #'\* probs))) (/ prod (+ prod (apply #'\* (mapcar #'(lambda (x) (- 1 x)) probs))))) One question that arises in practice is what probability to assign to a word you've never seen, i.e. one that doesn't occur in the hash table of word probabilities. I've found, again by trial and error, that .4 is a good number to use. If you've never seen a word before, it is probably fairly innocent; spam words tend to be all too familiar.
There are examples of this algorithm being applied to actual emails in an appendix at the end.
I treat mail as spam if the algorithm above gives it a probability of more than .9 of being spam. But in practice it would not matter much where I put this threshold, because few probabilities end up in the middle of the range.
\_ \_ \_
One great advantage of the statistical approach is that you don't have to read so many spams. Over the past six months, I've read literally thousands of spams, and it is really kind of demoralizing. Norbert Wiener said if you compete with slaves you become a slave, and there is something similarly degrading about competing with spammers. To recognize individual spam features you have to try to get into the mind of the spammer, and frankly I want to spend as little time inside the minds of spammers as possible.
But the real advantage of the Bayesian approach, of course, is that you know what you're measuring. Feature-recognizing filters like SpamAssassin assign a spam "score" to email. The Bayesian approach assigns an actual probability. The problem with a "score" is that no one knows what it means. The user doesn't know what it means, but worse still, neither does the developer of the filter. How many _points_ should an email get for having the word "sex" in it? A probability can of course be mistaken, but there is little ambiguity about what it means, or how evidence should be combined to calculate it. Based on my corpus, "sex" indicates a .97 probability of the containing email being a spam, whereas "sexy" indicates .99 probability. And Bayes' Rule, equally unambiguous, says that an email containing both words would, in the (unlikely) absence of any other evidence, have a 99.97% chance of being a spam.
Because it is measuring probabilities, the Bayesian approach considers all the evidence in the email, both good and bad. Words that occur disproportionately _rarely_ in spam (like "though" or "tonight" or "apparently") contribute as much to decreasing the probability as bad words like "unsubscribe" and "opt-in" do to increasing it. So an otherwise innocent email that happens to include the word "sex" is not going to get tagged as spam.
Ideally, of course, the probabilities should be calculated individually for each user. I get a lot of email containing the word "Lisp", and (so far) no spam that does. So a word like that is effectively a kind of password for sending mail to me. In my earlier spam-filtering software, the user could set up a list of such words and mail containing them would automatically get past the filters. On my list I put words like "Lisp" and also my zipcode, so that (otherwise rather spammy-sounding) receipts from online orders would get through. I thought I was being very clever, but I found that the Bayesian filter did the same thing for me, and moreover discovered of a lot of words I hadn't thought of.
When I said at the start that our filters let through less than 5 spams per 1000 with 0 false positives, I'm talking about filtering my mail based on a corpus of my mail. But these numbers are not misleading, because that is the approach I'm advocating: filter each user's mail based on the spam and nonspam mail he receives. Essentially, each user should have two delete buttons, ordinary delete and delete-as-spam. Anything deleted as spam goes into the spam corpus, and everything else goes into the nonspam corpus.
You could start users with a seed filter, but ultimately each user should have his own per-word probabilities based on the actual mail he receives. This (a) makes the filters more effective, (b) lets each user decide their own precise definition of spam, and (c) perhaps best of all makes it hard for spammers to tune mails to get through the filters. If a lot of the brain of the filter is in the individual databases, then merely tuning spams to get through the seed filters won't guarantee anything about how well they'll get through individual users' varying and much more trained filters.
Content-based spam filtering is often combined with a whitelist, a list of senders whose mail can be accepted with no filtering. One easy way to build such a whitelist is to keep a list of every address the user has ever sent mail to. If a mail reader has a delete-as-spam button then you could also add the from address of every email the user has deleted as ordinary trash.
I'm an advocate of whitelists, but more as a way to save computation than as a way to improve filtering. I used to think that whitelists would make filtering easier, because you'd only have to filter email from people you'd never heard from, and someone sending you mail for the first time is constrained by convention in what they can say to you. Someone you already know might send you an email talking about sex, but someone sending you mail for the first time would not be likely to. The problem is, people can have more than one email address, so a new from-address doesn't guarantee that the sender is writing to you for the first time. It is not unusual for an old friend (especially if he is a hacker) to suddenly send you an email with a new from-address, so you can't risk false positives by filtering mail from unknown addresses especially stringently.
In a sense, though, my filters do themselves embody a kind of whitelist (and blacklist) because they are based on entire messages, including the headers. So to that extent they "know" the email addresses of trusted senders and even the routes by which mail gets from them to me. And they know the same about spam, including the server names, mailer versions, and protocols.
\_ \_ \_
If I thought that I could keep up current rates of spam filtering, I would consider this problem solved. But it doesn't mean much to be able to filter out most present-day spam, because spam evolves. Indeed, most [antispam techniques](falsepositives.html) so far have been like pesticides that do nothing more than create a new, resistant strain of bugs.
I'm more hopeful about Bayesian filters, because they evolve with the spam. So as spammers start using "c0ck" instead of "cock" to evade simple-minded spam filters based on individual words, Bayesian filters automatically notice. Indeed, "c0ck" is far more damning evidence than "cock", and Bayesian filters know precisely how much more.
Still, anyone who proposes a plan for spam filtering has to be able to answer the question: if the spammers knew exactly what you were doing, how well could they get past you? For example, I think that if checksum-based spam filtering becomes a serious obstacle, the spammers will just switch to mad-lib techniques for generating message bodies.
To beat Bayesian filters, it would not be enough for spammers to make their emails unique or to stop using individual naughty words. They'd have to make their mails indistinguishable from your ordinary mail. And this I think would severely constrain them. Spam is mostly sales pitches, so unless your regular mail is all sales pitches, spams will inevitably have a different character. And the spammers would also, of course, have to change (and keep changing) their whole infrastructure, because otherwise the headers would look as bad to the Bayesian filters as ever, no matter what they did to the message body. I don't know enough about the infrastructure that spammers use to know how hard it would be to make the headers look innocent, but my guess is that it would be even harder than making the message look innocent.
Assuming they could solve the problem of the headers, the spam of the future will probably look something like this: Hey there. Thought you should check out the following: http://www.27meg.com/foo because that is about as much sales pitch as content-based filtering will leave the spammer room to make. (Indeed, it will be hard even to get this past filters, because if everything else in the email is neutral, the spam probability will hinge on the url, and it will take some effort to make that look neutral.)
Spammers range from businesses running so-called opt-in lists who don't even try to conceal their identities, to guys who hijack mail servers to send out spams promoting porn sites. If we use filtering to whittle their options down to mails like the one above, that should pretty much put the spammers on the "legitimate" end of the spectrum out of business; they feel obliged by various state laws to include boilerplate about why their spam is not spam, and how to cancel your "subscription," and that kind of text is easy to recognize.
(I used to think it was naive to believe that stricter laws would decrease spam. Now I think that while stricter laws may not decrease the amount of spam that spammers _send,_ they can certainly help filters to decrease the amount of spam that recipients actually see.)
All along the spectrum, if you restrict the sales pitches spammers can make, you will inevitably tend to put them out of business. That word _business_ is an important one to remember. The spammers are businessmen. They send spam because it works. It works because although the response rate is abominably low (at best 15 per million, vs 3000 per million for a catalog mailing), the cost, to them, is practically nothing. The cost is enormous for the recipients, about 5 man-weeks for each million recipients who spend a second to delete the spam, but the spammer doesn't have to pay that.
Sending spam does cost the spammer something, though. \[2\] So the lower we can get the response rate-- whether by filtering, or by using filters to force spammers to dilute their pitches-- the fewer businesses will find it worth their while to send spam.
The reason the spammers use the kinds of [sales pitches](http://www.milliondollaremails.com) that they do is to increase response rates. This is possibly even more disgusting than getting inside the mind of a spammer, but let's take a quick look inside the mind of someone who _responds_ to a spam. This person is either astonishingly credulous or deeply in denial about their sexual interests. In either case, repulsive or idiotic as the spam seems to us, it is exciting to them. The spammers wouldn't say these things if they didn't sound exciting. And "thought you should check out the following" is just not going to have nearly the pull with the spam recipient as the kinds of things that spammers say now. Result: if it can't contain exciting sales pitches, spam becomes less effective as a marketing vehicle, and fewer businesses want to use it.
That is the big win in the end. I started writing spam filtering software because I didn't want have to look at the stuff anymore. But if we get good enough at filtering out spam, it will stop working, and the spammers will actually stop sending it.
\_ \_ \_
Of all the approaches to fighting spam, from software to laws, I believe Bayesian filtering will be the single most effective. But I also think that the more different kinds of antispam efforts we undertake, the better, because any measure that constrains spammers will tend to make filtering easier. And even within the world of content-based filtering, I think it will be a good thing if there are many different kinds of software being used simultaneously. The more different filters there are, the harder it will be for spammers to tune spams to get through them.
**Appendix: Examples of Filtering**
[Here](https://sep.yimg.com/ty/cdn/paulgraham/spam1.txt?t=1595850613&) is an example of a spam that arrived while I was writing this article. The fifteen most interesting words in this spam are: qvp0045 indira mx-05 intimail $7500 freeyankeedom cdo bluefoxmedia jpg unsecured platinum 3d0 qves 7c5 7c266675 The words are a mix of stuff from the headers and from the message body, which is typical of spam. Also typical of spam is that every one of these words has a spam probability, in my database, of .99. In fact there are more than fifteen words with probabilities of .99, and these are just the first fifteen seen.
Unfortunately that makes this email a boring example of the use of Bayes' Rule. To see an interesting variety of probabilities we have to look at [this](https://sep.yimg.com/ty/cdn/paulgraham/spam2.txt?t=1595850613&) actually quite atypical spam.
The fifteen most interesting words in this spam, with their probabilities, are: madam 0.99 promotion 0.99 republic 0.99 shortest 0.047225013 mandatory 0.047225013 standardization 0.07347802 sorry 0.08221981 supported 0.09019077 people's 0.09019077 enter 0.9075001 quality 0.8921298 organization 0.12454646 investment 0.8568143 very 0.14758544 valuable 0.82347786 This time the evidence is a mix of good and bad. A word like "shortest" is almost as much evidence for innocence as a word like "madam" or "promotion" is for guilt. But still the case for guilt is stronger. If you combine these numbers according to Bayes' Rule, the resulting probability is .9027.
"Madam" is obviously from spams beginning "Dear Sir or Madam." They're not very common, but the word "madam" _never_ occurs in my legitimate email, and it's all about the ratio.
"Republic" scores high because it often shows up in Nigerian scam emails, and also occurs once or twice in spams referring to Korea and South Africa. You might say that it's an accident that it thus helps identify this spam. But I've found when examining spam probabilities that there are a lot of these accidents, and they have an uncanny tendency to push things in the right direction rather than the wrong one. In this case, it is not entirely a coincidence that the word "Republic" occurs in Nigerian scam emails and this spam. There is a whole class of dubious business propositions involving less developed countries, and these in turn are more likely to have names that specify explicitly (because they aren't) that they are republics.\[3\]
On the other hand, "enter" is a genuine miss. It occurs mostly in unsubscribe instructions, but here is used in a completely innocent way. Fortunately the statistical approach is fairly robust, and can tolerate quite a lot of misses before the results start to be thrown off.
For comparison, [here](https://sep.yimg.com/ty/cdn/paulgraham/hostexspam.txt?t=1595850613&) is an example of that rare bird, a spam that gets through the filters. Why? Because by sheer chance it happens to be loaded with words that occur in my actual email: perl 0.01 python 0.01 tcl 0.01 scripting 0.01 morris 0.01 graham 0.01491078 guarantee 0.9762507 cgi 0.9734398 paul 0.027040077 quite 0.030676773 pop3 0.042199217 various 0.06080265 prices 0.9359873 managed 0.06451222 difficult 0.071706355 There are a couple pieces of good news here. First, this mail probably wouldn't get through the filters of someone who didn't happen to specialize in programming languages and have a good friend called Morris. For the average user, all the top five words here would be neutral and would not contribute to the spam probability.
Second, I think filtering based on word pairs (see below) might well catch this one: "cost effective", "setup fee", "money back" -- pretty incriminating stuff. And of course if they continued to spam me (or a network I was part of), "Hostex" itself would be recognized as a spam term.
Finally, [here](https://sep.yimg.com/ty/cdn/paulgraham/legit.txt?t=1595850613&) is an innocent email. Its fifteen most interesting words are as follows: continuation 0.01 describe 0.01 continuations 0.01 example 0.033600237 programming 0.05214485 i'm 0.055427782 examples 0.07972858 color 0.9189189 localhost 0.09883721 hi 0.116539136 california 0.84421706 same 0.15981844 spot 0.1654587 us-ascii 0.16804294 what 0.19212411 Most of the words here indicate the mail is an innocent one. There are two bad smelling words, "color" (spammers love colored fonts) and "California" (which occurs in testimonials and also in menus in forms), but they are not enough to outweigh obviously innocent words like "continuation" and "example".
It's interesting that "describe" rates as so thoroughly innocent. It hasn't occurred in a single one of my 4000 spams. The data turns out to be full of such surprises. One of the things you learn when you analyze spam texts is how narrow a subset of the language spammers operate in. It's that fact, together with the equally characteristic vocabulary of any individual user's mail, that makes Bayesian filtering a good bet.
**Appendix: More Ideas**
One idea that I haven't tried yet is to filter based on word pairs, or even triples, rather than individual words. This should yield a much sharper estimate of the probability. For example, in my current database, the word "offers" has a probability of .96. If you based the probabilities on word pairs, you'd end up with "special offers" and "valuable offers" having probabilities of .99 and, say, "approach offers" (as in "this approach offers") having a probability of .1 or less.
The reason I haven't done this is that filtering based on individual words already works so well. But it does mean that there is room to tighten the filters if spam gets harder to detect. (Curiously, a filter based on word pairs would be in effect a Markov-chaining text generator running in reverse.)
Specific spam features (e.g. not seeing the recipient's address in the to: field) do of course have value in recognizing spam. They can be considered in this algorithm by treating them as virtual words. I'll probably do this in future versions, at least for a handful of the most egregious spam indicators. Feature-recognizing spam filters are right in many details; what they lack is an overall discipline for combining evidence.
Recognizing nonspam features may be more important than recognizing spam features. False positives are such a worry that they demand extraordinary measures. I will probably in future versions add a second level of testing designed specifically to avoid false positives. If a mail triggers this second level of filters it will be accepted even if its spam probability is above the threshold.
I don't expect this second level of filtering to be Bayesian. It will inevitably be not only ad hoc, but based on guesses, because the number of false positives will not tend to be large enough to notice patterns. (It is just as well, anyway, if a backup system doesn't rely on the same technology as the primary system.)
Another thing I may try in the future is to focus extra attention on specific parts of the email. For example, about 95% of current spam includes the url of a site they want you to visit. (The remaining 5% want you to call a phone number, reply by email or to a US mail address, or in a few cases to buy a certain stock.) The url is in such cases practically enough by itself to determine whether the email is spam.
