A ramble about cookbooks and implicit knowledge

Recipes have evolved over the years to include more and more detail. In the Middle Ages, cryptic and elliptic were de rigeur. Take this one, 45 words including the title: 

Frytour blaunched. Take almaundes blaunched, and grynde hem al to doust withouten eny lycour. Do þerto poudour of gyngeuer, sugur, and salt; do þise in a thynne foile. Close it þerinne fast, and frye it in oile; clarifie hony with wyne, & bake it þerwith.

Go to the website I got it from, and you'll see the recipe translated into 211 words of modern English, which are still well short of modern standards of specificity, neglecting to indicate any quantities, the temperature of the oven, or the duration of baking.

Now I welcome modern cookbooks. They are based on the idea that with adequate detail, a cookbook can be more than an aid to the memory of a cook who learned the basics from another cook but cannot quite remember whether the "hony" was to be clarified with "wyne" or ale. It can be a guide to self-study, the sort that still requires trial and error, but no one-on-one apprenticeship. I particularly cherish an old, batter-stained copy of the "Blue Band Kinderkookboek," a Dutch language cookbook targeted to children that is exemplary in its attention to detail and the careful attempts to replace use of the big chef's knife with child-safe, if cumbersome, use of potato peelers, food mills, and such like.

But we are often tempted to think that modern cookbooks are even more than a guide to study. They are, their authors like us to believe, a step-by-step guide that explains all the details, and therefore you cannot go wrong if you follow the recipe.

Bullshit.

Here's a dish I cooked today, described in the sort of mnemonic style I would need to reproduce it:

Heat frozen French beans over high. Cook roux of flour and roasted garlic infused olive oil until it barely browns, add milk, cook some more, season with Chachere's and Parmesan.

Sounds simple (and delicious, I promise). But let's look at the implicit knowledge here.

It turns out that to make a roux, you need a pan that is hot enough overall without any particular part of it being so hot as to burn the roux. This means, in practice, that you need something kind of heavy. If your saucepan is the stamped sheet metal variety, forget it. Or at least, be prepared to work very slowly over very low heat. 

All right, so the pan needs to be kind of heavy. That, of course, is just a heuristic. Really what it needs is good thermal conductivity.

And yet, there are many people who can make excellent sauces who have never heard of "thermal conductivity." All they know is that for sauce, you use the old saucepan from the second cabinet from the left, and it works. Besides, this sort of implicit knowledge depends on the type of heat source, and it is a source of much frustration to experienced cooks switching from gas to electric or vice versa, because there are all these details that you didn't even know you knew that are now different.

Or take the phrase "cook some more." I mean, by that, of course, "cook until the taste suddenly switches from nasty porridge to tasty béchamel." Good luck if you don't know what béchamel tastes like. "No more raw flour taste," as the cookbooks often say? Bleh, doesn't work. Dirty little secret is that even when it's done, the sauce still tastes a little bit like raw flour; you need salt to mask it.

Or "add milk." By which of course is meant "add milk slowly, while stirring fast enough to dissolve lumps in the sauce, but not worrying about lumps of the sort of consistency and size that you know will dissolve just fine later."

Or even the simple "heat frozen French beans." That means "heat frozen French beans until they are just a little bit crispier than you want them to be, because of course they cook some more once you take them off the heat." And "if they are the kind that are clumped together, you need to pry them apart once they defrost a little bit, lest the cooking is uneven." And "it helps to put a lid on the pot, to keep the steam inside so it will defrost and cook the beans on top, but don't leave the lid on too long lest so much water accumulate at the bottom that the bottom beans cook to mush." Et cetera.

Then there's the issue of exactly how a roux is to be stirred, and with what utensil. The process really has two goals: to keep any particular patch of the sauce from drying out and burning, and to dissolve (strictly, to cause to go into suspension) any remaining undissolved flour. It's a lot easier with a balk whisk than with a wire whisk, or, if you prefer a wooden spoon, it's easier with the kind that has a flat side (so you can get into the corners of the pan.)

Now these are details that I know consciously. But my experience is that once you try to teach this sort of thing to somebody else, you suddenly become aware of more details yet.

And there's the real point of this post: it's hard to write teaching material if you haven't attempted to teach from it. 

And that was a simple recipe. 

 

Posted

Unit testing considered awesome

I decided that while I have a whole winter break ahead of me before I start my new job, I might as well brush up on some of the skills I'm likely to need. So this morning, I wrote a (somewhat simplistic) skip list implementation in C++. (I link to it, but please don't take it too seriously—I have no experience whatsoever with C++; that's precisely why I was trying to write some C++.)

But what turned out to be more interesting than C++ was Google Test. I have heard over and over that Unit Testing and Test Driven Development is the One True Way, but I have been so spoiled with REPLs that I never bothered to learn—in Haskell or in Python it's too easy to just test things informally interactively. I think I may be seeing the light, though.

I won't go into any of the details as to why exactly I'm starting to like this unit testing business: they are precisely the same ones that are listed by any of the unit testing freaks proselytizing on the internet. And yet, alas, it seems that no amount of their missionary zeal can approximate the effect of actually trying it...

Posted

Browser marketing

This is from Google's "whatbrowser" page:

Try a new browser for your PC:

  • Safari
  • Internet Explorer
  • Firefox
  • Opera
  • Google Chrome
  • I was initially a little skeptical of Google's marketing efforts with Chrome. It all seems a little heavy-handedly educational.

    But apparently, many people really really have no clue what a browser even is. (http://www.youtube.com/watch?v=o4MwTvtyrUQ)

    Therefore, it makes sense for Google to focus their marketing on simply educating people about the fact that they use something called a browser, what it is, and that they might want a different one.

