Home | Archive

how to explain: an explainer

🌱 tags: inkhaven pck updated: ,

I have been spending the past four days trying to figure out how to explain things to people in text form. You will see live migrations is my scratchpad for the same efforts. With all my best intentions, it is rapidly devolving into a narrativized, purple mess. It neither clearly explains things to people without context, nor uses a legible shorthand or translation layer that people with some context will enjoy, nor entertains a reader who has all the context. It's a soggy mess of dessert salad that once aspired to be a trifle.

This shouldn't be this hard – and yet, it is. I've read other people writing in this mode for basically my entire internet life. I am flabbergasted at how inaccessibe I find it. I'm used to writing just fucking working but this just fucking doesn't.

Many possible things could be going wrong here. Here is a list of them:

Daystar has suggested a recipe, and Scott Aaronson has suggested that you get good at this by doing it a lot. I'm therefore favoring that last explanation.

So! Let's assemble some explainer recipes here, and think through them, and study them in place. Nothing like a good bootstrappy fixpoint to get one started.

layer cake

There is a sentence you want to say, so just say it. Then you'll have to explain what it means, so just write the explanations that suggest themselves. Each such explanation is a layer of the cake.

– Daystar

So whenever I'm writing a slide deck for a talk, for each slide I ask myself, I imagine someone were to raise their hand and say, "I don't get this", what would I say? —and I insert that explanation as a slide. I repeat this until the thing I would tell that guy is, "I don't know what to tell you, man, it's a picture of a blueberry."

– Valentino, Slime Mold Time Mold

well, why didn't you just say that?

Scott Aaronson also suggests to begin with you thesis statement, but in terms that anyone within your intended audience would understand. He also suggests either preceding or following the thesis with why they're supposed to care. Over the course of the conversation, the question that seemed to show me the shortcut and tell me what came next was the above: "why didn't you just say the sentence you just told me?" –or, forward looking, "why are you telling me this bit? what is this for?"

I kept answering him, either: "I need all this machinery for this problem", or "I didn't know that was the exact sentence you needed to hear."

So now, my takeaway algorithm is:

  1. You who you're talking to.
  2. write down what they care about. speak to that.
  3. write down what information you think they need.
  4. editing pass: annotate the whole thing, for every claim and every term and every concept, write down whether you think your audience knows it.
  5. explain everything they don't know in the appropriate place.

I think this is recursive: for each new thing to explain, you have to run the whole thing again, smaller. I also think it takes a lot of taste and art and experience to do this well. Well, that tracks. But if it's learnable, it's probably learnable in many passes on small-to-medium things.

nested puzzles

This is my favorite kind of explainer: Lisp in Lisp, Learn You a Haskell, The Agda book. You're given a simple problem to solve. The solution gives you part of what you need to solve the next puzzle. You've turned your explainer into a videogame. It's beautiful, when executed well it's unforgettable.

How do you write these? Worth a close read of one of these to see if I can see how they wrote them. (Do you ever do that? See through a piece to see how someone must have written it? I do this all the time, it's how I learned to write.)

PCK, or the Learning How to Learn playbook

What do I teach you that lets you generate or learn the rest of it? Which bit is the real summary?

Related questions, at least for me: What's the shitposty version I'd tell you if we were both drunk at a party?

Somehow, these always end up being a question. But sometimes they're a formula. The clearest exemplar memory I have is a teacher telling me, E = 1/2mv2 contains the other five kinematics equations, and deriving them all in the scratchspace of an exam answer book instead of struggling, in my panic, to remember the one I needed.

So maybe, rather than the durnken question, we need the strong version of this: the meditative question.

the recipe

or, do as I do, and see what happens.

This is how you teach procedural knowledge. Sometimes, there is the nicest, shortest, cleanest thing you can do. This explainer is perhaps a sufficient handle for you to observe, infer, or intuit the rest of your way to my thoughts and feelings about this.