Domain names differ from the rest of the text in a (non-German) email in that they often consist of several words stuck together. Though computationally expensive in the general case, it might be worth trying to decompose them. If a filter has never seen the token "xxxporn" before it will have an individual spam probability of .4, whereas "xxx" and "porn" individually have probabilities (in my corpus) of .9889 and .99 respectively, and a combined probability of .9998.
I expect decomposing domain names to become more important as spammers are gradually forced to stop using incriminating words in the text of their messages. (A url with an ip address is of course an extremely incriminating sign, except in the mail of a few sysadmins.)
It might be a good idea to have a cooperatively maintained list of urls promoted by spammers. We'd need a trust metric of the type studied by Raph Levien to prevent malicious or incompetent submissions, but if we had such a thing it would provide a boost to any filtering software. It would also be a convenient basis for boycotts.
Another way to test dubious urls would be to send out a crawler to look at the site before the user looked at the email mentioning it. You could use a Bayesian filter to rate the site just as you would an email, and whatever was found on the site could be included in calculating the probability of the email being a spam. A url that led to a redirect would of course be especially suspicious.
One cooperative project that I think really would be a good idea would be to accumulate a giant corpus of spam. A large, clean corpus is the key to making Bayesian filtering work well. Bayesian filters could actually use the corpus as input. But such a corpus would be useful for other kinds of filters too, because it could be used to test them.
Creating such a corpus poses some technical problems. We'd need trust metrics to prevent malicious or incompetent submissions, of course. We'd also need ways of erasing personal information (not just to-addresses and ccs, but also e.g. the arguments to unsubscribe urls, which often encode the to-address) from mails in the corpus. If anyone wants to take on this project, it would be a good thing for the world.
**Appendix: Defining Spam**
I think there is a rough consensus on what spam is, but it would be useful to have an explicit definition. We'll need to do this if we want to establish a central corpus of spam, or even to compare spam filtering rates meaningfully.
To start with, spam is not unsolicited commercial email. If someone in my neighborhood heard that I was looking for an old Raleigh three-speed in good condition, and sent me an email offering to sell me one, I'd be delighted, and yet this email would be both commercial and unsolicited. The defining feature of spam (in fact, its _raison d'etre_) is not that it is unsolicited, but that it is automated.
It is merely incidental, too, that spam is usually commercial. If someone started sending mass email to support some political cause, for example, it would be just as much spam as email promoting a porn site.
I propose we define spam as **unsolicited automated email**. This definition thus includes some email that many legal definitions of spam don't. Legal definitions of spam, influenced presumably by lobbyists, tend to exclude mail sent by companies that have an "existing relationship" with the recipient. But buying something from a company, for example, does not imply that you have solicited ongoing email from them. If I order something from an online store, and they then send me a stream of spam, it's still spam.
Companies sending spam often give you a way to "unsubscribe," or ask you to go to their site and change your "account preferences" if you want to stop getting spam. This is not enough to stop the mail from being spam. Not opting out is not the same as opting in. Unless the recipient explicitly checked a clearly labelled box (whose default was no) asking to receive the email, then it is spam.
In some business relationships, you do implicitly solicit certain kinds of mail. When you order online, I think you implicitly solicit a receipt, and notification when the order ships. I don't mind when Verisign sends me mail warning that a domain name is about to expire (at least, if they are the [actual registrar](http://siliconvalley.internet.com/news/article.php/1441651) for it). But when Verisign sends me email offering a FREE Guide to Building My E-Commerce Web Site, that's spam.
**Notes:**
\[1\] The examples in this article are translated into Common Lisp for, believe it or not, greater accessibility. The application described here is one that we wrote in order to test a new Lisp dialect called [Arc](arc.html) that is not yet released.
\[2\] Currently the lowest rate seems to be about $200 to send a million spams. That's very cheap, 1/50th of a cent per spam. But filtering out 95% of spam, for example, would increase the spammers' cost to reach a given audience by a factor of 20. Few can have margins big enough to absorb that.
\[3\] As a rule of thumb, the more qualifiers there are before the name of a country, the more corrupt the rulers. A country called The Socialist People's Democratic Republic of X is probably the last place in the world you'd want to live.
**Thanks** to Sarah Harlin for reading drafts of this; Daniel Giffin (who is also writing the production Arc interpreter) for several good ideas about filtering and for creating our mail infrastructure; Robert Morris, Trevor Blackwell and Erann Gat for many discussions about spam; Raph Levien for advice about trust metrics; and Chip Coldwell and Sam Steingold for advice about statistics.
You'll find this essay and 14 others in [**_Hackers & Painters_**](http://www.amazon.com/gp/product/0596006624).
**More Info:**
[Plan for Spam FAQ](spamfaq.html)
[Better Bayesian Filtering](http://paulgraham.com/better.html)
[Filters that Fight Back](ffb.html)
[Will Filters Kill Spam?](wfks.html)
[Probability](naivebayes.html)
[Spam is Different](spamdiff.html)
[Filters vs. Blacklists](falsepositives.html)
[Trust Metrics](http://www.levien.com/free/tmetric-HOWTO.html)
[Filtering Research](bayeslinks.html)
[Microsoft Patent](msftpatent.html)
[Slashdot Article](http://developers.slashdot.org/article.pl?sid=02/08/16/1428238&mode=thread&tid=156)
[The Wrong Way](http://office.microsoft.com/Assistance/9798/newfilters.aspx)
[LWN: Filter Comparison](http://lwn.net/Articles/9460/)
[CRM114 gets 99.87%](wsy.html) |
2 | The Trouble with the Segway | July 2009 | The Segway hasn't delivered on its initial promise, to put it mildly. There are several reasons why, but one is that people don't want to be seen riding them. Someone riding a Segway looks like a dork.
My friend Trevor Blackwell built [his own Segway](http://tlb.org/#scooter), which we called the Segwell. He also built a one-wheeled version, [the Eunicycle](http://tlb.org/#eunicycle), which looks exactly like a regular unicycle till you realize the rider isn't pedaling. He has ridden them both to downtown Mountain View to get coffee. When he rides the Eunicycle, people smile at him. But when he rides the Segwell, they shout abuse from their cars: "Too lazy to walk, ya fuckin homo?"
Why do Segways provoke this reaction? The reason you look like a dork riding a Segway is that you look _smug_. You don't seem to be working hard enough.
Someone riding a motorcycle isn't working any harder. But because he's sitting astride it, he seems to be making an effort. When you're riding a Segway you're just standing there. And someone who's being whisked along while seeming to do no work — someone in a sedan chair, for example — can't help but look smug.
Try this thought experiment and it becomes clear: imagine something that worked like the Segway, but that you rode with one foot in front of the other, like a skateboard. That wouldn't seem nearly as uncool.
So there may be a way to capture more of the market Segway hoped to reach: make a version that doesn't look so easy for the rider. It would also be helpful if the styling was in the tradition of skateboards or bicycles rather than medical devices.
Curiously enough, what got Segway into this problem was that the company was itself a kind of Segway. It was too easy for them; they were too successful raising money. If they'd had to grow the company gradually, by iterating through several versions they sold to real users, they'd have learned pretty quickly that people looked stupid riding them. Instead they had enough to work in secret. They had focus groups aplenty, I'm sure, but they didn't have the people yelling insults out of cars. So they never realized they were zooming confidently down a blind alley. |
3 | After Credentials | December 2008 | A few months ago I read a _New York Times_ article on South Korean cram schools that said
> Admission to the right university can make or break an ambitious young South Korean.
A parent added:
> "In our country, college entrance exams determine 70 to 80 percent of a person's future."
It was striking how old fashioned this sounded. And yet when I was in high school it wouldn't have seemed too far off as a description of the US. Which means things must have been changing here.
The course of people's lives in the US now seems to be determined less by credentials and more by performance than it was 25 years ago. Where you go to college still matters, but not like it used to.
What happened?
\_\_\_\_\_
Judging people by their academic credentials was in its time an advance. The practice seems to have begun in China, where starting in 587 candidates for the imperial civil service had to take an exam on classical literature. \[[1](#f1n)\] It was also a test of wealth, because the knowledge it tested was so specialized that passing required years of expensive training. But though wealth was a necessary condition for passing, it was not a sufficient one. By the standards of the rest of the world in 587, the Chinese system was very enlightened. Europeans didn't introduce formal civil service exams till the nineteenth century, and even then they seem to have been influenced by the Chinese example.
Before credentials, government positions were obtained mainly by family influence, if not outright bribery. It was a great step forward to judge people by their performance on a test. But by no means a perfect solution. When you judge people that way, you tend to get cram schools—which they did in Ming China and nineteenth century England just as much as in present day South Korea.
What cram schools are, in effect, is leaks in a seal. The use of credentials was an attempt to seal off the direct transmission of power between generations, and cram schools represent that power finding holes in the seal. Cram schools turn wealth in one generation into credentials in the next.
It's hard to beat this phenomenon, because the schools adjust to suit whatever the tests measure. When the tests are narrow and predictable, you get cram schools on the classic model, like those that prepared candidates for Sandhurst (the British West Point) or the classes American students take now to improve their SAT scores. But as the tests get broader, the schools do too. Preparing a candidate for the Chinese imperial civil service exams took years, as prep school does today. But the raison d'etre of all these institutions has been the same: to beat the system. \[[2](#f2n)\]
\_\_\_\_\_
History suggests that, all other things being equal, a society prospers in proportion to its ability to prevent parents from influencing their children's success directly. It's a fine thing for parents to help their children indirectly—for example, by helping them to become smarter or more disciplined, which then makes them more successful. The problem comes when parents use direct methods: when they are able to use their own wealth or power as a substitute for their children's qualities.
Parents will tend to do this when they can. Parents will die for their kids, so it's not surprising to find they'll also push their scruples to the limits for them. Especially if other parents are doing it.
Sealing off this force has a double advantage. Not only does a society get "the best man for the job," but parents' ambitions are diverted from direct methods to indirect ones—to actually trying to raise their kids well.
But we should expect it to be very hard to contain parents' efforts to obtain an unfair advantage for their kids. We're dealing with one of the most powerful forces in human nature. We shouldn't expect naive solutions to work, any more than we'd expect naive solutions for keeping heroin out of a prison to work.
\_\_\_\_\_
The obvious way to solve the problem is to make credentials better. If the tests a society uses are currently hackable, we can study the way people beat them and try to plug the holes. You can use the cram schools to show you where most of the holes are. They also tell you when you're succeeding in fixing them: when cram schools become less popular.
A more general solution would be to push for increased transparency, especially at critical social bottlenecks like college admissions. In the US this process still shows many outward signs of corruption. For example, legacy admissions. The official story is that legacy status doesn't carry much weight, because all it does is break ties: applicants are bucketed by ability, and legacy status is only used to decide between the applicants in the bucket that straddles the cutoff. But what this means is that a university can make legacy status have as much or as little weight as they want, by adjusting the size of the bucket that straddles the cutoff.
By gradually chipping away at the abuse of credentials, you could probably make them more airtight. But what a long fight it would be. Especially when the institutions administering the tests don't really want them to be airtight.
\_\_\_\_\_
Fortunately there's a better way to prevent the direct transmission of power between generations. Instead of trying to make credentials harder to hack, we can also make them matter less.
Let's think about what credentials are for. What they are, functionally, is a way of predicting performance. If you could measure actual performance, you wouldn't need them.
So why did they even evolve? Why haven't we just been measuring actual performance? Think about where credentialism first appeared: in selecting candidates for large organizations. Individual performance is hard to measure in large organizations, and the harder performance is to measure, the more important it is to predict it. If an organization could immediately and cheaply measure the performance of recruits, they wouldn't need to examine their credentials. They could take everyone and keep just the good ones.
Large organizations can't do this. But a bunch of small organizations in a market can come close. A market takes every organization and keeps just the good ones. As organizations get smaller, this approaches taking every person and keeping just the good ones. So all other things being equal, a society consisting of more, smaller organizations will care less about credentials.
\_\_\_\_\_
That's what's been happening in the US. That's why those quotes from Korea sound so old fashioned. They're talking about an economy like America's a few decades ago, dominated by a few big companies. The route for the ambitious in that sort of environment is to join one and climb to the top. Credentials matter a lot then. In the culture of a large organization, an elite pedigree becomes a self-fulfilling prophecy.
This doesn't work in small companies. Even if your colleagues were impressed by your credentials, they'd soon be parted from you if your performance didn't match, because the company would go out of business and the people would be dispersed.
In a world of small companies, performance is all anyone cares about. People hiring for a startup don't care whether you've even graduated from college, let alone which one. All they care about is what you can do. Which is in fact all that should matter, even in a large organization. The reason credentials have such prestige is that for so long the large organizations in a society tended to be the most powerful. But in the US at least they don't have the monopoly on power they once did, precisely because they can't measure (and thus reward) individual performance. Why spend twenty years climbing the corporate ladder when you can get rewarded directly by the market?
I realize I see a more exaggerated version of the change than most other people. As a partner at an early stage venture funding firm, I'm like a jumpmaster shoving people out of the old world of credentials and into the new one of performance. I'm an agent of the change I'm seeing. But I don't think I'm imagining it. It was not so easy 25 years ago for an ambitious person to choose to be judged directly by the market. You had to go through bosses, and they were influenced by where you'd been to college.
\_\_\_\_\_
What made it possible for small organizations to succeed in America? I'm still not entirely sure. Startups are certainly a large part of it. Small organizations can develop new ideas faster than large ones, and new ideas are increasingly valuable.
But I don't think startups account for all the shift from credentials to measurement. My friend Julian Weber told me that when he went to work for a New York law firm in the 1950s they paid associates far less than firms do today. Law firms then made no pretense of paying people according to the value of the work they'd done. Pay was based on seniority. The younger employees were paying their dues. They'd be rewarded later.
The same principle prevailed at industrial companies. When my father was working at Westinghouse in the 1970s, he had people working for him who made more than he did, because they'd been there longer.
Now companies increasingly have to pay employees market price for the work they do. One reason is that employees no longer trust companies to deliver [deferred rewards](ladder.html): why work to accumulate deferred rewards at a company that might go bankrupt, or be taken over and have all its implicit obligations wiped out? The other is that some companies broke ranks and started to pay young employees large amounts. This was particularly true in consulting, law, and finance, where it led to the phenomenon of yuppies. The word is rarely used today because it's no longer surprising to see a 25 year old with money, but in 1985 the sight of a 25 year old _professional_ able to afford a new BMW was so novel that it called forth a new word.
The classic yuppie worked for a small organization. He didn't work for General Widget, but for the law firm that handled General Widget's acquisitions or the investment bank that floated their bond issues.
Startups and yuppies entered the American conceptual vocabulary roughly simultaneously in the late 1970s and early 1980s. I don't think there was a causal connection. Startups happened because technology started to change so fast that big companies could no longer keep a lid on the smaller ones. I don't think the rise of yuppies was inspired by it; it seems more as if there was a change in the social conventions (and perhaps the laws) governing the way big companies worked. But the two phenomena rapidly fused to produce a principle that now seems obvious: paying energetic young people market rates, and getting correspondingly high performance from them.
