For some people, there’s nothing more demotivating than the idea of the universe’s heat death. Imagining the sun fizzling to ash and endless galaxies swallowed by quiescent gray static just flattens them and steals away their will to do anything at all. Thinking about how little I remember of what I read has the same effect on me. When I look back at my Goodreads account, I always find books I’ve read and can now tell you almost nothing about. It’s enough to make me wonder why I bother.
Because of this anxiety (and, really, I don’t know what other word to use for it), I’ve taken the idea of managing knowledge quite seriously. I’ve iterated through a number of systems but have stuck with my current system for about a year now. This is long enough to make me feel like I’ve really integrated this system as a habit and long enough to be comfortable sharing it, on the off chance it helps you too.
A word about what I do
I work as a data scientist and researcher for a medical specialty board. Most researchers in similar positions to mine focus on psychometrics or how certification affects clinical practices. I’m in a pretty unusual position, though, because my employer has spent the last 7 years building up an enormous database of electronic medical records from primary care practices. This means that most of my work focuses on understanding the vast uncharacterized terrain of primary care in America, and supporting our academic collaborators as they build a more robust primary care research pipeline.
My job, then, is very similar to an academic research position. I read, design studies, code, write, and present. There aren’t many boundaries on what I can research, so it gives me an opportunity to bring together all sorts of ideas. This has meant that I need to not only catalog knowledge, but also to bring it back into circulation along with whatever’s in my head at the moment. Because of my own flaws and proclivities, I lean towards breadth over depth. So, caveat lector, this might not work for you.
The system
Almost everything I read comes either directly or indirectly from one of four sources: X, Substack, Hacker News, and various RSS feeds. I’ve set up all my RSS feeds to appear directly within Readwise Reader, which is a combination read-it-later app and RSS reader. I can send any article from my RSS feed to my Readwise Reader inbox for later, in-depth reading. Similarly, I can send any articles from Substack or Hacker News to Readwise Reader through the Readwise browser extension.
I’ve found that it’s helpful for me to separate the act of finding articles from the act of reading them. The two tasks demand different mindsets — explore versus exploit — and using Readwise Reader means that I always have a store of interesting articles ready for whenever I feel like focusing on reading over searching.
The inbox of Readwise Reader is essentially a giant list of unread articles. You can tag each article, but I usually only do that for articles that will contribute to a writing project (either here or for academic journals). These tags mean that I can find project-specific articles quickly. I export any articles that I’ll use in my academic writing to Zotero, which makes citation easy in Word.
The other way I engage with articles through Readwise Reader is by highlighting passages. If I read anything that I think I’ll want to remember longer term, I simply highlight it. Then, all the highlights from within Readwise Reader are automatically synced to Readwise, a spaced repetition program that also allows syncing with other data sources like tweets or Kindle highlights.
Every day, I review 15 highlights in Readwise. Even though Readwise gives me the option to convert highlights into flashcards and other formats, virtually all of my highlights are non-interactive. During a review, I have the option of keeping a highlight in the rotation or deleting it. I also have the option to favorite it, which will cause it to appear in my reviews more frequently, and to add tags. I find that tagging each passage’s topic is an important way of actively engaging with the highlights by thinking about the contexts where I might someday apply them.
In all, I’ve collected 3,787 highlights. At least once a week one of these highlights will unlock a new approach to some problem or idea. This is the key value of Readwise: having an ongoing dialogue between what’s on your mind now and the states of mind that are recoverable by reencountering what you read in the past.
Whether new ideas come to me by reading an article, encountering an older highlight in spaced repetition, or just mentally masticating my daily experiences, I add them all to Apple Reminders. This app syncs across my devices, from my laptop to my phone. Apple Reminders contains my academic article ideas, Substack article ideas, questions to research in greater depth, and my daily to-dos. For each of these, Reminders forces me to think of a concise formulation for everything I add to it, making it easy to review at a glance. Every few months, I’ll go through my article ideas and delete anything that, with a little distance, I realize is trash.
Once ideas start inching towards realization, I add them to Apple Notes. At first, these notes will just be a randomly ordered list of things I want to say about that idea, but will shift into an outline with links and sources as they come into focus. The main advantage is, again, that it syncs across devices. It’s also distraction-free, unlike web apps that are too close to easy amusement.
This process is worth little if it doesn’t support my output, which usually takes the form of writing. I send my writing to the same sources that I get them from. I post my Substack articles on X and occasionally on Hacker News. I tweet random musings. Academic articles that I publish appear in many of the same journals that I subscribe to via RSS, or they’re tweeted out by the journals. I see it as a virtuous circle where I contribute to the same sources that often fuel my original ideas.
