Learn a Difficult Topic From First Principles

A REUSABLE METHOD FOR LEARNING FROM FIRST PRINCIPLES Treat it as an iterative loop, not a linear path. Each pass goes deeper. Stop when you can do all six phases without external reference. PHASE 0 — DEFINE THE CORE PROBLEM 1. Write one sentence: "This topic exists to solve ______." If you can't, you haven't found the problem yet. Search for it — read the origin story, the paper, the first use case. 2. State the constraints the solution must respect (performance, correctness, cost, scale, safety). These constraints are WHY the topic looks the way it does. 3. Write the naive solution you'd invent yourself before learning the real one. This gives you an anchor: every new concept is now "a fix for a specific flaw in my naive version." Rule: never memorize a mechanism before you can say what failure it prevents. PHASE 1 — BUILD VOCABULARY DELIBERATELY 1. Collect the term list: every bolded term, acronym, and symbol from one authoritative source (textbook chapter, official docs, survey paper). 2. For each term, write a definition in YOUR words, plus the problem it solves, plus one concrete instance. Not a dictionary copy. 3. Group terms into a dependency map: term A must be understood before B. Learn in dependency order, not document order. 4. Keep a running glossary file. Every time a term appears that you can't define instantly, add it. Rule: you don't know a term until you can use it correctly in a sentence about the problem. PHASE 2 — EXAMPLES AND COUNTEREXAMPLES 1. For every concept, collect at least three examples: the canonical one (from the source), one you construct yourself, and one from a different domain that still fits. 2. Collect at least one counterexample per concept — the case where the concept does NOT apply, or the naive version breaks. Counterexamples are where real understanding lives; they expose the boundary of the definition. 3. Trace every worked example by hand, line by line, without looking. Then check. Where you diverge from the source is your ignorance map. 4. Find the edge cases: empty input, zero, extremes, degenerate values. Ask "what happens when X is 0/infinite/negative/missing?" Rule: if you can't produce a counterexample, you likely have the definition memorized, not understood. PHASE 3 — TEST UNDERSTANDING ACTIVELY 1. Closed-book recall: after studying, write everything you know about the topic on a blank page. This is retrieval practice — it's what actually builds memory, not rereading. 2. Explain it to a novice (or a rubber duck, or an empty doc) in plain language. Any point where you hand-wave or say "it just works" is a gap. 3. Reconstruct, don't recognize: close the book and re-derive the key result. Recognition ("I've seen this") is weak; reconstruction is strong. 4. Apply it to a novel problem you've never seen. If you can only solve problems identical to the examples, you don't have transferable understanding yet. 5. Use the Feynman test: if you can't explain it simply, you don't understand it well enough. Find the specific sentence where your explanation breaks and fix exactly that. PHASE 4 — AVOID MISTAKING FLUENCY FOR KNOWLEDGE This is the critical failure mode. Fluency = the feeling of familiarity from exposure. Knowledge = the ability to produce and apply. They feel identical while reading; they separate only under testing. 1. Beware the illusion of competence: rereading, highlighting, and watching videos all feel productive and all create fluency without knowledge. They are the lowest-value activities. Replace them with recall and production. 2. Test in the hard direction: knowledge flows from you to the page (writing, solving, explaining), not from the page to you (reading). If you can't produce it, you don't know it, regardless of how familiar it feels. 3. Add spacing: revisit after 1 day, 3 days, 1 week, 1 month. Fluency fades fast; knowledge that survives spacing is real. 4. Add interleaving: mix this topic with adjacent topics and older material. Blocked practice (drilling one thing) inflates apparent mastery. 5. Set a "can I do it cold" bar: at random intervals, without warning, produce the explanation or solve a problem from scratch. If you need a warm-up or a hint, it's fluency, not knowledge. PHASE 5 — CLOSE THE LOOP 1. After each test, update the glossary, fix the naive solution, and add the counterexamples you missed. 2. Re-derive the core problem: has your one-sentence definition changed? If yes, your mental model deepened — good. If it's identical after deep study, you probably never left the surface. 3. When you can do all of the above from memory, teach it to someone else for real. Teaching is the final exam. THE COMPACT VERSION (print this) 1. What problem does it solve, and what's my naive version? 2. Can I define every term in my own words, in dependency order? 3. Three examples, one counterexample, all traced by hand? 4. Can I produce it from a blank page, explain it simply, and apply it to a novel case? 5. Am I confident because I tested myself, or because it felt familiar? (If the latter, retest cold.) The whole method reduces to one principle: knowledge is what you can produce, not what you recognize. Design every step to force production, and schedule the production to outlast the fluency.

The whole method reduces to one principle: knowledge is what you can produce, not what you recognize. Design every step to force production, and schedule that production to outlast the fluency. If a topic ever feels "easy," that's the warning sign to test yourself cold — the feeling of ease while reading is exactly the illusion the method exists to defeat.