At about the same time the US economy rocketed out of the doldrums that had afflicted it for most of the 1970s. Was there a connection? I don't know enough to say, but it felt like it at the time. There was a lot of energy released.
\_\_\_\_\_
Countries worried about their competitiveness are right to be concerned about the number of startups started within them. But they would do even better to examine the underlying principle. Do they let energetic young people get paid market rate for the work they do? The young are the test, because when people aren't rewarded according to performance, they're invariably rewarded according to seniority instead.
All it takes is a few beachheads in your economy that pay for performance. Measurement spreads like heat. If one part of a society is better at measurement than others, it tends to push the others to do better. If people who are young but smart and driven can make more by starting their own companies than by working for existing ones, the existing companies are forced to pay more to keep them. So market rates gradually permeate every organization, even the government. \[[3](#f3n)\]
The measurement of performance will tend to push even the organizations issuing credentials into line. When we were kids I used to annoy my sister by ordering her to do things I knew she was about to do anyway. As credentials are superseded by performance, a similar role is the best former gatekeepers can hope for. Once credential granting institutions are no longer in the self-fullfilling prophecy business, they'll have to work harder to predict the future.
\_\_\_\_\_
Credentials are a step beyond bribery and influence. But they're not the final step. There's an even better way to block the transmission of power between generations: to encourage the trend toward an economy made of more, smaller units. Then you can measure what credentials merely predict.
No one likes the transmission of power between generations—not the left or the right. But the market forces favored by the right turn out to be a better way of preventing it than the credentials the left are forced to fall back on.
The era of credentials began to end when the power of large organizations [peaked](highres.html) in the late twentieth century. Now we seem to be entering a new era based on measurement. The reason the new model has advanced so rapidly is that it works so much better. It shows no sign of slowing.
**Notes**
\[1\] Miyazaki, Ichisada (Conrad Schirokauer trans.), _China's Examination Hell: The Civil Service Examinations of Imperial China,_ Yale University Press, 1981.
Scribes in ancient Egypt took exams, but they were more the type of proficiency test any apprentice might have to pass.
\[2\] When I say the raison d'etre of prep schools is to get kids into better colleges, I mean this in the narrowest sense. I'm not saying that's all prep schools do, just that if they had zero effect on college admissions there would be far less demand for them.
\[3\] Progressive tax rates will tend to damp this effect, however, by decreasing the difference between good and bad measurers.
**Thanks** to Trevor Blackwell, Sarah Harlin, Jessica Livingston, and David Sloo for reading drafts of this. |
4 | High Resolution Fundraising | September 2010 | The reason startups have been using [more convertible notes](http://twitter.com/paulg/status/22319113993) in angel rounds is that they make deals close faster. By making it easier for startups to give different prices to different investors, they help them break the sort of deadlock that happens when investors all wait to see who else is going to invest.
By far the biggest influence on investors' opinions of a startup is the opinion of other investors. There are very, very few who simply decide for themselves. Any startup founder can tell you the most common question they hear from investors is not about the founders or the product, but "who else is investing?"
That tends to produce deadlocks. Raising an old-fashioned fixed-size equity round can take weeks, because all the angels sit around waiting for the others to commit, like competitors in a bicycle sprint who deliberately ride slowly at the start so they can follow whoever breaks first.
Convertible notes let startups beat such deadlocks by rewarding investors willing to move first with lower (effective) valuations. Which they deserve because they're taking more risk. It's much safer to invest in a startup Ron Conway has already invested in; someone who comes after him should pay a higher price.
The reason convertible notes allow more flexibility in price is that valuation caps aren't actual valuations, and notes are cheap and easy to do. So you can do high-resolution fundraising: if you wanted you could have a separate note with a different cap for each investor.
That cap need not simply rise monotonically. A startup could also give better deals to investors they expected to help them most. The point is simply that different investors, whether because of the help they offer or their willingness to commit, have different values for startups, and their terms should reflect that.
Different terms for different investors is clearly the way of the future. Markets always evolve toward higher resolution. You may not need to use convertible notes to do it. With sufficiently lightweight standardized equity terms (and some changes in investors' and lawyers' expectations about equity rounds) you might be able to do the same thing with equity instead of debt. Either would be fine with startups, so long as they can easily change their valuation.
Deadlocks weren't the only problem with fixed-size equity rounds. Another was that startups had to decide in advance how much to raise. I think it's a mistake for a startup to fix upon a specific number. If investors are easily convinced, the startup should raise more now, and if investors are skeptical, the startup should take a smaller amount and use that to get the company to the point where it's more convincing.
It's just not reasonable to expect startups to pick an optimal round size in advance, because that depends on the reactions of investors, and those are impossible to predict.
Fixed-size, multi-investor angel rounds are such a bad idea for startups that one wonders why things were ever done that way. One possibility is that this custom reflects the way investors like to collude when they can get away with it. But I think the actual explanation is less sinister. I think angels (and their lawyers) organized rounds this way in unthinking imitation of VC series A rounds. In a series A, a fixed-size equity round with a lead makes sense, because there is usually just one big investor, who is unequivocally the lead. Fixed-size series A rounds already are high res. But the more investors you have in a round, the less sense it makes for everyone to get the same price.
The most interesting question here may be what high res fundraising will do to the world of investors. Bolder investors will now get rewarded with lower prices. But more important, in a hits-driven business, is that they'll be able to get into the deals they want. Whereas the "who else is investing?" type of investors will not only pay higher prices, but may not be able to get into the best deals at all.
**Thanks** to Immad Akhund, Sam Altman, John Bautista, Pete Koomen, Jessica Livingston, Dan Siroker, Harj Taggar, and Fred Wilson for reading drafts of this. |
5 | Where to See Silicon Valley | October 2010 | Silicon Valley proper is mostly suburban sprawl. At first glance it doesn't seem there's anything to see. It's not the sort of place that has conspicuous monuments. But if you look, there are subtle signs you're in a place that's different from other places.
**1\. [Stanford University](http://maps.google.com/maps?q=stanford+university)**
Stanford is a strange place. Structurally it is to an ordinary university what suburbia is to a city. It's enormously spread out, and feels surprisingly empty much of the time. But notice the weather. It's probably perfect. And notice the beautiful mountains to the west. And though you can't see it, cosmopolitan San Francisco is 40 minutes to the north. That combination is much of the reason Silicon Valley grew up around this university and not some other one.
**2\. [University Ave](http://maps.google.com/maps?q=university+and+ramona+palo+alto)**
A surprising amount of the work of the Valley is done in the cafes on or just off University Ave in Palo Alto. If you visit on a weekday between 10 and 5, you'll often see founders pitching investors. In case you can't tell, the founders are the ones leaning forward eagerly, and the investors are the ones sitting back with slightly pained expressions.
**3\. [The Lucky Office](http://maps.google.com/maps?q=165+university+ave+palo+alto)**
The office at 165 University Ave was Google's first. Then it was Paypal's. (Now it's [Wepay](http://wepay.com)'s.) The interesting thing about it is the location. It's a smart move to put a startup in a place with restaurants and people walking around instead of in an office park, because then the people who work there want to stay there, instead of fleeing as soon as conventional working hours end. They go out for dinner together, talk about ideas, and then come back and implement them.
It's important to realize that Google's current location in an office park is not where they started; it's just where they were forced to move when they needed more space. Facebook was till recently across the street, till they too had to move because they needed more space.
**4\. [Old Palo Alto](http://maps.google.com/maps?q=old+palo+alto)**
Palo Alto was not originally a suburb. For the first 100 years or so of its existence, it was a college town out in the countryside. Then in the mid 1950s it was engulfed in a wave of suburbia that raced down the peninsula. But Palo Alto north of Oregon expressway still feels noticeably different from the area around it. It's one of the nicest places in the Valley. The buildings are old (though increasingly they are being torn down and replaced with generic McMansions) and the trees are tall. But houses are very expensive—around $1000 per square foot. This is post-exit Silicon Valley.
**5\. [Sand Hill Road](http://maps.google.com/maps?q=2900+sand+hill+road+menlo+park)**
It's interesting to see the VCs' offices on the north side of Sand Hill Road precisely because they're so boringly uniform. The buildings are all more or less the same, their exteriors express very little, and they are arranged in a confusing maze. (I've been visiting them for years and I still occasionally get lost.) It's not a coincidence. These buildings are a pretty accurate reflection of the VC business.
If you go on a weekday you may see groups of founders there to meet VCs. But mostly you won't see anyone; bustling is the last word you'd use to describe the atmos. Visiting Sand Hill Road reminds you that the opposite of "down and dirty" would be "up and clean."
**6\. [Castro Street](http://maps.google.com/maps?q=castro+and+villa+mountain+view)**
It's a tossup whether Castro Street or University Ave should be considered the heart of the Valley now. University Ave would have been 10 years ago. But Palo Alto is getting expensive. Increasingly startups are located in Mountain View, and Palo Alto is a place they come to meet investors. Palo Alto has a lot of different cafes, but there is one that clearly dominates in Mountain View: [Red Rock](http://maps.google.com/places/us/ca/mountain-view/castro-st/201/-red-rock-coffee).
**7\. [Google](http://maps.google.com/maps?q=charleston+road+mountain+view)**
Google spread out from its first building in Mountain View to a lot of the surrounding ones. But because the buildings were built at different times by different people, the place doesn't have the sterile, walled-off feel that a typical large company's headquarters have. It definitely has a flavor of its own though. You sense there is something afoot. The general atmos is vaguely utopian; there are lots of Priuses, and people who look like they drive them.
You can't get into Google unless you know someone there. It's very much worth seeing inside if you can, though. Ditto for Facebook, at the end of California Ave in Palo Alto, though there is nothing to see outside.
**8\. [Skyline Drive](http://maps.google.com/maps?q=skylonda)**
Skyline Drive runs along the crest of the Santa Cruz mountains. On one side is the Valley, and on the other is the sea—which because it's cold and foggy and has few harbors, plays surprisingly little role in the lives of people in the Valley, considering how close it is. Along some parts of Skyline the dominant trees are huge redwoods, and in others they're live oaks. Redwoods mean those are the parts where the fog off the coast comes in at night; redwoods condense rain out of fog. The MROSD manages a collection of [great walking trails](http://www.openspace.org/) off Skyline.
**9\. [280](http://maps.google.com/maps?q=interstate+280+san+mateo)**
Silicon Valley has two highways running the length of it: 101, which is pretty ugly, and 280, which is one of the more beautiful highways in the world. I always take 280 when I have a choice. Notice the long narrow lake to the west? That's the San Andreas Fault. It runs along the base of the hills, then heads uphill through Portola Valley. One of the MROSD trails runs [right along the fault](http://www.openspace.org/preserves/pr_los_trancos.asp). A string of rich neighborhoods runs along the foothills to the west of 280: Woodside, Portola Valley, Los Altos Hills, Saratoga, Los Gatos.
[SLAC](http://www.flickr.com/photos/38037974@N00/3890299362/) goes right under 280 a little bit south of Sand Hill Road. And a couple miles south of that is the Valley's equivalent of the "Welcome to Las Vegas" sign: [The Dish](http://www.flickr.com/photos/paulbarroga/3443486941/).
**Notes**
I skipped the [Computer History Museum](http://www.computerhistory.org/) because this is a list of where to see the Valley itself, not where to see artifacts from it. I also skipped San Jose. San Jose calls itself the capital of Silicon Valley, but when people in the Valley use the phrase "the city," they mean San Francisco. San Jose is a dotted line on a map.
**Thanks** to Sam Altman, Paul Buchheit, Patrick Collison, and Jessica Livingston for reading drafts of this. |
6 | Keep Your Identity Small | February 2009 | I finally realized today why politics and religion yield such uniquely useless discussions.
As a rule, any mention of religion on an online forum degenerates into a religious argument. Why? Why does this happen with religion and not with Javascript or baking or other topics people talk about on forums?
What's different about religion is that people don't feel they need to have any particular expertise to have opinions about it. All they need is strongly held beliefs, and anyone can have those. No thread about Javascript will grow as fast as one about religion, because people feel they have to be over some threshold of expertise to post comments about that. But on religion everyone's an expert.
Then it struck me: this is the problem with politics too. Politics, like religion, is a topic where there's no threshold of expertise for expressing an opinion. All you need is strong convictions.
Do religion and politics have something in common that explains this similarity? One possible explanation is that they deal with questions that have no definite answers, so there's no back pressure on people's opinions. Since no one can be proven wrong, every opinion is equally valid, and sensing this, everyone lets fly with theirs.
But this isn't true. There are certainly some political questions that have definite answers, like how much a new government policy will cost. But the more precise political questions suffer the same fate as the vaguer ones.
I think what religion and politics have in common is that they become part of people's identity, and people can never have a fruitful argument about something that's part of their identity. By definition they're partisan.
Which topics engage people's identity depends on the people, not the topic. For example, a discussion about a battle that included citizens of one or more of the countries involved would probably degenerate into a political argument. But a discussion today about a battle that took place in the Bronze Age probably wouldn't. No one would know what side to be on. So it's not politics that's the source of the trouble, but identity. When people say a discussion has degenerated into a religious war, what they really mean is that it has started to be driven mostly by people's identities. \[[1](#f1n)\]
Because the point at which this happens depends on the people rather than the topic, it's a mistake to conclude that because a question tends to provoke religious wars, it must have no answer. For example, the question of the relative merits of programming languages often degenerates into a religious war, because so many programmers identify as X programmers or Y programmers. This sometimes leads people to conclude the question must be unanswerable—that all languages are equally good. Obviously that's false: anything else people make can be well or badly designed; why should this be uniquely impossible for programming languages? And indeed, you can have a fruitful discussion about the relative merits of programming languages, so long as you exclude people who respond from identity.
More generally, you can have a fruitful discussion about a topic only if it doesn't engage the identities of any of the participants. What makes politics and religion such minefields is that they engage so many people's identities. But you could in principle have a useful conversation about them with some people. And there are other topics that might seem harmless, like the relative merits of Ford and Chevy pickup trucks, that you couldn't safely talk about with [others](http://www.theledger.com/apps/pbcs.dll/article?AID=/20060418/NEWS/604180378/1039).
The most intriguing thing about this theory, if it's right, is that it explains not merely which kinds of discussions to avoid, but how to have better ideas. If people can't think clearly about anything that has become part of their identity, then all other things being equal, the best plan is to let as few things into your identity as possible. \[[2](#f2n)\]
Most people reading this will already be fairly tolerant. But there is a step beyond thinking of yourself as x but tolerating y: not even to consider yourself an x. The more labels you have for yourself, the dumber they make you.
**Notes**
\[1\] When that happens, it tends to happen fast, like a core going critical. The threshold for participating goes down to zero, which brings in more people. And they tend to say incendiary things, which draw more and angrier counterarguments.
\[2\] There may be some things it's a net win to include in your identity. For example, being a scientist. But arguably that is more of a placeholder than an actual label—like putting NMI on a form that asks for your middle initial—because it doesn't commit you to believing anything in particular. A scientist isn't committed to believing in natural selection in the same way a biblical literalist is committed to rejecting it. All he's committed to is following the evidence wherever it leads.
