Digest What You Read: Walkthrough of my process for understanding Rationality by Eliezer Yudkowsky

A slow, deliberate “rumination” process—turning raw reading into visual models, then into permanent notes—is presented as the antidote to information overload and today’s speed-and-productivity obsession. The core claim is that real understanding requires time for ideas to settle, be reprocessed from new angles, and be stored in a way that can be retrieved and reused later. The cow’s digestive system becomes the guiding metaphor: grass is pre-processed, regurgitated, chewed again, and only then fully digested. Likewise, reading should be followed by structured re-chewing—concept maps, chapter-level visuals, summaries, and linked notes—so knowledge becomes part of one’s durable mental system.

That workflow is framed as both a cognitive necessity and a practical response to modern “eating disorders” of information. The argument targets the productivity fetish: the belief that speed-reading, efficiency, and doing more tasks quickly automatically create value. Instead, the process is described as intentionally long and complicated because digestion—biological or informational—cannot be rushed. The workshop’s six-week structure is positioned as a guided method for this: read a difficult book, process it into a “book on a page,” create permanent notes, and present the material. The emphasis lands on rumination as the heart of the method: time plus repeated transformation of the same material.

The walkthrough then uses Eliezer Yudkowsky’s Rationality as the test case. Early confusion during chapter one is met with a concept map that distinguishes epistemic rationality (building accurate maps of reality by improving beliefs) from instrumental rationality (systematically achieving values by choosing actions that lead to desired outcomes). After importing highlights into Obsidian, the workflow shifts into progressive summarization (Thiago Forte’s approach) and chapter-by-chapter “atomic visuals”—simple diagrams meant to reframe each chapter’s ideas. Because the book is hard, external assistance is added carefully: text is copied from LessWrong, then ChatGPT is asked to produce two-sentence summaries per chapter, recommend simple SVG icons, and generate image prompts for Stable Diffusion via Midjourney. The resulting images are standardized with consistent style settings and later embedded into an organized “map of the book” that lays out the logical flow across roughly 45 chapters.

The “book on a page” and permanent notes focus on the book’s major themes: biases (like availability bias, conjunction fallacy, planning fallacy, and illusion of transparency), fake beliefs (including anticipation/proof problems, applause-seeking language, and “belief as attire”), and the mechanics of evidence. Evidence is treated as cause-and-effect-linked to what’s being inferred, with categories such as rational, historical, legal, and scientific evidence. The method also stresses experimental discipline—writing predictions before testing to reduce confirmation bias—and warns against misuse of Occam’s razor and “scientific-sounding” language without understanding. Blockers to rationality include curiosity stoppers (mysterious words, emergence-like jargon), semantic stop signs (terms that end questioning), and “guessing the teacher’s password” rather than understanding.

The closing takeaway ties the whole system back to the workshop’s purpose: build a digestible, retrievable knowledge structure through time, visuals, and reprocessing—so understanding survives beyond the initial reading and can be regenerated when needed.