Aspirations of polymathic cyborghood
As important as software is to my system of managing knowledge, my physical tools are also vital to its success. I mostly search for articles on my phone and read them on my iPad, although I do a little bit of both activities on both devices plus my laptop.
I reserve my most eccentric setup for more prolonged reading sessions, during which I like to lie down. I use a pair of XReal Air glasses to display my iPad screen as if it were on a giant display a comfortable distance away from me. The glasses were pretty affordable, they’re light, and the display is quite nice. On the other hand, they don’t have a battery and so connect directly to my iPad. This makes them light, but drains my iPad’s battery relatively quickly. The glasses’ arms also tend to press against my mastoid processes, which get achy if I’m not attentive to padding them or shifting the glasses around frequently.
The problem with the glasses connecting to the iPad is that there’s no obvious way to interact with the content on the screen. The glasses themselves don’t accept inputs, and the iPad is ideally off to the side while I’m reading. Thus, I got a handheld mouse — specifically the ELECOM Relacon Handheld Trackball Mouse. I don’t love it. It’s too small for my hands, and the scroll mechanism can be unpredictable. But it’s got Bluetooth and gets the job done. There just aren’t a ton of options for handheld mice.
Suffice it to say, there’s room for improvement with this setup, but it’s a first step towards a more absorptive kind of reading.
What hasn’t worked for me
Like I alluded to at the top of this article, I’ve tried a number of knowledge management systems that simply haven’t worked long-term for me. Two are worth noting because they failed in interesting and different ways: Notion and Roam Research.
Notion can be thought of as an interface for connected databases. You can set up a database of to-dos, with each item connected to a member of a project database, for example. There are some very shiny tools for project management in Notion, but there’s a kind of cold start problem: to benefit from Notion, you have to impose structure on it, but without using it, you don’t know what kind of structures you’ll need. I found that Notion tends to incentivize the creation of complex structures that are difficult to maintain. As a set of connected databases, it’s never adequate to just update one thing in Notion — there’s always another database to update. Maintaining it was unsustainable for me from almost the start.
Unlike Notion, Roam Research lets you build up structure gradually. The core idea is similar to the original imaginings of hypertext: you can turn any text passage into a page, which can then be used to link ideas across contexts. For example, a note about scurvy treatments and a recipe for key lime pie might both appear on a page called “citrus”. It’s so easy to create a page, though, that I found myself forgetting what I’d called them: “Did I call it ‘citrus’ or ‘citruses’?" I’d think to myself. Things got messy because I couldn’t keep up with similarly titled pages, and the hubs stopped working as intended. I stuck with Roam for a while, but its advantages diminished as I used it longer and the pages became less reliable.
From these experiences, I found that I need parsimonious, opinionated tools. They can’t be too capable, or I’ll want to push those capabilities. They also can’t be too open-ended, or my own faults will win out over the system’s strengths before long. The tools I’ve landed on manage to keep me focused by having a well-curated set of features and by not trying to solve every use case with a single app.
I like the prospect of the heat death of the universe and, before that, our destruction by the expansion of the sun. It puts things into perspective.
I had a long conversation with a friend on Friday about how the world will recover when humanity dies out. In The World Without Us — https://www.goodreads.com/en/book/show/248787 — animals will quickly take back their spaces, the forests will grow back and, eventually, a glacier will come down from the north and wipe the island of Manhattan clean.
Your job sounds very cool.
Thanks for this. I thinks it’s helpful to be reflective and intentional about this process in such a noisy environment. I guess my main sources are: manifold for current events, discord for ML papers and news, and then substack for everything else.
Overall, I try to focus on recall, rather than tools. But I often lose links so that’s clearly not perfect. I think forcing my brain to output something about the content rather than just storing a note really helps me to recall it. I have a file where I store links of things I want to remember and come back to and write a paragraph summary of it (one is a link to your nosology post). Then I also write short summaries of a couple pages or so of nonfiction books I read.
Most of my machine learning news comes from discords these days. Everything online has gotten so flooded in the last couple years that a lot of people just talk in private discords now. I wish I had some way of taking things from X and getting them emailed to me as a daily digest. That’s one of the main things I like about consuming stuff from substack. I find going on social media to be really frustrating, even when it’s technical commentary. I’m missing useful stuff by not being on X, but I find the cost to be too high.