Considering yourself a scientist is equivalent to putting a sign in a cupboard saying "this cupboard must be kept empty." Yes, strictly speaking, you're putting something in the cupboard, but not in the ordinary sense.
**Thanks** to Sam Altman, Trevor Blackwell, Paul Buchheit, and Robert Morris for reading drafts of this. |
7 | What Kate Saw in Silicon Valley | August 2009 | Kate Courteau is the architect who designed Y Combinator's office. Recently we managed to recruit her to help us run YC when she's not busy with architectural projects. Though she'd heard a lot about YC since the beginning, the last 9 months have been a total immersion.
I've been around the startup world for so long that it seems normal to me, so I was curious to hear what had surprised her most about it. This was her list:
**1\. How many startups fail.**
Kate knew in principle that startups were very risky, but she was surprised to see how constant the threat of failure was — not just for the minnows, but even for the famous startups whose founders came to speak at YC dinners.
**2\. How much startups' ideas change.**
As usual, by Demo Day about half the startups were doing something significantly different than they started with. We encourage that. Starting a startup is like science in that you have to follow the truth wherever it leads. In the rest of the world, people don't start things till they're sure what they want to do, and once started they tend to continue on their initial path even if it's mistaken.
**3\. How little money it can take to start a startup.**
In Kate's world, everything is still physical and expensive. You can barely renovate a bathroom for the cost of starting a startup.
**4\. How scrappy founders are.**
That was her actual word. I agree with her, but till she mentioned this it never occurred to me how little this quality is appreciated in most of the rest of the world. It wouldn't be a compliment in most organizations to call someone scrappy.
What does it mean, exactly? It's basically the diminutive form of belligerent. Someone who's scrappy manages to be both threatening and undignified at the same time. Which seems to me exactly what one would want to be, in any kind of work. If you're not threatening, you're probably not doing anything new, and dignity is merely a sort of plaque.
**5\. How tech-saturated Silicon Valley is.**
"It seems like everybody here is in the industry." That isn't literally true, but there is a qualitative difference between Silicon Valley and other places. You tend to keep your voice down, because there's a good chance the person at the next table would know some of the people you're talking about. I never felt that in Boston. The good news is, there's also a good chance the person at the next table could help you in some way.
**6\. That the speakers at YC were so consistent in their advice.**
Actually, I've noticed this too. I always worry the speakers will put us in an embarrassing position by contradicting what we tell the startups, but it happens surprisingly rarely.
When I asked her what specific things she remembered speakers always saying, she mentioned: that the way to succeed was to launch something fast, listen to users, and then iterate; that startups required resilience because they were always an emotional rollercoaster; and that most VCs were sheep.
I've been impressed by how consistently the speakers advocate launching fast and iterating. That was contrarian advice 10 years ago, but it's clearly now the established practice.
**7\. How casual successful startup founders are.**
Most of the famous founders in Silicon Valley are people you'd overlook on the street. It's not merely that they don't dress up. They don't project any kind of aura of power either. "They're not trying to impress anyone."
Interestingly, while Kate said that she could never pick out successful founders, she could recognize VCs, both by the way they dressed and the way they carried themselves.
**8\. How important it is for founders to have people to ask for advice.**
(I swear I didn't prompt this one.) Without advice "they'd just be sort of lost." Fortunately, there are a lot of people to help them. There's a strong tradition within YC of helping other YC-funded startups. But we didn't invent that idea: it's just a slightly more concentrated form of existing Valley culture.
**9\. What a solitary task startups are.**
Architects are constantly interacting face to face with other people, whereas doing a technology startup, at least, tends to require long stretches of uninterrupted time to work. "You could do it in a box."
By inverting this list, we can get a portrait of the "normal" world. It's populated by people who talk a lot with one another as they work slowly but harmoniously on conservative, expensive projects whose destinations are decided in advance, and who carefully adjust their manner to reflect their position in the hierarchy.
That's also a fairly accurate description of the past. So startup culture may not merely be different in the way you'd expect any subculture to be, but a leading indicator. |
8 | Revenge of the Nerds | May 2002 | "We were after the C++ programmers. We managed to drag a lot of them about halfway to Lisp."
\- Guy Steele, co-author of the Java spec
In the software business there is an ongoing struggle between the pointy-headed academics, and another equally formidable force, the pointy-haired bosses. Everyone knows who the pointy-haired boss is, right? I think most people in the technology world not only recognize this cartoon character, but know the actual person in their company that he is modelled upon.
The pointy-haired boss miraculously combines two qualities that are common by themselves, but rarely seen together: (a) he knows nothing whatsoever about technology, and (b) he has very strong opinions about it.
Suppose, for example, you need to write a piece of software. The pointy-haired boss has no idea how this software has to work, and can't tell one programming language from another, and yet he knows what language you should write it in. Exactly. He thinks you should write it in Java.
Why does he think this? Let's take a look inside the brain of the pointy-haired boss. What he's thinking is something like this. Java is a standard. I know it must be, because I read about it in the press all the time. Since it is a standard, I won't get in trouble for using it. And that also means there will always be lots of Java programmers, so if the programmers working for me now quit, as programmers working for me mysteriously always do, I can easily replace them.
Well, this doesn't sound that unreasonable. But it's all based on one unspoken assumption, and that assumption turns out to be false. The pointy-haired boss believes that all programming languages are pretty much equivalent. If that were true, he would be right on target. If languages are all equivalent, sure, use whatever language everyone else is using.
But all languages are not equivalent, and I think I can prove this to you without even getting into the differences between them. If you asked the pointy-haired boss in 1992 what language software should be written in, he would have answered with as little hesitation as he does today. Software should be written in C++. But if languages are all equivalent, why should the pointy-haired boss's opinion ever change? In fact, why should the developers of Java have even bothered to create a new language?
Presumably, if you create a new language, it's because you think it's better in some way than what people already had. And in fact, Gosling makes it clear in the first Java white paper that Java was designed to fix some problems with C++. So there you have it: languages are not all equivalent. If you follow the trail through the pointy-haired boss's brain to Java and then back through Java's history to its origins, you end up holding an idea that contradicts the assumption you started with.
So, who's right? James Gosling, or the pointy-haired boss? Not surprisingly, Gosling is right. Some languages _are_ better, for certain problems, than others. And you know, that raises some interesting questions. Java was designed to be better, for certain problems, than C++. What problems? When is Java better and when is C++? Are there situations where other languages are better than either of them?
Once you start considering this question, you have opened a real can of worms. If the pointy-haired boss had to think about the problem in its full complexity, it would make his brain explode. As long as he considers all languages equivalent, all he has to do is choose the one that seems to have the most momentum, and since that is more a question of fashion than technology, even he can probably get the right answer. But if languages vary, he suddenly has to solve two simultaneous equations, trying to find an optimal balance between two things he knows nothing about: the relative suitability of the twenty or so leading languages for the problem he needs to solve, and the odds of finding programmers, libraries, etc. for each. If that's what's on the other side of the door, it is no surprise that the pointy-haired boss doesn't want to open it.
The disadvantage of believing that all programming languages are equivalent is that it's not true. But the advantage is that it makes your life a lot simpler. And I think that's the main reason the idea is so widespread. It is a _comfortable_ idea.
We know that Java must be pretty good, because it is the cool, new programming language. Or is it? If you look at the world of programming languages from a distance, it looks like Java is the latest thing. (From far enough away, all you can see is the large, flashing billboard paid for by Sun.) But if you look at this world up close, you find that there are degrees of coolness. Within the hacker subculture, there is another language called Perl that is considered a lot cooler than Java. Slashdot, for example, is generated by Perl. I don't think you would find those guys using Java Server Pages. But there is another, newer language, called Python, whose users tend to look down on Perl, and [more](accgen.html) waiting in the wings.
If you look at these languages in order, Java, Perl, Python, you notice an interesting pattern. At least, you notice this pattern if you are a Lisp hacker. Each one is progressively more like Lisp. Python copies even features that many Lisp hackers consider to be mistakes. You could translate simple Lisp programs into Python line for line. It's 2002, and programming languages have almost caught up with 1958.
**Catching Up with Math**
What I mean is that Lisp was first discovered by John McCarthy in 1958, and popular programming languages are only now catching up with the ideas he developed then.
Now, how could that be true? Isn't computer technology something that changes very rapidly? I mean, in 1958, computers were refrigerator-sized behemoths with the processing power of a wristwatch. How could any technology that old even be relevant, let alone superior to the latest developments?
I'll tell you how. It's because Lisp was not really designed to be a programming language, at least not in the sense we mean today. What we mean by a programming language is something we use to tell a computer what to do. McCarthy did eventually intend to develop a programming language in this sense, but the Lisp that we actually ended up with was based on something separate that he did as a [theoretical exercise](rootsoflisp.html)\-- an effort to define a more convenient alternative to the Turing Machine. As McCarthy said later,
> Another way to show that Lisp was neater than Turing machines was to write a universal Lisp function and show that it is briefer and more comprehensible than the description of a universal Turing machine. This was the Lisp function [_eval_](https://sep.yimg.com/ty/cdn/paulgraham/jmc.lisp?t=1595850613&)..., which computes the value of a Lisp expression.... Writing _eval_ required inventing a notation representing Lisp functions as Lisp data, and such a notation was devised for the purposes of the paper with no thought that it would be used to express Lisp programs in practice.
What happened next was that, some time in late 1958, Steve Russell, one of McCarthy's grad students, looked at this definition of _eval_ and realized that if he translated it into machine language, the result would be a Lisp interpreter.
This was a big surprise at the time. Here is what McCarthy said about it later in an interview:
> Steve Russell said, look, why don't I program this _eval_..., and I said to him, ho, ho, you're confusing theory with practice, this _eval_ is intended for reading, not for computing. But he went ahead and did it. That is, he compiled the _eval_ in my paper into \[IBM\] 704 machine code, fixing bugs, and then advertised this as a Lisp interpreter, which it certainly was. So at that point Lisp had essentially the form that it has today....
Suddenly, in a matter of weeks I think, McCarthy found his theoretical exercise transformed into an actual programming language-- and a more powerful one than he had intended.
So the short explanation of why this 1950s language is not obsolete is that it was not technology but math, and math doesn't get stale. The right thing to compare Lisp to is not 1950s hardware, but, say, the Quicksort algorithm, which was discovered in 1960 and is still the fastest general-purpose sort.
There is one other language still surviving from the 1950s, Fortran, and it represents the opposite approach to language design. Lisp was a piece of theory that unexpectedly got turned into a programming language. Fortran was developed intentionally as a programming language, but what we would now consider a very low-level one.
[Fortran I](history.html), the language that was developed in 1956, was a very different animal from present-day Fortran. Fortran I was pretty much assembly language with math. In some ways it was less powerful than more recent assembly languages; there were no subroutines, for example, only branches. Present-day Fortran is now arguably closer to Lisp than to Fortran I.
Lisp and Fortran were the trunks of two separate evolutionary trees, one rooted in math and one rooted in machine architecture. These two trees have been converging ever since. Lisp started out powerful, and over the next twenty years got fast. So-called mainstream languages started out fast, and over the next forty years gradually got more powerful, until now the most advanced of them are fairly close to Lisp. Close, but they are still missing a few things....
**What Made Lisp Different**
When it was first developed, Lisp embodied nine new ideas. Some of these we now take for granted, others are only seen in more advanced languages, and two are still unique to Lisp. The nine ideas are, in order of their adoption by the mainstream,
1. Conditionals. A conditional is an if-then-else construct. We take these for granted now, but Fortran I didn't have them. It had only a conditional goto closely based on the underlying machine instruction.
2. A function type. In Lisp, functions are a data type just like integers or strings. They have a literal representation, can be stored in variables, can be passed as arguments, and so on.
3. Recursion. Lisp was the first programming language to support it.
4. Dynamic typing. In Lisp, all variables are effectively pointers. Values are what have types, not variables, and assigning or binding variables means copying pointers, not what they point to.
5. Garbage-collection.
6. Programs composed of expressions. Lisp programs are trees of expressions, each of which returns a value. This is in contrast to Fortran and most succeeding languages, which distinguish between expressions and statements.
It was natural to have this distinction in Fortran I because you could not nest statements. And so while you needed expressions for math to work, there was no point in making anything else return a value, because there could not be anything waiting for it.
This limitation went away with the arrival of block-structured languages, but by then it was too late. The distinction between expressions and statements was entrenched. It spread from Fortran into Algol and then to both their descendants.
7. A symbol type. Symbols are effectively pointers to strings stored in a hash table. So you can test equality by comparing a pointer, instead of comparing each character.
8. A notation for code using trees of symbols and constants.
9. The whole language there all the time. There is no real distinction between read-time, compile-time, and runtime. You can compile or run code while reading, read or run code while compiling, and read or compile code at runtime.
Running code at read-time lets users reprogram Lisp's syntax; running code at compile-time is the basis of macros; compiling at runtime is the basis of Lisp's use as an extension language in programs like Emacs; and reading at runtime enables programs to communicate using s-expressions, an idea recently reinvented as XML.
When Lisp first appeared, these ideas were far removed from ordinary programming practice, which was dictated largely by the hardware available in the late 1950s. Over time, the default language, embodied in a succession of popular languages, has gradually evolved toward Lisp. Ideas 1-5 are now widespread. Number 6 is starting to appear in the mainstream. Python has a form of 7, though there doesn't seem to be any syntax for it.
As for number 8, this may be the most interesting of the lot. Ideas 8 and 9 only became part of Lisp by accident, because Steve Russell implemented something McCarthy had never intended to be implemented. And yet these ideas turn out to be responsible for both Lisp's strange appearance and its most distinctive features. Lisp looks strange not so much because it has a strange syntax as because it has no syntax; you express programs directly in the parse trees that get built behind the scenes when other languages are parsed, and these trees are made of lists, which are Lisp data structures.
Expressing the language in its own data structures turns out to be a very powerful feature. Ideas 8 and 9 together mean that you can write programs that write programs. That may sound like a bizarre idea, but it's an everyday thing in Lisp. The most common way to do it is with something called a _macro._
The term "macro" does not mean in Lisp what it means in other languages. A Lisp macro can be anything from an abbreviation to a compiler for a new language. If you want to really understand Lisp, or just expand your programming horizons, I would [learn more](onlisp.html) about macros.
Macros (in the Lisp sense) are still, as far as I know, unique to Lisp. This is partly because in order to have macros you probably have to make your language look as strange as Lisp. It may also be because if you do add that final increment of power, you can no longer claim to have invented a new language, but only a new dialect of Lisp.
I mention this mostly as a joke, but it is quite true. If you define a language that has car, cdr, cons, quote, cond, atom, eq, and a notation for functions expressed as lists, then you can build all the rest of Lisp out of it. That is in fact the defining quality of Lisp: it was in order to make this so that McCarthy gave Lisp the shape it has.
**Where Languages Matter**
So suppose Lisp does represent a kind of limit that mainstream languages are approaching asymptotically-- does that mean you should actually use it to write software? How much do you lose by using a less powerful language? Isn't it wiser, sometimes, not to be at the very edge of innovation? And isn't popularity to some extent its own justification? Isn't the pointy-haired boss right, for example, to want to use a language for which he can easily hire programmers?
