*Published September 2026* 1. Accumulate some reading material of interest. I prefer to consider questions rather than academic subjects, see [[Open questions]] for some examples. I use [Reader by Readwise](https://readwise.io/read) for articles, papers, and newsletters, and [Kindle](https://en.wikipedia.org/wiki/Fire_making) for books. The simplest workflow I've found is to use the [Readwise Highlighter extension](https://chromewebstore.google.com/detail/readwise-highlighter/jjhefcfhmnkfeepcpnilbbkaadhngkbi) and the [Zotero extension for citation management](https://chromewebstore.google.com/detail/zotero-connector/ekhagklcjbdpajgpjgmbionohlpdbjgc?hl=en) 2. While reading, I highlight insightful passages. Ideally these highlights make sense independent of the surrounding context. Academic papers are frustrating for this very reason, they are seemingly designed to make such extraction impossible, relying on dense prose that adheres to [Parkinson's Law](https://en.wikipedia.org/wiki/Parkinson%27s_Law) but for page limits.^[For an alternative model of the academic paper, check out the excellent [Things could be better](https://www.experimental-history.com/p/things-could-be-better)] For papers, I write a short summary, inspired by the [*five C's*](https://www.lib.sfu.ca/system/files/32376/paper-reading.pdf). If the work is highly technical, I may still highlight relevant equations + definitions. 3. I consider these highlights and notes to be [hard-earned and immensely valuable](https://en.wikipedia.org/wiki/Desirable_difficulty). I use them in a few ways: 1. Spaced Repetition: Readwise offers a daily review for highlights. Additionally, there's a [Mastery feature](https://docs.readwise.io/readwise/guides/mastery) that assists in creating flashcards. 2. Search: The [Readwise MCP](https://docs.readwise.io/tools/mcp) allows you to connect your repository to AI apps. I have it installed on all platforms (ChatGPT, Claude, Codex, etc.), and include a custom instruction to prefer retrieval from my highlights when applicable. I can then issue requests such as: "Pull highlights relating to management" or "Retrieve paper summaries relevant to Human-AI Interaction." 3. Reread: Occasionally, I will reread a source but just go through all of my highlights. This is much faster, on the order of tens of minutes. That's about it. This process is geared towards *discovery* and *recall*. I've found that I don't need to explicitly spend time on application or drawing connections, rather that part comes naturally. Note that this isn't really intended for procedural knowledge (probably better to learn by doing) or language acquisition (I have little insight to offer on that front).