    And in fact, they have taken this to the extreme where they don't even go out of their way to specifically recommend Chrome. (cf. http://www.youtube.com/watch?v=BrXPcaRlBqo)

    Which is kinda neat.

    Posted

    Why today's internet is just the beginning

    I attended a talk by Peter Thiel at Stanford this week. He had a fairly long shpiel about the stagnation of US real wages. While income statistics are always controversial, he has a point. 

    I would not attempt to cram a full theory of productivity slowdown into a single blog post even if I had one, but I do have a clue. This clue is related to a result in computer architecture known as Amdahl's law, which describes the effect of speeding up (usually parallellizing) part of a computation. What it comes down to is that if some part of the computation cannot be sped up—even if it is just a small part—then the more you speed up all the rest of the computation, the more the pesky little part that cannot be sped up begins to dominate the efficiency of the entire program.

    A real world example of Amdahl's law involves picking and pitting cherries. If you have one person picking the cherries and pitting them, then picking might take 10 hours and pitting another 10 hours. 

    Now suppose 9 of your friends come to help you, but you only have one cherry pitter. Can you get the job done in one tenth of the time? Nope! You can sure pick all the cherries in 1/10th of the time, that is, one hour, but it still takes 10 hours to do the pitting, because you only have one pitter. 

    And the more people you add to the process, the less the proportional effect is on how long the whole process takes.

    Now let's get back to the economy. Before the industrial revolution, many people worked in agriculture. One argument about why it took so long for an industrial revolution to really take hold, despite lots of false starts in various stages of world history, is that initially there was no improvement in agriculture, meaning that no matter how much more efficient manufacturing got, the fact that the vast majority of the population had to work in agriculture to keep everyone fed implied that overall productivity wasn't increased by all that much. I don't know if the argument is true—the origins of the industrial revolution are among the most contentious topics in economic history—but it surely makes logical sense.

    Similarly, I think we are now in a stage of economic development where we have gotten so good at certain things—manufacturing affordable and reliable automobiles, for instance, or furniture—that in order to push productivity much further we are going to have to innovate now in some of the domains that have so far been largely immune to productivity increases, and that therefore have been taking up ever growing fractions of the economy.

    I refer, of course, to transaction facilitation. The US, like many other Western countries, has become a nation of brokers, accountants, agents, lawyers, buyers, sellers, traders, analysts, recruiters, payroll clerks, and so on. All of these people are involved not in producing physical gadgets, but in coordinating human action. They facilitate transactions and the flow of information and trust.

    In order to make a real impact on overall productivity, technology is going to have to focus on facilitating transactions. 

    And that is why I believe that we have only just seen the beginning of the economic impact of computer and communications technology.

    But that's a story for another day.

    Posted

    Authors@Google: Daniel Solin

    I had a post recently about an article I read about a man whose sensible, if elementary, investment advice was being taken seriously only because he had a brain tumor. I wondered out loud just how ground-breaking the advice could be, since it boiled down to what I would think of as conventional wisdom: "buy and hold a little of everything; more stocks if you're young, more bonds if you're old."

    This guy, who presumably doesn't have a brain tumor, preaches much the same, but he adds an interesting perspective: he has some numbers and anecdotes about the popularity and phoniness of popular and phony investment advice. He takes a particular jab at one TV stock picker who is right *less* than half the time...

    (You may want to skip the lengthy introduction where he deferentially and loquaciously explains to the audience that they are much more sophisticated than the target audience of his book.)

    Posted

    Maxymiser to Optimizely: A/B testing’s a thing of the past | VentureBeat

    But Optimizely competitor Maxymiser says A/B testing alone is outdated. Maxymiser, whose clients include Sony, HarperCollins, Hewlett-Packard, Lufthansa and Virgin Mobile, says despite a recent $1.2 million all-angel round of funding for its competitor, the future for “smart websites” now lies in “multivariate” testing — and it’s likely to stay that way.

    It is heartening to know that statistical techniques for UI testing are moving from the 18th century into the 20th...

    Posted

    A Dying Banker’s Last Financial Instructions - Your Money - NYTimes.com

    This is one of the true benefits of having a brain tumor,” Mr. Murray said, laughing. “Everyone wants to hear what you have to say.

    So what is it, you may ask, that Mr. Murray has to say? It is surprisingly unsurprising...

    My summary: Knowing things is not enough to beat the market—you have to know something that nobody else knows. Most likely you don't. (If you read it in the Journal, you're late.) So instead, spread out your risk over various sorts of assets (internationally, too!), avoid high-overhead mutual funds, and adjust the ratio of stocks vs bonds to match your risk preference.

    I don't know about professional investment advisors, but every economics professor could've told you this.

    But most of them don't get brain tumors.

    Posted

    From the Modula 3 language definition...

    C++, has enriched C by adding objects; but it has also given up C's best virtue---simplicity---without relieving C's worst drawback---its low-level programming model.

    Posted

    LINK: Early Quora Design Notes

    A lot of thought went into the Quora product design and even at this early stage many details have been revisited multiple times. So, I thought I'd share a few of the decisions and principles that went into the first major version of the beta product.

    Rebekah Cox on Quora design.

    Posted

    Facebook Engineering Tech Talks: Front End Tech Talk

    Facebook has a bunch of engineering "tech talks" online. For some reason, they've turned off embedding, so you'll have to go to the facebook site to watch this one, but it's interesting. Basically, this guy managed to remove a large fraction of the Javascript on various pages by writing one piece of generic Javascript that interprets a domain-specific language implemented as extra attributes in the HTML.

    http://www.facebook.com/video/video.php?v=596368660334#

    Posted