There are, of course, projects where the choice of programming language doesn't matter much. As a rule, the more demanding the application, the more leverage you get from using a powerful language. But plenty of projects are not demanding at all. Most programming probably consists of writing little glue programs, and for little glue programs you can use any language that you're already familiar with and that has good libraries for whatever you need to do. If you just need to feed data from one Windows app to another, sure, use Visual Basic.
You can write little glue programs in Lisp too (I use it as a desktop calculator), but the biggest win for languages like Lisp is at the other end of the spectrum, where you need to write sophisticated programs to solve hard problems in the face of fierce competition. A good example is the [airline fare search program](carl.html) that ITA Software licenses to Orbitz. These guys entered a market already dominated by two big, entrenched competitors, Travelocity and Expedia, and seem to have just humiliated them technologically.
The core of ITA's application is a 200,000 line Common Lisp program that searches many orders of magnitude more possibilities than their competitors, who apparently are still using mainframe-era programming techniques. (Though ITA is also in a sense using a mainframe-era programming language.) I have never seen any of ITA's code, but according to one of their top hackers they use a lot of macros, and I am not surprised to hear it.
**Centripetal Forces**
I'm not saying there is no cost to using uncommon technologies. The pointy-haired boss is not completely mistaken to worry about this. But because he doesn't understand the risks, he tends to magnify them.
I can think of three problems that could arise from using less common languages. Your programs might not work well with programs written in other languages. You might have fewer libraries at your disposal. And you might have trouble hiring programmers.
How much of a problem is each of these? The importance of the first varies depending on whether you have control over the whole system. If you're writing software that has to run on a remote user's machine on top of a buggy, closed operating system (I mention no names), there may be advantages to writing your application in the same language as the OS. But if you control the whole system and have the source code of all the parts, as ITA presumably does, you can use whatever languages you want. If any incompatibility arises, you can fix it yourself.
In server-based applications you can get away with using the most advanced technologies, and I think this is the main cause of what Jonathan Erickson calls the "[programming language renaissance](http://www.byte.com/documents/s=1821/byt20011214s0003/)." This is why we even hear about new languages like Perl and Python. We're not hearing about these languages because people are using them to write Windows apps, but because people are using them on servers. And as software shifts [off the desktop](road.html) and onto servers (a future even Microsoft seems resigned to), there will be less and less pressure to use middle-of-the-road technologies.
As for libraries, their importance also depends on the application. For less demanding problems, the availability of libraries can outweigh the intrinsic power of the language. Where is the breakeven point? Hard to say exactly, but wherever it is, it is short of anything you'd be likely to call an application. If a company considers itself to be in the software business, and they're writing an application that will be one of their products, then it will probably involve several hackers and take at least six months to write. In a project of that size, powerful languages probably start to outweigh the convenience of pre-existing libraries.
The third worry of the pointy-haired boss, the difficulty of hiring programmers, I think is a red herring. How many hackers do you need to hire, after all? Surely by now we all know that software is best developed by teams of less than ten people. And you shouldn't have trouble hiring hackers on that scale for any language anyone has ever heard of. If you can't find ten Lisp hackers, then your company is probably based in the wrong city for developing software.
In fact, choosing a more powerful language probably decreases the size of the team you need, because (a) if you use a more powerful language you probably won't need as many hackers, and (b) hackers who work in more advanced languages are likely to be smarter.
I'm not saying that you won't get a lot of pressure to use what are perceived as "standard" technologies. At Viaweb (now Yahoo Store), we raised some eyebrows among VCs and potential acquirers by using Lisp. But we also raised eyebrows by using generic Intel boxes as servers instead of "industrial strength" servers like Suns, for using a then-obscure open-source Unix variant called FreeBSD instead of a real commercial OS like Windows NT, for ignoring a supposed e-commerce standard called [SET](http://news.com.com/2100-1017-225723.html) that no one now even remembers, and so on.
You can't let the suits make technical decisions for you. Did it alarm some potential acquirers that we used Lisp? Some, slightly, but if we hadn't used Lisp, we wouldn't have been able to write the software that made them want to buy us. What seemed like an anomaly to them was in fact cause and effect.
If you start a startup, don't design your product to please VCs or potential acquirers. _Design your product to please the users._ If you win the users, everything else will follow. And if you don't, no one will care how comfortingly orthodox your technology choices were.
**The Cost of Being Average**
How much do you lose by using a less powerful language? There is actually some data out there about that.
The most convenient measure of power is probably [code size](power.html). The point of high-level languages is to give you bigger abstractions-- bigger bricks, as it were, so you don't need as many to build a wall of a given size. So the more powerful the language, the shorter the program (not simply in characters, of course, but in distinct elements).
How does a more powerful language enable you to write shorter programs? One technique you can use, if the language will let you, is something called [bottom-up programming](progbot.html). Instead of simply writing your application in the base language, you build on top of the base language a language for writing programs like yours, then write your program in it. The combined code can be much shorter than if you had written your whole program in the base language-- indeed, this is how most compression algorithms work. A bottom-up program should be easier to modify as well, because in many cases the language layer won't have to change at all.
Code size is important, because the time it takes to write a program depends mostly on its length. If your program would be three times as long in another language, it will take three times as long to write-- and you can't get around this by hiring more people, because beyond a certain size new hires are actually a net lose. Fred Brooks described this phenomenon in his famous book _The Mythical Man-Month,_ and everything I've seen has tended to confirm what he said.
So how much shorter are your programs if you write them in Lisp? Most of the numbers I've heard for Lisp versus C, for example, have been around 7-10x. But a recent article about ITA in [_New Architect_](http://www.newarchitectmag.com/documents/s=2286/new1015626014044/) magazine said that "one line of Lisp can replace 20 lines of C," and since this article was full of quotes from ITA's president, I assume they got this number from ITA. If so then we can put some faith in it; ITA's software includes a lot of C and C++ as well as Lisp, so they are speaking from experience.
My guess is that these multiples aren't even constant. I think they increase when you face harder problems and also when you have smarter programmers. A really good hacker can squeeze more out of better tools.
As one data point on the curve, at any rate, if you were to compete with ITA and chose to write your software in C, they would be able to develop software twenty times faster than you. If you spent a year on a new feature, they'd be able to duplicate it in less than three weeks. Whereas if they spent just three months developing something new, it would be _five years_ before you had it too.
And you know what? That's the best-case scenario. When you talk about code-size ratios, you're implicitly assuming that you can actually write the program in the weaker language. But in fact there are limits on what programmers can do. If you're trying to solve a hard problem with a language that's too low-level, you reach a point where there is just too much to keep in your head at once.
So when I say it would take ITA's imaginary competitor five years to duplicate something ITA could write in Lisp in three months, I mean five years if nothing goes wrong. In fact, the way things work in most companies, any development project that would take five years is likely never to get finished at all.
I admit this is an extreme case. ITA's hackers seem to be unusually smart, and C is a pretty low-level language. But in a competitive market, even a differential of two or three to one would be enough to guarantee that you'd always be behind.
**A Recipe**
This is the kind of possibility that the pointy-haired boss doesn't even want to think about. And so most of them don't. Because, you know, when it comes down to it, the pointy-haired boss doesn't mind if his company gets their ass kicked, so long as no one can prove it's his fault. The safest plan for him personally is to stick close to the center of the herd.
Within large organizations, the phrase used to describe this approach is "industry best practice." Its purpose is to shield the pointy-haired boss from responsibility: if he chooses something that is "industry best practice," and the company loses, he can't be blamed. He didn't choose, the industry did.
I believe this term was originally used to describe accounting methods and so on. What it means, roughly, is _don't do anything weird._ And in accounting that's probably a good idea. The terms "cutting-edge" and "accounting" do not sound good together. But when you import this criterion into decisions about technology, you start to get the wrong answers.
Technology often _should_ be cutting-edge. In programming languages, as Erann Gat has pointed out, what "industry best practice" actually gets you is not the best, but merely the average. When a decision causes you to develop software at a fraction of the rate of more aggressive competitors, "best practice" is a misnomer.
So here we have two pieces of information that I think are very valuable. In fact, I know it from my own experience. Number 1, languages vary in power. Number 2, most managers deliberately ignore this. Between them, these two facts are literally a recipe for making money. ITA is an example of this recipe in action. If you want to win in a software business, just take on the hardest problem you can find, use the most powerful language you can get, and wait for your competitors' pointy-haired bosses to revert to the mean.
**Appendix: Power**
As an illustration of what I mean about the relative power of programming languages, consider the following problem. We want to write a function that generates accumulators-- a function that takes a number n, and returns a function that takes another number i and returns n incremented by i.
(That's _incremented by_, not plus. An accumulator has to accumulate.)
In Common Lisp this would be (defun foo (n) (lambda (i) (incf n i))) and in Perl 5, sub foo { my ($n) = @\_; sub {$n += shift} } which has more elements than the Lisp version because you have to extract parameters manually in Perl.
In Smalltalk the code is slightly longer than in Lisp foo: n |s| s := n. ^\[:i| s := s+i. \] because although in general lexical variables work, you can't do an assignment to a parameter, so you have to create a new variable s.
In Javascript the example is, again, slightly longer, because Javascript retains the distinction between statements and expressions, so you need explicit return statements to return values: function foo(n) { return function (i) { return n += i } } (To be fair, Perl also retains this distinction, but deals with it in typical Perl fashion by letting you omit returns.)
If you try to translate the Lisp/Perl/Smalltalk/Javascript code into Python you run into some limitations. Because Python doesn't fully support lexical variables, you have to create a data structure to hold the value of n. And although Python does have a function data type, there is no literal representation for one (unless the body is only a single expression) so you need to create a named function to return. This is what you end up with: def foo(n): s = \[n\] def bar(i): s\[0\] += i return s\[0\] return bar Python users might legitimately ask why they can't just write def foo(n): return lambda i: return n += i or even def foo(n): lambda i: n += i and my guess is that they probably will, one day. (But if they don't want to wait for Python to evolve the rest of the way into Lisp, they could always just...)
In OO languages, you can, to a limited extent, simulate a closure (a function that refers to variables defined in enclosing scopes) by defining a class with one method and a field to replace each variable from an enclosing scope. This makes the programmer do the kind of code analysis that would be done by the compiler in a language with full support for lexical scope, and it won't work if more than one function refers to the same variable, but it is enough in simple cases like this.
Python experts seem to agree that this is the preferred way to solve the problem in Python, writing either def foo(n): class acc: def \_\_init\_\_(self, s): self.s = s def inc(self, i): self.s += i return self.s return acc(n).inc or class foo: def \_\_init\_\_(self, n): self.n = n def \_\_call\_\_(self, i): self.n += i return self.n I include these because I wouldn't want Python advocates to say I was misrepresenting the language, but both seem to me more complex than the first version. You're doing the same thing, setting up a separate place to hold the accumulator; it's just a field in an object instead of the head of a list. And the use of these special, reserved field names, especially \_\_call\_\_, seems a bit of a hack.
In the rivalry between Perl and Python, the claim of the Python hackers seems to be that that Python is a more elegant alternative to Perl, but what this case shows is that power is the ultimate elegance: the Perl program is simpler (has fewer elements), even if the syntax is a bit uglier.
How about other languages? In the other languages mentioned in this talk-- Fortran, C, C++, Java, and Visual Basic-- it is not clear whether you can actually solve this problem. Ken Anderson says that the following code is about as close as you can get in Java: public interface Inttoint { public int call(int i); } public static Inttoint foo(final int n) { return new Inttoint() { int s = n; public int call(int i) { s = s + i; return s; }}; } This falls short of the spec because it only works for integers. After many email exchanges with Java hackers, I would say that writing a properly polymorphic version that behaves like the preceding examples is somewhere between damned awkward and impossible. If anyone wants to write one I'd be very curious to see it, but I personally have timed out.
It's not literally true that you can't solve this problem in other languages, of course. The fact that all these languages are Turing-equivalent means that, strictly speaking, you can write any program in any of them. So how would you do it? In the limit case, by writing a Lisp interpreter in the less powerful language.
That sounds like a joke, but it happens so often to varying degrees in large programming projects that there is a name for the phenomenon, Greenspun's Tenth Rule:
> Any sufficiently complicated C or Fortran program contains an ad hoc informally-specified bug-ridden slow implementation of half of Common Lisp.
If you try to solve a hard problem, the question is not whether you will use a powerful enough language, but whether you will (a) use a powerful language, (b) write a de facto interpreter for one, or (c) yourself become a human compiler for one. We see this already begining to happen in the Python example, where we are in effect simulating the code that a compiler would generate to implement a lexical variable.
This practice is not only common, but institutionalized. For example, in the OO world you hear a good deal about "patterns". I wonder if these patterns are not sometimes evidence of case (c), the human compiler, at work. When I see patterns in my programs, I consider it a sign of trouble. The shape of a program should reflect only the problem it needs to solve. Any other regularity in the code is a sign, to me at least, that I'm using abstractions that aren't powerful enough-- often that I'm generating by hand the expansions of some macro that I need to write.
**Notes**
* The IBM 704 CPU was about the size of a refrigerator, but a lot heavier. The CPU weighed 3150 pounds, and the 4K of RAM was in a separate box weighing another 4000 pounds. The Sub-Zero 690, one of the largest household refrigerators, weighs 656 pounds.
* Steve Russell also wrote the first (digital) computer game, Spacewar, in 1962.
* If you want to trick a pointy-haired boss into letting you write software in Lisp, you could try telling him it's XML.
* Here is the accumulator generator in other Lisp dialects: Scheme: (define (foo n) (lambda (i) (set! n (+ n i)) n)) Goo: (df foo (n) (op incf n \_))) Arc: (def foo (n) \[++ n \_\])
* Erann Gat's sad tale about "industry best practice" at JPL inspired me to address this generally misapplied phrase.
* Peter Norvig found that 16 of the 23 patterns in _Design Patterns_ were "[invisible or simpler](http://www.norvig.com/design-patterns/)" in Lisp.
* Thanks to the many people who answered my questions about various languages and/or read drafts of this, including Ken Anderson, Trevor Blackwell, Erann Gat, Dan Giffin, Sarah Harlin, Jeremy Hylton, Robert Morris, Peter Norvig, Guy Steele, and Anton van Straaten. They bear no blame for any opinions expressed.
**Related:**
Many people have responded to this talk, so I have set up an additional page to deal with the issues they have raised: [Re: Revenge of the Nerds](icadmore.html).
It also set off an extensive and often useful discussion on the [LL1](http://www.ai.mit.edu/~gregs/ll1-discuss-archive-html/threads.html) mailing list. See particularly the mail by Anton van Straaten on semantic compression.
Some of the mail on LL1 led me to try to go deeper into the subject of language power in [Succinctness is Power](power.html).
A larger set of canonical implementations of the [accumulator generator benchmark](accgen.html) are collected together on their own page.
You'll find this essay and 14 others in [**_Hackers & Painters_**](http://www.amazon.com/gp/product/0596006624). |
9 | The Python Paradox | August 2004 | In a recent [talk](gh.html) I said something that upset a lot of people: that you could get smarter programmers to work on a Python project than you could to work on a Java project.
