The Feynman Technique is a learning method that uses teaching as a mechanism for identifying gaps in understanding. It consists of four steps: (1) choose a concept, (2) teach it to a novice in plain language, (3) identify where your explanation breaks down, and (4) return to the source material to fill those gaps, then simplify the explanation further. The cycle repeats until the explanation holds together from start to finish.
Choose a single, bounded concept. Not "machine learning." Not "quantum computing." Something you can hold in your head: "how gradient descent works," "what a transformer attention mechanism does," "why the Fed raises interest rates to fight inflation." The more specific the concept, the more useful the exercise.
AI role at this stage: if you are not sure how to bound a concept, ask the AI to break a broad topic into teachable sub-concepts. "I want to learn about retrieval-augmented generation. Break it into 5 sub-concepts I can explain one at a time." This structures the learning before you begin.
Write or speak an explanation as if your audience has no prior knowledge of the field. No acronyms without definition. No "as we all know." No skipping steps because they feel obvious. The test is not whether the explanation is elegant. The test is whether someone who has never encountered the concept could follow it from beginning to end.
The breakdown will happen at predictable places: the transition between two ideas where you used the word "therefore" without justifying the connection. The analogy that felt clarifying in your head but reads as hand-waving on the page. The technical term you used without defining. Mark every one of these. They are the map of what you do not yet understand.
AI role: after the AI identifies gaps, do not let it fill them. Return to the source material yourself. Re-read the relevant section. Reconstruct the explanation in your own words. Then test it again. The AI is the diagnostic tool, not the cure.
Rewrite the explanation using simpler language and shorter sentences. Remove every word that is not doing work. Replace jargon with plain English. If a concept required three sentences, try to explain it in one. If the one-sentence version loses accuracy, you have identified the minimum complexity the concept requires. That threshold is itself a form of understanding.
Repeat the cycle until the explanation holds together from beginning to end. Some concepts take one pass. Some take five. The number of cycles is not a measure of intelligence. It is a measure of how much new ground you are covering.
The risk is that the AI becomes a crutch. When the explanation breaks down and the AI immediately provides the correct answer, the learning opportunity is lost. The rule is simple: the AI asks. You answer. Only after you have attempted your best answer do you check against the source material or ask the AI to verify. The struggle is the learning. The AI's job is to make sure the struggle happens at the right points.
Q: How is the Feynman Technique different from active recall?
A: They are complementary mechanisms. Active recall tests whether you can retrieve information from memory. The Feynman Technique tests whether you can organize that information into a coherent explanation. You can recall every fact about a concept and still fail to explain it clearly. The Feynman Technique catches that failure. The two techniques work best together: use active recall to surface what you know, use the Feynman Technique to structure it for understanding.
Q: What if I can't simplify a concept without losing accuracy?
A: That is the point. The loss of accuracy tells you something real: either the concept genuinely requires a certain level of complexity, or your understanding is not deep enough to find the simpler explanation. In the first case, you have learned a boundary. In the second, you have found a gap. Both are valuable.