This past month I’ve been experimenting with video posts, as you’ve probably noticed (unless you read all these posts on Less Wrong, in which case, now you know). It started because I didn’t think I had time to make a post, but then on a whim decided to record a video. People responded well, so I’ve tried it two more times.
I like that videos do a better job of netting me new subscribers, which is important because this blog ultimately exists for the purposes of promoting the book. Otherwise I’d blog only on Less Wrong, write marginally fewer posts, and not worry about having a subscriber list.
I don’t like that the written transcripts of the videos don’t read as well as written posts would. Or at least that’s what I think. They contain a lot more fluff, which is more tolerable when speaking, but less so in writing. Or at least it is to me.
I’m not yet sure what to do about it, but I’m going to continue to experiment a little longer. I’ll likely settle somewhere that includes a mix of written posts and video posts once I figure out what kinds of posts work best which way.
Speaking of writing, I spent August wrestling with Chapter 4. Turns out this one had a lot of stuff in it I left behind for myself to fix later in editing, I just didn’t remember that. So I spent most of the month trying to get the opening section right. I’m finally there, so now I’m on to the other big task of editing this chapter, which is fixing the last section by addressing the Löbian Obstacle (thanks to Abram Demski for many useful comments two years ago that I’m just now finally addressing!)
All this editing work has made a few things clear to me about how to fix writing faster. My current methods:
If some text doesn’t work, the problem is usually not with the text itself, but earlier text that forced this text to take the shape it’s taking. The solution is usually to go up the page and change whatever is making the problem text exists so that some different text can take its place.
Figuring out that I need to change up-page text usually takes a few passes, since it’s not always immediately obvious that what’s off about a sentence or paragraph is that it’s been written into a corner by some previous statement rather than normal draft writing that needs to be cleaned up.
When I get stuck on how to move forward, LLMs are a good source of creativity. What I usually do is ask Claude to generate several possible completions for a snippet of text. It never gives me back anything I end up using directly, but it gets me thinking about ideas I hadn’t considered talking about, which usually leads to figuring out what to do.
If I’m really stuck, Claude and I will have some back and forth. I turn on my annoyed, short-on-time manager mode to press it by simply rejecting the results and saying what I don’t like. After a couple rounds, this is usually enough feedback to get it to give me something useful.
In other writing outside the blog, I did a dialog with fellow Less Wrong user SpectrumDT about ’s hypertext book Meaningness. In the end, SpectrumDT says I didn’t end up explaining anything better, but he enjoyed the exchange. You can read it here.
Paul Graham discusses that good thinking requires good writing and vice versa.
You have written a lot, but maybe what you notice in your video transcripts is that one of the effects of writing is missing. I don’t think it has to be literal writing. Generalizing Paul Graham, I think that for clear thinking you need to put ideas in a form that forces precision and makes them grow into more. The grow into more seems to be clearly the case with the engagement and the posts here. But video often doesn’t have the precision—maybe that’s what shows in the fluff?