I didn't mean by this that Java programmers are dumb. I meant that Python programmers are smart. It's a lot of work to learn a new programming language. And people don't learn Python because it will get them a job; they learn it because they genuinely like to program and aren't satisfied with the languages they already know.
Which makes them exactly the kind of programmers companies should want to hire. Hence what, for lack of a better name, I'll call the Python paradox: if a company chooses to write its software in a comparatively esoteric language, they'll be able to hire better programmers, because they'll attract only those who cared enough to learn it. And for programmers the paradox is even more pronounced: the language to learn, if you want to get a good job, is a language that people don't learn merely to get a job.
Only a few companies have been smart enough to realize this so far. But there is a kind of selection going on here too: they're exactly the companies programmers would most like to work for. Google, for example. When they advertise Java programming jobs, they also want Python experience.
A friend of mine who knows nearly all the widely used languages uses Python for most of his projects. He says the main reason is that he likes the way source code looks. That may seem a frivolous reason to choose one language over another. But it is not so frivolous as it sounds: when you program, you spend more time reading code than writing it. You push blobs of source code around the way a sculptor does blobs of clay. So a language that makes source code ugly is maddening to an exacting programmer, as clay full of lumps would be to a sculptor.
At the mention of ugly source code, people will of course think of Perl. But the superficial ugliness of Perl is not the sort I mean. Real ugliness is not harsh-looking syntax, but having to build programs out of the wrong concepts. Perl may look like a cartoon character swearing, but there are [cases](icad.html) where it surpasses Python conceptually.
So far, anyway. Both languages are of course [moving](hundred.html) targets. But they share, along with Ruby (and Icon, and Joy, and J, and Lisp, and Smalltalk) the fact that they're created by, and used by, people who really care about programming. And those tend to be the ones who do it well.
If you liked this, you may also like [**_Hackers & Painters_**](http://www.amazon.com/gp/product/0596006624). |
10 | Design and Research | January 2003 | _(This article is derived from a keynote talk at the fall 2002 meeting of NEPLS.)_
Visitors to this country are often surprised to find that Americans like to begin a conversation by asking "what do you do?" I've never liked this question. I've rarely had a neat answer to it. But I think I have finally solved the problem. Now, when someone asks me what I do, I look them straight in the eye and say "I'm designing a [new dialect of Lisp](arc.html)." I recommend this answer to anyone who doesn't like being asked what they do. The conversation will turn immediately to other topics.
I don't consider myself to be doing research on programming languages. I'm just designing one, in the same way that someone might design a building or a chair or a new typeface. I'm not trying to discover anything new. I just want to make a language that will be good to program in. In some ways, this assumption makes life a lot easier.
The difference between design and research seems to be a question of new versus good. Design doesn't have to be new, but it has to be good. Research doesn't have to be good, but it has to be new. I think these two paths converge at the top: the best design surpasses its predecessors by using new ideas, and the best research solves problems that are not only new, but actually worth solving. So ultimately we're aiming for the same destination, just approaching it from different directions.
What I'm going to talk about today is what your target looks like from the back. What do you do differently when you treat programming languages as a design problem instead of a research topic?
The biggest difference is that you focus more on the user. Design begins by asking, who is this for and what do they need from it? A good architect, for example, does not begin by creating a design that he then imposes on the users, but by studying the intended users and figuring out what they need.
Notice I said "what they need," not "what they want." I don't mean to give the impression that working as a designer means working as a sort of short-order cook, making whatever the client tells you to. This varies from field to field in the arts, but I don't think there is any field in which the best work is done by the people who just make exactly what the customers tell them to.
The customer _is_ always right in the sense that the measure of good design is how well it works for the user. If you make a novel that bores everyone, or a chair that's horribly uncomfortable to sit in, then you've done a bad job, period. It's no defense to say that the novel or the chair is designed according to the most advanced theoretical principles.
And yet, making what works for the user doesn't mean simply making what the user tells you to. Users don't know what all the choices are, and are often mistaken about what they really want.
The answer to the paradox, I think, is that you have to design for the user, but you have to design what the user needs, not simply what he says he wants. It's much like being a doctor. You can't just treat a patient's symptoms. When a patient tells you his symptoms, you have to figure out what's actually wrong with him, and treat that.
This focus on the user is a kind of axiom from which most of the practice of good design can be derived, and around which most design issues center.
If good design must do what the user needs, who is the user? When I say that design must be for users, I don't mean to imply that good design aims at some kind of lowest common denominator. You can pick any group of users you want. If you're designing a tool, for example, you can design it for anyone from beginners to experts, and what's good design for one group might be bad for another. The point is, you have to pick some group of users. I don't think you can even talk about good or bad design except with reference to some intended user.
You're most likely to get good design if the intended users include the designer himself. When you design something for a group that doesn't include you, it tends to be for people you consider to be less sophisticated than you, not more sophisticated.
That's a problem, because looking down on the user, however benevolently, seems inevitably to corrupt the designer. I suspect that very few housing projects in the US were designed by architects who expected to live in them. You can see the same thing in programming languages. C, Lisp, and Smalltalk were created for their own designers to use. Cobol, Ada, and Java, were created for other people to use.
If you think you're designing something for idiots, the odds are that you're not designing something good, even for idiots.
Even if you're designing something for the most sophisticated users, though, you're still designing for humans. It's different in research. In math you don't choose abstractions because they're easy for humans to understand; you choose whichever make the proof shorter. I think this is true for the sciences generally. Scientific ideas are not meant to be ergonomic.
Over in the arts, things are very different. Design is all about people. The human body is a strange thing, but when you're designing a chair, that's what you're designing for, and there's no way around it. All the arts have to pander to the interests and limitations of humans. In painting, for example, all other things being equal a painting with people in it will be more interesting than one without. It is not merely an accident of history that the great paintings of the Renaissance are all full of people. If they hadn't been, painting as a medium wouldn't have the prestige that it does.
Like it or not, programming languages are also for people, and I suspect the human brain is just as lumpy and idiosyncratic as the human body. Some ideas are easy for people to grasp and some aren't. For example, we seem to have a very limited capacity for dealing with detail. It's this fact that makes programing languages a good idea in the first place; if we could handle the detail, we could just program in machine language.
Remember, too, that languages are not primarily a form for finished programs, but something that programs have to be developed in. Anyone in the arts could tell you that you might want different mediums for the two situations. Marble, for example, is a nice, durable medium for finished ideas, but a hopelessly inflexible one for developing new ideas.
A program, like a proof, is a pruned version of a tree that in the past has had false starts branching off all over it. So the test of a language is not simply how clean the finished program looks in it, but how clean the path to the finished program was. A design choice that gives you elegant finished programs may not give you an elegant design process. For example, I've written a few macro-defining macros full of nested backquotes that look now like little gems, but writing them took hours of the ugliest trial and error, and frankly, I'm still not entirely sure they're correct.
We often act as if the test of a language were how good finished programs look in it. It seems so convincing when you see the same program written in two languages, and one version is much shorter. When you approach the problem from the direction of the arts, you're less likely to depend on this sort of test. You don't want to end up with a programming language like marble.
For example, it is a huge win in developing software to have an interactive toplevel, what in Lisp is called a read-eval-print loop. And when you have one this has real effects on the design of the language. It would not work well for a language where you have to declare variables before using them, for example. When you're just typing expressions into the toplevel, you want to be able to set x to some value and then start doing things to x. You don't want to have to declare the type of x first. You may dispute either of the premises, but if a language has to have a toplevel to be convenient, and mandatory type declarations are incompatible with a toplevel, then no language that makes type declarations mandatory could be convenient to program in.
In practice, to get good design you have to get close, and stay close, to your users. You have to calibrate your ideas on actual users constantly, especially in the beginning. One of the reasons Jane Austen's novels are so good is that she read them out loud to her family. That's why she never sinks into self-indulgently arty descriptions of landscapes, or pretentious philosophizing. (The philosophy's there, but it's woven into the story instead of being pasted onto it like a label.) If you open an average "literary" novel and imagine reading it out loud to your friends as something you'd written, you'll feel all too keenly what an imposition that kind of thing is upon the reader.
In the software world, this idea is known as Worse is Better. Actually, there are several ideas mixed together in the concept of Worse is Better, which is why people are still arguing about whether worse is actually better or not. But one of the main ideas in that mix is that if you're building something new, you should get a prototype in front of users as soon as possible.
The alternative approach might be called the Hail Mary strategy. Instead of getting a prototype out quickly and gradually refining it, you try to create the complete, finished, product in one long touchdown pass. As far as I know, this is a recipe for disaster. Countless startups destroyed themselves this way during the Internet bubble. I've never heard of a case where it worked.
What people outside the software world may not realize is that Worse is Better is found throughout the arts. In drawing, for example, the idea was discovered during the Renaissance. Now almost every drawing teacher will tell you that the right way to get an accurate drawing is not to work your way slowly around the contour of an object, because errors will accumulate and you'll find at the end that the lines don't meet. Instead you should draw a few quick lines in roughly the right place, and then gradually refine this initial sketch.
In most fields, prototypes have traditionally been made out of different materials. Typefaces to be cut in metal were initially designed with a brush on paper. Statues to be cast in bronze were modelled in wax. Patterns to be embroidered on tapestries were drawn on paper with ink wash. Buildings to be constructed from stone were tested on a smaller scale in wood.
What made oil paint so exciting, when it first became popular in the fifteenth century, was that you could actually make the finished work _from_ the prototype. You could make a preliminary drawing if you wanted to, but you weren't held to it; you could work out all the details, and even make major changes, as you finished the painting.
You can do this in software too. A prototype doesn't have to be just a model; you can refine it into the finished product. I think you should always do this when you can. It lets you take advantage of new insights you have along the way. But perhaps even more important, it's good for morale.
Morale is key in design. I'm surprised people don't talk more about it. One of my first drawing teachers told me: if you're bored when you're drawing something, the drawing will look boring. For example, suppose you have to draw a building, and you decide to draw each brick individually. You can do this if you want, but if you get bored halfway through and start making the bricks mechanically instead of observing each one, the drawing will look worse than if you had merely suggested the bricks.
Building something by gradually refining a prototype is good for morale because it keeps you engaged. In software, my rule is: always have working code. If you're writing something that you'll be able to test in an hour, then you have the prospect of an immediate reward to motivate you. The same is true in the arts, and particularly in oil painting. Most painters start with a blurry sketch and gradually refine it. If you work this way, then in principle you never have to end the day with something that actually looks unfinished. Indeed, there is even a saying among painters: "A painting is never finished, you just stop working on it." This idea will be familiar to anyone who has worked on software.
Morale is another reason that it's hard to design something for an unsophisticated user. It's hard to stay interested in something you don't like yourself. To make something good, you have to be thinking, "wow, this is really great," not "what a piece of shit; those fools will love it."
Design means making things for humans. But it's not just the user who's human. The designer is human too.
Notice all this time I've been talking about "the designer." Design usually has to be under the control of a single person to be any good. And yet it seems to be possible for several people to collaborate on a research project. This seems to me one of the most interesting differences between research and design.
There have been famous instances of collaboration in the arts, but most of them seem to have been cases of molecular bonding rather than nuclear fusion. In an opera it's common for one person to write the libretto and another to write the music. And during the Renaissance, journeymen from northern Europe were often employed to do the landscapes in the backgrounds of Italian paintings. But these aren't true collaborations. They're more like examples of Robert Frost's "good fences make good neighbors." You can stick instances of good design together, but within each individual project, one person has to be in control.
I'm not saying that good design requires that one person think of everything. There's nothing more valuable than the advice of someone whose judgement you trust. But after the talking is done, the decision about what to do has to rest with one person.
Why is it that research can be done by collaborators and design can't? This is an interesting question. I don't know the answer. Perhaps, if design and research converge, the best research is also good design, and in fact can't be done by collaborators. A lot of the most famous scientists seem to have worked alone. But I don't know enough to say whether there is a pattern here. It could be simply that many famous scientists worked when collaboration was less common.
Whatever the story is in the sciences, true collaboration seems to be vanishingly rare in the arts. Design by committee is a synonym for bad design. Why is that so? Is there some way to beat this limitation?
I'm inclined to think there isn't-- that good design requires a dictator. One reason is that good design has to be all of a piece. Design is not just for humans, but for individual humans. If a design represents an idea that fits in one person's head, then the idea will fit in the user's head too.
**Related:**
[Taste for Makers](http://www.paulgraham.com/taste.html) |
11 | What I've Learned from Hacker News | February 2009 | Hacker News was two years old last week. Initially it was supposed to be a side project—an application to sharpen Arc on, and a place for current and future Y Combinator founders to exchange news. It's grown bigger and taken up more time than I expected, but I don't regret that because I've learned so much from working on it.
**Growth**
When we launched in February 2007, weekday traffic was around 1600 daily uniques. It's since [grown](http://ycombinator.com/images/2yeartraffic.png) to around 22,000. This growth rate is a bit higher than I'd like. I'd like the site to grow, since a site that isn't growing at least slowly is probably dead. But I wouldn't want it to grow as large as Digg or Reddit—mainly because that would dilute the character of the site, but also because I don't want to spend all my time dealing with scaling.
I already have problems enough with that. Remember, the original motivation for HN was to test a new programming language, and moreover one that's focused on experimenting with language design, not performance. Every time the site gets slow, I fortify myself by recalling McIlroy and Bentley's famous quote
> The key to performance is elegance, not battalions of special cases.
and look for the bottleneck I can remove with least code. So far I've been able to keep up, in the sense that performance has remained consistently mediocre despite 14x growth. I don't know what I'll do next, but I'll probably think of something.
This is my attitude to the site generally. Hacker News is an experiment, and an experiment in a very young field. Sites of this type are only a few years old. Internet conversation generally is only a few decades old. So we've probably only discovered a fraction of what we eventually will.
That's why I'm so optimistic about HN. When a technology is this young, the existing solutions are usually terrible; which means it must be possible to do much better; which means many problems that seem insoluble aren't. Including, I hope, the problem that has afflicted so many previous communities: being ruined by growth.
**Dilution**
Users have worried about that since the site was a few months old. So far these alarms have been false, but they may not always be. Dilution is a hard problem. But probably soluble; it doesn't mean much that open conversations have "always" been destroyed by growth when "always" equals 20 instances.
But it's important to remember we're trying to solve a new problem, because that means we're going to have to try new things, most of which probably won't work. A couple weeks ago I tried displaying the names of users with the highest average comment scores in orange. \[[1](#f1n)\] That was a mistake. Suddenly a culture that had been more or less united was divided into haves and have-nots. I didn't realize how united the culture had been till I saw it divided. It was painful to watch. \[[2](#f2n)\]
So orange usernames won't be back. (Sorry about that.) But there will be other equally broken-seeming ideas in the future, and the ones that turn out to work will probably seem just as broken as those that don't.
Probably the most important thing I've learned about dilution is that it's measured more in behavior than users. It's bad behavior you want to keep out more than bad people. User behavior turns out to be surprisingly malleable. If people are [expected](http://ycombinator.com/newswelcome.html) to behave well, they tend to; and vice versa.
