Teaching to Learn: The Feynman Technique, Run With AI
You don't understand it until you can explain it simply
There's a specific, familiar feeling: you read something, it makes sense, you nod along — and then someone asks you to explain it and you can't. You reach for the words and find you were only recognising the material, not understanding it. Recognition ("yes, that looks right") is easy and feels like knowledge; the ability to reconstruct and explain something is real knowledge, and they're very different things.
The Feynman technique exploits the gap. It's built on one move: try to teach the thing, simply, and watch where you fail. The moments you stumble, reach for jargon you can't unpack, or hand-wave — those are the exact spots you don't actually understand, made visible. Teaching is the most demanding possible test of understanding, which is why "the best way to learn something is to teach it" is literally true, not a motivational poster. Studies find that even expecting to teach material makes people learn it better, and explaining it out loud beats re-studying.
The old obstacle was needing a patient audience. That obstacle is gone: AI is an infinitely patient student that asks "why?" forever — so you can run the technique any time, on anything.
What you'll have at the end
- The four-step Feynman loop
- Three ways to run it with AI as the student or the examiner
- One concept you now genuinely understand, with its gaps closed
The four steps
1. Pick a concept and write its name
One specific thing you're trying to learn — "the electron transport chain", "how a for-loop works", "why bond prices fall when interest rates rise", "the causes of the French Revolution."
2. Explain it in plain language, as if to a beginner
Write or say an explanation as though to a smart twelve-year-old who knows none of the jargon. This is the whole engine, and it has one strict rule: no undefined jargon. Every technical term either gets unpacked in plain words or gets cut. "It works via the electron transport chain" is hiding — what is that, in normal words? The constraint of simplicity is what forces real understanding to the surface, because you cannot fake-simple something you don't grasp.
Talking to an inanimate object works too (the classic "rubber duck") — the point is producing the explanation out loud, from memory.
3. Find the gaps — where you stumbled or hid
Now read your explanation honestly and mark:
- Where you stalled or couldn't continue.
- Where you fell back on jargon because you couldn't say it plainly.
- Where you went vague or hand-wavy ("and then it sort of… happens").
- Where you're not sure it's actually true.
Those marks are a map of exactly what you don't understand — which is gold, because most study wastes time re-covering what you already know while these specific holes stay open.
4. Go back to the source, then re-explain simpler
Return to your notes, the textbook, the source — but only for the gaps you found — relearn those, and then explain the whole thing again, more simply. Loop steps 2–4 until you can give a clean, jargon-free explanation start to finish with no stumbles. That clean version is proof of understanding, and it also makes a perfect revision note (feed it into LN-02).
Running it with AI
AI turns the technique from a solo exercise into an interactive one. Three modes (see also AI-06):
- AI as the confused student. Explain the concept to it and instruct it: "You're a curious beginner. After my explanation, ask me the three 'but why?' questions a smart 12-year-old would ask where I was vaguest." Its questions land precisely on your gaps — the ones you glided over.
- AI as the examiner. Paste your plain-language explanation and ask: "Where is this vague, wrong, or hiding behind jargon? What would a beginner still not understand?" It marks step 3 for you, ruthlessly, which is hard to do to your own writing.
- AI as the source, carefully. For closing gaps, you can ask it to explain the specific bit you missed — but verify it (AI-07), because a confident wrong explanation will teach you a confident wrong thing. Cross-check anything load-bearing against a real source.
The AI doesn't replace the work — you still have to produce the explanation, which is where the learning is. It replaces the patient audience and the honest examiner.
Why this works
1. Producing beats recognising. Explaining forces retrieval and reconstruction (LN-01) — the effortful production that builds memory — where re-reading only gives you recognition, which doesn't.
2. Simplicity is an honesty test. You can nod along to a complex explanation you don't understand; you cannot generate a simple one for something you don't grasp. The no-jargon rule makes the gaps undeniable.
3. It targets your actual holes. Step 3 tells you exactly what to restudy, so step 4 is efficient — you fix the specific gaps instead of re-reading everything and hoping.
What this means for you
- Recognition isn't understanding. If you can't explain it simply, you don't yet know it.
- Teach it simply and watch where you fail — the stumbles and jargon are a precise map of your gaps.
- No undefined jargon is the rule that forces real understanding out.
- Relearn only the gaps, then re-explain — loop until it's clean and stumble-free.
- Run it with AI as the curious student, the ruthless examiner, or (with verification, AI-07) the source. You still do the explaining — that's the point.
Exercise (35 min, verifiable output)
- Pick one concept you're studying and think you understand.
- Explain it in plain language — written or spoken aloud, from memory, no undefined jargon.
- Have AI mark the gaps: paste your explanation and ask it where you're vague, wrong, or hiding behind jargon; or explain to it and have it ask the "but why?" questions. List the gaps it finds.
- Relearn one gap from a real source (verify anything AI told you, AI-07).
- Re-explain that part, simpler. Save the clean explanation into your study system (LN-02).
✅ Finish check: a plain-language explanation of one concept, a listed set of gaps it exposed, and one gap re-learned and re-explained clearly — proof you moved from recognising to understanding.
Summary card
- If you can't explain it simply, you don't understand it — recognition ≠ understanding.
- The loop: name it → explain in plain language (no undefined jargon) → find where you stumble/hide → relearn only those gaps → re-explain simpler. Repeat until clean.
- Simplicity is an honesty test; the stumbles are a precise map of what to restudy.
- Producing beats recognising — explaining is retrieval, which builds memory (LN-01).
- Run it with AI as curious student / ruthless examiner / (verified) source — you still do the explaining.
Sources
- Feynman, R. — via Gleick, J., Genius, 1992
- Nestojko, J. et al. — Expecting to Teach Enhances Learning, 2014
- Fiorella, L. & Mayer, R. — Learning by Teaching, 2013
Next lesson: AI-06 — Studying With AI: Feynman, Mock Exams, Error Hunting (L2) Related: LN-01 Active Recall and Spaced Repetition · LN-02 Building Your Own Study System · LN-06 Reading Technical Material · AI-07 Hallucination Hunting Path: related — the strongest test of understanding