Though of course forbidding bad behavior does tend to keep away bad people, because they feel uncomfortably constrained in a place where they have to behave well. But this way of keeping them out is gentler and probably also more effective than overt barriers.
It's pretty clear now that the broken windows theory applies to community sites as well. The theory is that minor forms of bad behavior encourage worse ones: that a neighborhood with lots of graffiti and broken windows becomes one where robberies occur. I was living in New York when Giuliani introduced the reforms that made the broken windows theory famous, and the transformation was miraculous. And I was a Reddit user when the opposite happened there, and the transformation was equally dramatic.
I'm not criticizing Steve and Alexis. What happened to Reddit didn't happen out of neglect. From the start they had a policy of censoring nothing except spam. Plus Reddit had different goals from Hacker News. Reddit was a startup, not a side project; its goal was to grow as fast as possible. Combine rapid growth and zero censorship, and the result is a free for all. But I don't think they'd do much differently if they were doing it again. Measured by traffic, Reddit is much more successful than Hacker News.
But what happened to Reddit won't inevitably happen to HN. There are several local maxima. There can be places that are free for alls and places that are more thoughtful, just as there are in the real world; and people will behave differently depending on which they're in, just as they do in the real world.
I've observed this in the wild. I've seen people cross-posting on Reddit and Hacker News who actually took the trouble to write two versions, a flame for Reddit and a more subdued version for HN.
**Submissions**
There are two major types of problems a site like Hacker News needs to avoid: bad stories and bad comments. So far the danger of bad stories seems smaller. The stories on the frontpage now are still roughly the ones that would have been there when HN started.
I once thought I'd have to weight votes to keep crap off the frontpage, but I haven't had to yet. I wouldn't have predicted the frontpage would hold up so well, and I'm not sure why it has. Perhaps only the more thoughtful users care enough to submit and upvote links, so the marginal cost of one random new user approaches zero. Or perhaps the frontpage protects itself, by advertising what type of submission is expected.
The most dangerous thing for the frontpage is stuff that's too easy to upvote. If someone proves a new theorem, it takes some work by the reader to decide whether or not to upvote it. An amusing cartoon takes less. A rant with a rallying cry as the title takes zero, because people vote it up without even reading it.
Hence what I call the Fluff Principle: on a user-voted news site, the links that are easiest to judge will take over unless you take specific measures to prevent it.
Hacker News has two kinds of protections against fluff. The most common types of fluff links are banned as off-topic. Pictures of kittens, political diatribes, and so on are explicitly banned. This keeps out most fluff, but not all of it. Some links are both fluff, in the sense of being very short, and also on topic.
There's no single solution to that. If a link is just an empty rant, editors will sometimes kill it even if it's on topic in the sense of being about hacking, because it's not on topic by the real standard, which is to engage one's intellectual curiosity. If the posts on a site are characteristically of this type I sometimes ban it, which means new stuff at that url is auto-killed. If a post has a linkbait title, editors sometimes rephrase it to be more matter-of-fact. This is especially necessary with links whose titles are rallying cries, because otherwise they become implicit "vote up if you believe such-and-such" posts, which are the most extreme form of fluff.
The techniques for dealing with links have to evolve, because the links do. The existence of aggregators has already affected what they aggregate. Writers now deliberately write things to draw traffic from aggregators—sometimes even specific ones. (No, the irony of this statement is not lost on me.) Then there are the more sinister mutations, like linkjacking—posting a paraphrase of someone else's article and submitting that instead of the original. These can get a lot of upvotes, because a lot of what's good in an article often survives; indeed, the closer the paraphrase is to plagiarism, the more survives. \[[3](#f3n)\]
I think it's important that a site that kills submissions provide a way for users to see what got killed if they want to. That keeps editors honest, and just as importantly, makes users confident they'd know if the editors stopped being honest. HN users can do this by flipping a switch called showdead in their profile. \[[4](#f4n)\]
**Comments**
Bad comments seem to be a harder problem than bad submissions. While the quality of links on the frontpage of HN hasn't changed much, the quality of the median comment may have decreased somewhat.
There are two main kinds of badness in comments: meanness and stupidity. There is a lot of overlap between the two—mean comments are disproportionately likely also to be dumb—but the strategies for dealing with them are different. Meanness is easier to control. You can have rules saying one shouldn't be mean, and if you enforce them it seems possible to keep a lid on meanness.
Keeping a lid on stupidity is harder, perhaps because stupidity is not so easily distinguishable. Mean people are more likely to know they're being mean than stupid people are to know they're being stupid.
The most dangerous form of stupid comment is not the long but mistaken argument, but the dumb joke. Long but mistaken arguments are actually quite rare. There is a strong correlation between comment quality and length; if you wanted to compare the quality of comments on community sites, average length would be a good predictor. Probably the cause is human nature rather than anything specific to comment threads. Probably it's simply that stupidity more often takes the form of having few ideas than wrong ones.
Whatever the cause, stupid comments tend to be short. And since it's hard to write a short comment that's distinguished for the amount of information it conveys, people try to distinguish them instead by being funny. The most tempting format for stupid comments is the supposedly witty put-down, probably because put-downs are the easiest form of humor. \[[5](#f5n)\] So one advantage of forbidding meanness is that it also cuts down on these.
Bad comments are like kudzu: they take over rapidly. Comments have much more effect on new comments than submissions have on new submissions. If someone submits a lame article, the other submissions don't all become lame. But if someone posts a stupid comment on a thread, that sets the tone for the region around it. People reply to dumb jokes with dumb jokes.
Maybe the solution is to add a delay before people can respond to a comment, and make the length of the delay inversely proportional to some prediction of its quality. Then dumb threads would grow slower. \[[6](#f6n)\]
**People**
I notice most of the techniques I've described are conservative: they're aimed at preserving the character of the site rather than enhancing it. I don't think that's a bias of mine. It's due to the shape of the problem. Hacker News had the good fortune to start out good, so in this case it's literally a matter of preservation. But I think this principle would also apply to sites with different origins.
The good things in a community site come from people more than technology; it's mainly in the prevention of bad things that technology comes into play. Technology certainly can enhance discussion. Nested comments do, for example. But I'd rather use a site with primitive features and smart, nice users than a more advanced one whose users were idiots or [trolls](trolls.html).
So the most important thing a community site can do is attract the kind of people it wants. A site trying to be as big as possible wants to attract everyone. But a site aiming at a particular subset of users has to attract just those—and just as importantly, repel everyone else. I've made a conscious effort to do this on HN. The graphic design is as plain as possible, and the site rules discourage dramatic link titles. The goal is that the only thing to interest someone arriving at HN for the first time should be the ideas expressed there.
The downside of tuning a site to attract certain people is that, to those people, it can be too attractive. I'm all too aware how addictive Hacker News can be. For me, as for many users, it's a kind of virtual town square. When I want to take a break from working, I walk into the square, just as I might into Harvard Square or University Ave in the physical world. \[[7](#f7n)\] But an online square is more dangerous than a physical one. If I spent half the day loitering on University Ave, I'd notice. I have to walk a mile to get there, and sitting in a cafe feels different from working. But visiting an online forum takes just a click, and feels superficially very much like working. You may be wasting your time, but you're not idle. Someone is [wrong](http://xkcd.com/386/) on the Internet, and you're fixing the problem.
Hacker News is definitely useful. I've learned a lot from things I've read on HN. I've written several essays that began as comments there. So I wouldn't want the site to go away. But I would like to be sure it's not a net drag on productivity. What a disaster that would be, to attract thousands of smart people to a site that caused them to waste lots of time. I wish I could be 100% sure that's not a description of HN.
I feel like the addictiveness of games and social applications is still a mostly unsolved problem. The situation now is like it was with crack in the 1980s: we've invented terribly addictive new things, and we haven't yet evolved ways to protect ourselves from them. We will eventually, and that's one of the problems I hope to focus on next.
**Notes**
\[1\] I tried ranking users by both average and median comment score, and average (with the high score thrown out) seemed the more accurate predictor of high quality. Median may be the more accurate predictor of low quality though.
\[2\] Another thing I learned from this experiment is that if you're going to distinguish between people, you better be sure you do it right. This is one problem where rapid prototyping doesn't work.
Indeed, that's the intellectually honest argument for not discriminating between various types of people. The reason not to do it is not that everyone's the same, but that it's bad to do wrong and hard to do right.
\[3\] When I catch egregiously linkjacked posts I replace the url with that of whatever they copied. Sites that habitually linkjack get banned.
\[4\] Digg is notorious for its lack of transparency. The root of the problem is not that the guys running Digg are especially sneaky, but that they use the wrong algorithm for generating their frontpage. Instead of bubbling up from the bottom as they get more votes, as on Reddit, stories start at the top and get pushed down by new arrivals.
The reason for the difference is that Digg is derived from Slashdot, while Reddit is derived from Delicious/popular. Digg is Slashdot with voting instead of editors, and Reddit is Delicious/popular with voting instead of bookmarking. (You can still see fossils of their origins in their graphic design.)
Digg's algorithm is very vulnerable to gaming, because any story that makes it onto the frontpage is the new top story. Which in turn forces Digg to respond with extreme countermeasures. A lot of startups have some kind of secret about the subterfuges they had to resort to in the early days, and I suspect Digg's is the extent to which the top stories were de facto chosen by human editors.
\[5\] The dialog on Beavis and Butthead was composed largely of these, and when I read comments on really bad sites I can hear them in their voices.
\[6\] I suspect most of the techniques for discouraging stupid comments have yet to be discovered. Xkcd implemented a particularly clever one in its IRC channel: don't allow the same thing twice. Once someone has said "fail," no one can ever say it again. This would penalize short comments especially, because they have less room to avoid collisions in.
Another promising idea is the [stupid filter](http://stupidfilter.org), which is just like a probabilistic spam filter, but trained on corpora of stupid and non-stupid comments instead.
You may not have to kill bad comments to solve the problem. Comments at the bottom of a long thread are rarely seen, so it may be enough to incorporate a prediction of quality in the comment sorting algorithm.
\[7\] What makes most suburbs so demoralizing is that there's no center to walk to.
**Thanks** to Justin Kan, Jessica Livingston, Robert Morris, Alexis Ohanian, Emmet Shear, and Fred Wilson for reading drafts of this. |
12 | The List of N Things | September 2009 | I bet you the current issue of _Cosmopolitan_ has an article whose title begins with a number. "7 Things He Won't Tell You about Sex," or something like that. Some popular magazines feature articles of this type on the cover of every issue. That can't be happening by accident. Editors must know they attract readers.
Why do readers like the list of n things so much? Mainly because it's easier to read than a regular article. \[[1](#f1n)\] Structurally, the list of n things is a degenerate case of essay. An essay can go anywhere the writer wants. In a list of n things the writer agrees to constrain himself to a collection of points of roughly equal importance, and he tells the reader explicitly what they are.
Some of the work of reading an article is understanding its structure—figuring out what in high school we'd have called its "outline." Not explicitly, of course, but someone who really understands an article probably has something in his brain afterward that corresponds to such an outline. In a list of n things, this work is done for you. Its structure is an exoskeleton.
As well as being explicit, the structure is guaranteed to be of the simplest possible type: a few main points with few to no subordinate ones, and no particular connection between them.
Because the main points are unconnected, the list of n things is random access. There's no thread of reasoning you have to follow. You could read the list in any order. And because the points are independent of one another, they work like watertight compartments in an unsinkable ship. If you get bored with, or can't understand, or don't agree with one point, you don't have to give up on the article. You can just abandon that one and skip to the next. A list of n things is parallel and therefore fault tolerant.
There are times when this format is what a writer wants. One, obviously, is when what you have to say actually is a list of n things. I once wrote an essay about the [mistakes that kill startups](startupmistakes.html), and a few people made fun of me for writing something whose title began with a number. But in that case I really was trying to make a complete catalog of a number of independent things. In fact, one of the questions I was trying to answer was how many there were.
There are other less legitimate reasons for using this format. For example, I use it when I get close to a deadline. If I have to give a talk and I haven't started it a few days beforehand, I'll sometimes play it safe and make the talk a list of n things.
The list of n things is easier for writers as well as readers. When you're writing a real essay, there's always a chance you'll hit a dead end. A real essay is a train of thought, and some trains of thought just peter out. That's an alarming possibility when you have to give a talk in a few days. What if you run out of ideas? The compartmentalized structure of the list of n things protects the writer from his own stupidity in much the same way it protects the reader. If you run out of ideas on one point, no problem: it won't kill the essay. You can take out the whole point if you need to, and the essay will still survive.
Writing a list of n things is so relaxing. You think of n/2 of them in the first 5 minutes. So bang, there's the structure, and you just have to fill it in. As you think of more points, you just add them to the end. Maybe you take out or rearrange or combine a few, but at every stage you have a valid (though initially low-res) list of n things. It's like the sort of programming where you write a version 1 very quickly and then gradually modify it, but at every point have working code—or the style of painting where you begin with a complete but very blurry sketch done in an hour, then spend a week cranking up the resolution.
Because the list of n things is easier for writers too, it's not always a damning sign when readers prefer it. It's not necessarily evidence readers are lazy; it could also mean they don't have much confidence in the writer. The list of n things is in that respect the cheeseburger of essay forms. If you're eating at a restaurant you suspect is bad, your best bet is to order the cheeseburger. Even a bad cook can make a decent cheeseburger. And there are pretty strict conventions about what a cheeseburger should look like. You can assume the cook isn't going to try something weird and artistic. The list of n things similarly limits the damage that can be done by a bad writer. You know it's going to be about whatever the title says, and the format prevents the writer from indulging in any flights of fancy.
Because the list of n things is the easiest essay form, it should be a good one for beginning writers. And in fact it is what most beginning writers are taught. The classic 5 paragraph essay is really a list of n things for n = 3. But the students writing them don't realize they're using the same structure as the articles they read in _Cosmopolitan_. They're not allowed to include the numbers, and they're expected to spackle over the gaps with gratuitous transitions ("Furthermore...") and cap the thing at either end with introductory and concluding paragraphs so it will look superficially like a real essay. \[[2](#f2n)\]
It seems a fine plan to start students off with the list of n things. It's the easiest form. But if we're going to do that, why not do it openly? Let them write lists of n things like the pros, with numbers and no transitions or "conclusion."
There is one case where the list of n things is a dishonest format: when you use it to attract attention by falsely claiming the list is an exhaustive one. I.e. if you write an article that purports to be about _the_ 7 secrets of success. That kind of title is the same sort of reflexive challenge as a whodunit. You have to at least look at the article to check whether they're the same 7 you'd list. Are you overlooking one of the secrets of success? Better check.
It's fine to put "The" before the number if you really believe you've made an exhaustive list. But evidence suggests most things with titles like this are linkbait.
The greatest weakness of the list of n things is that there's so little room for new thought. The main point of essay writing, when done right, is the new ideas you have while doing it. A real essay, as the name implies, is [dynamic](essay.html): you don't know what you're going to write when you start. It will be about whatever you discover in the course of writing it.
This can only happen in a very limited way in a list of n things. You make the title first, and that's what it's going to be about. You can't have more new ideas in the writing than will fit in the watertight compartments you set up initially. And your brain seems to know this: because you don't have room for new ideas, you don't have them.
Another advantage of admitting to beginning writers that the 5 paragraph essay is really a list of n things is that we can warn them about this. It only lets you experience the defining characteristic of essay writing on a small scale: in thoughts of a sentence or two. And it's particularly dangerous that the 5 paragraph essay buries the list of n things within something that looks like a more sophisticated type of essay. If you don't know you're using this form, you don't know you need to escape it.
**Notes**
\[1\] Articles of this type are also startlingly popular on Delicious, but I think that's because [delicious/popular](http://delicious.com/popular) is driven by bookmarking, not because Delicious users are stupid. Delicious users are collectors, and a list of n things seems particularly collectible because it's a collection itself.
\[2\] Most "word problems" in school math textbooks are similarly misleading. They look superficially like the application of math to real problems, but they're not. So if anything they reinforce the impression that math is merely a complicated but pointless collection of stuff to be memorized. |
13 | The Anatomy of Determination | September 2009 | Like all investors, we spend a lot of time trying to learn how to predict which startups will succeed. We probably spend more time thinking about it than most, because we invest the earliest. Prediction is usually all we have to rely on.
We learned quickly that the most important predictor of success is determination. At first we thought it might be intelligence. Everyone likes to believe that's what makes startups succeed. It makes a better story that a company won because its founders were so smart. The PR people and reporters who spread such stories probably believe them themselves. But while it certainly helps to be smart, it's not the deciding factor. There are plenty of people as smart as Bill Gates who achieve nothing.
In most domains, talent is overrated compared to determination—partly because it makes a better story, partly because it gives onlookers an excuse for being lazy, and partly because after a while determination starts to look like talent.
I can't think of any field in which determination is overrated, but the relative importance of determination and talent probably do vary somewhat. Talent probably matters more in types of work that are purer, in the sense that one is solving mostly a single type of problem instead of many different types. I suspect determination would not take you as far in math as it would in, say, organized crime.
I don't mean to suggest by this comparison that types of work that depend more on talent are always more admirable. Most people would agree it's more admirable to be good at math than memorizing long strings of digits, even though the latter depends more on natural ability.
Perhaps one reason people believe startup founders win by being smarter is that intelligence does matter more in technology startups than it used to in earlier types of companies. You probably do need to be a bit smarter to dominate Internet search than you had to be to dominate railroads or hotels or newspapers. And that's probably an ongoing trend. But even in the highest of high tech industries, success still depends more on determination than brains.
If determination is so important, can we isolate its components? Are some more important than others? Are there some you can cultivate?
The simplest form of determination is sheer willfulness. When you want something, you must have it, no matter what.
A good deal of willfulness must be inborn, because it's common to see families where one sibling has much more of it than another. Circumstances can alter it, but at the high end of the scale, nature seems to be more important than nurture. Bad circumstances can break the spirit of a strong-willed person, but I don't think there's much you can do to make a weak-willed person stronger-willed.
Being strong-willed is not enough, however. You also have to be hard on yourself. Someone who was strong-willed but self-indulgent would not be called determined. Determination implies your willfulness is balanced by discipline.
That word balance is a significant one. The more willful you are, the more disciplined you have to be. The stronger your will, the less anyone will be able to argue with you except yourself. And someone has to argue with you, because everyone has base impulses, and if you have more will than discipline you'll just give into them and end up on a local maximum like drug addiction.
We can imagine will and discipline as two fingers squeezing a slippery melon seed. The harder they squeeze, the further the seed flies, but they must both squeeze equally or the seed spins off sideways.
If this is true it has interesting implications, because discipline can be cultivated, and in fact does tend to vary quite a lot in the course of an individual's life. If determination is effectively the product of will and discipline, then you can become more determined by being more disciplined. \[[1](#f1n)\]
Another consequence of the melon seed model is that the more willful you are, the more dangerous it is to be undisciplined. There seem to be plenty of examples to confirm that. In some very energetic people's lives you see something like wing flutter, where they alternate between doing great work and doing absolutely nothing. Externally this would look a lot like bipolar disorder.
The melon seed model is inaccurate in at least one respect, however: it's static. In fact the dangers of indiscipline increase with temptation. Which means, interestingly, that determination tends to erode itself. If you're sufficiently determined to achieve great things, this will probably increase the number of temptations around you. Unless you become proportionally more disciplined, willfulness will then get the upper hand, and your achievement will revert to the mean.
That's why Shakespeare's Caesar thought thin men so dangerous. They weren't tempted by the minor perquisites of power.
The melon seed model implies it's possible to be too disciplined. Is it? I think there probably are people whose willfulness is crushed down by excessive discipline, and who would achieve more if they weren't so hard on themselves. One reason the young sometimes succeed where the old fail is that they don't realize how incompetent they are. This lets them do a kind of deficit spending. When they first start working on something, they overrate their achievements. But that gives them confidence to keep working, and their performance improves. Whereas someone clearer-eyed would see their initial incompetence for what it was, and perhaps be discouraged from continuing.
There's one other major component of determination: ambition. If willfulness and discipline are what get you to your destination, ambition is how you choose it.
I don't know if it's exactly right to say that ambition is a component of determination, but they're not entirely orthogonal. It would seem a misnomer if someone said they were very determined to do something trivially easy.
And fortunately ambition seems to be quite malleable; there's a lot you can do to increase it. Most people don't know how ambitious to be, especially when they're young. They don't know what's hard, or what they're capable of. And this problem is exacerbated by having few peers. Ambitious people are rare, so if everyone is mixed together randomly, as they tend to be early in people's lives, then the ambitious ones won't have many ambitious peers. When you take people like this and put them together with other ambitious people, they bloom like dying plants given water. Probably most ambitious people are starved for the sort of encouragement they'd get from ambitious peers, whatever their age. \[[2](#f2n)\]
Achievements also tend to increase your ambition. With each step you gain confidence to stretch further next time.
So here in sum is how determination seems to work: it consists of willfulness balanced with discipline, aimed by ambition. And fortunately at least two of these three qualities can be cultivated. You may be able to increase your strength of will somewhat; you can definitely learn self-discipline; and almost everyone is practically malnourished when it comes to ambition.
I feel like I understand determination a bit better now. But only a bit: willfulness, discipline, and ambition are all concepts almost as complicated as determination. \[[3](#f3n)\]
Note too that determination and talent are not the whole story. There's a third factor in achievement: how much you like the work. If you really [love](love.html) working on something, you don't need determination to drive you; it's what you'd do anyway. But most types of work have aspects one doesn't like, because most types of work consist of doing things for other people, and it's very unlikely that the tasks imposed by their needs will happen to align exactly with what you want to do.
Indeed, if you want to create the most [wealth](wealth.html), the way to do it is to focus more on their needs than your interests, and make up the difference with determination.
**Notes**
\[1\] Loosely speaking. What I'm claiming with the melon seed model is more like determination is proportionate to wd^m - k|w - d|^n, where w is will and d discipline.
\[2\] Which means one of the best ways to help a society generally is to create [events](http://startupschool.org) and [institutions](http://ycombinator.com) that bring ambitious people together. It's like pulling the control rods out of a reactor: the energy they emit encourages other ambitious people, instead of being absorbed by the normal people they're usually surrounded with.
Conversely, it's probably a mistake to do as some European countries have done and try to ensure none of your universities is significantly better than the others.
\[3\] For example, willfulness clearly has two subcomponents, stubbornness and energy. The first alone yields someone who's stubbornly inert. The second alone yields someone flighty. As willful people get older or otherwise lose their energy, they tend to become merely stubborn.
**Thanks** to Sam Altman, Jessica Livingston, and Robert Morris for reading drafts of this. |
14 | Startups in 13 Sentences | February 2009 | One of the things I always tell startups is a principle I learned from Paul Buchheit: it's better to make a few people really happy than to make a lot of people semi-happy. I was saying recently to a reporter that if I could only tell startups 10 things, this would be one of them. Then I thought: what would the other 9 be?
When I made the list there turned out to be 13:
**1\. Pick good cofounders.**
Cofounders are for a startup what location is for real estate. You can change anything about a house except where it is. In a startup you can change your idea easily, but changing your cofounders is hard. \[[1](#f1n)\] And the success of a startup is almost always a function of its founders.
**2\. Launch fast.**
The reason to launch fast is not so much that it's critical to get your product to market early, but that you haven't really started working on it till you've launched. Launching teaches you what you should have been building. Till you know that you're wasting your time. So the main value of whatever you launch with is as a pretext for engaging users.
**3\. Let your idea evolve.**
This is the second half of launching fast. Launch fast and iterate. It's a big mistake to treat a startup as if it were merely a matter of implementing some brilliant initial idea. As in an essay, most of the ideas appear in the implementing.
**4\. Understand your users.**
You can envision the wealth created by a startup as a rectangle, where one side is the number of users and the other is how much you improve their lives. \[[2](#f2n)\] The second dimension is the one you have most control over. And indeed, the growth in the first will be driven by how well you do in the second. As in science, the hard part is not answering questions but asking them: the hard part is seeing something new that users lack. The better you understand them the better the odds of doing that. That's why so many successful startups make something the founders needed.
**5\. Better to make a few users love you than a lot ambivalent.**
Ideally you want to make large numbers of users love you, but you can't expect to hit that right away. Initially you have to choose between satisfying all the needs of a subset of potential users, or satisfying a subset of the needs of all potential users. Take the first. It's easier to expand userwise than satisfactionwise. And perhaps more importantly, it's harder to lie to yourself. If you think you're 85% of the way to a great product, how do you know it's not 70%? Or 10%? Whereas it's easy to know how many users you have.
**6\. Offer surprisingly good customer service.**
Customers are used to being maltreated. Most of the companies they deal with are quasi-monopolies that get away with atrocious customer service. Your own ideas about what's possible have been unconsciously lowered by such experiences. Try making your customer service not merely good, but [surprisingly good](http://www.diaryofawebsite.com/blog/2008/07/wufoo-and-the-art-of-customer-service/). Go out of your way to make people happy. They'll be overwhelmed; you'll see. In the earliest stages of a startup, it pays to offer customer service on a level that wouldn't scale, because it's a way of learning about your users.
**7\. You make what you measure.**
I learned this one from Joe Kraus. \[[3](#f3n)\] Merely measuring something has an uncanny tendency to improve it. If you want to make your user numbers go up, put a big piece of paper on your wall and every day plot the number of users. You'll be delighted when it goes up and disappointed when it goes down. Pretty soon you'll start noticing what makes the number go up, and you'll start to do more of that. Corollary: be careful what you measure.
**8\. Spend little.**
I can't emphasize enough how important it is for a startup to be cheap. Most startups fail before they make something people want, and the most common form of failure is running out of money. So being cheap is (almost) interchangeable with iterating rapidly. \[[4](#f4n)\] But it's more than that. A culture of cheapness keeps companies young in something like the way exercise keeps people young.
**9\. Get ramen profitable.**
"Ramen profitable" means a startup makes just enough to pay the founders' living expenses. It's not rapid prototyping for business models (though it can be), but more a way of hacking the investment process. Once you cross over into ramen profitable, it completely changes your relationship with investors. It's also great for morale.
**10\. Avoid distractions.**
Nothing kills startups like distractions. The worst type are those that pay money: day jobs, consulting, profitable side-projects. The startup may have more long-term potential, but you'll always interrupt working on it to answer calls from people paying you now. Paradoxically, [fundraising](fundraising.html) is this type of distraction, so try to minimize that too.
**11\. Don't get demoralized.**
Though the immediate cause of death in a startup tends to be running out of money, the underlying cause is usually lack of focus. Either the company is run by stupid people (which can't be fixed with advice) or the people are smart but got demoralized. Starting a startup is a huge moral weight. Understand this and make a conscious effort not to be ground down by it, just as you'd be careful to bend at the knees when picking up a heavy box.
**12\. Don't give up.**
Even if you get demoralized, [don't give up](die.html). You can get surprisingly far by just not giving up. This isn't true in all fields. There are a lot of people who couldn't become good mathematicians no matter how long they persisted. But startups aren't like that. Sheer effort is usually enough, so long as you keep morphing your idea.
**13\. Deals fall through.**
One of the most useful skills we learned from Viaweb was not getting our hopes up. We probably had 20 deals of various types fall through. After the first 10 or so we learned to treat deals as background processes that we should ignore till they terminated. It's very dangerous to morale to start to depend on deals closing, not just because they so often don't, but because it makes them less likely to.
Having gotten it down to 13 sentences, I asked myself which I'd choose if I could only keep one.
Understand your users. That's the key. The essential task in a startup is to create wealth; the dimension of wealth you have most control over is how much you improve users' lives; and the hardest part of that is knowing what to make for them. Once you know what to make, it's mere effort to make it, and most decent hackers are capable of that.
Understanding your users is part of half the principles in this list. That's the reason to launch early, to understand your users. Evolving your idea is the embodiment of understanding your users. Understanding your users well will tend to push you toward making something that makes a few people deeply happy. The most important reason for having surprisingly good customer service is that it helps you understand your users. And understanding your users will even ensure your morale, because when everything else is collapsing around you, having just ten users who love you will keep you going.
**Notes**
\[1\] Strictly speaking it's impossible without a time machine.
\[2\] In practice it's more like a ragged comb.
\[3\] Joe thinks one of the founders of Hewlett Packard said it first, but he doesn't remember which.
\[4\] They'd be interchangeable if markets stood still. Since they don't, working twice as fast is better than having twice as much time. |
End of preview. Expand
in Dataset Viewer.
Dataset Card for Paul Graham Essay Collection Dataset
Dataset Description
This dataset contains a complete collection of essays written by Paul Graham, a renowned programmer, venture capitalist, and essayist. The essays cover a wide range of topics including startups, programming, technology, entrepreneurship, and personal growth. Each essay has been cleaned and processed to extract the title, date of publication, and the full text content.
Dataset Structure
The dataset is provided in a tabular format with the following columns:
title: The title of the essay.date: The date when the essay was originally published, in the format Month YYYY.text: The full text content of the essay.
Data Sources
The essays in this dataset have been sourced from Paul Graham's personal website at http://www.paulgraham.com/.
Data Preprocessing
The essays have undergone the following preprocessing steps:
- Extraction of title and publication date from the essay's metadata.
- Removal of promotional material unrelated to the essay's content.
- Conversion of the essay text to plain text format.
- Cleaning of the text to remove any extraneous whitespace or special characters.
Intended Use and Limitations
This dataset is intended for various natural language processing tasks such as question-answering, summarization, text generation, and text-to-text generation. The essays provide valuable insights and perspectives on a range of topics and can be used for training models or conducting analyses related to startups, technology, and personal growth.
However, it's important to note that the essays reflect the personal views and opinions of Paul Graham and may not be representative of a broader population. The content should be used with appropriate context and understanding of its subjective nature.
License and Attribution
This dataset is released under the MIT License. When using this dataset, please attribute it to Paul Graham and provide a link to his website at http://www.paulgraham.com/.
Contact Information
For any questions or inquiries about this dataset, please contact sgoel9@berkeley.edu.
- Downloads last month
- 12