The Feynman Technique: You Only Know It If You Can Teach It

The short answer: if you can't explain it, you don't know it yet

Richard Feynman is supposed to have said: "If you can't explain something in simple terms, you don't understand it." The attribution is contested — like much Feynman-folklore — but the method that bears his name maps onto something the learning-science literature treats as ground truth: retrieval beats re-exposure. Karpicke and Roediger's 2008 study in Science (source) is the canonical evidence. Explaining out loud is one of the most demanding retrieval tasks you can perform on yourself — and so it produces some of the highest-quality long-term learning available. The technique is just a structured way of forcing that retrieval to happen.

The four steps of the Feynman technique

The standard version is short enough to fit on an index card. Pick the concept. Explain it out loud, in plain language, to someone (real or imagined) who doesn't share your background. Notice exactly where you stumble, hedge, or fall back on jargon — those are the holes. Go back to the source material, fill the hole, and try the explanation again. The loop ends when you can explain the thing without notes, without jargon, and without those stuttering moments where you're hoping the listener doesn't ask the follow-up.

The whole leverage is in step two. Re-reading hides the holes; explaining exposes them. The discomfort of stumbling is the signal that learning is needed — the brain re-reading the source quietly tells you "oh, I knew that", and the next test reveals you didn't.

Lem as the safe student

The problem most readers hit is the absence of a willing listener. You can't grab a friend every time you want to retrieve last week's chapter, and explaining to yourself in your head feels different from explaining to someone — softer, with the option to skip past the difficult bit. An AI mentor — Lem, in moinaki's case — is the safe stand-in: a listener that won't be bored, won't judge, won't be impatient, and can be asked follow-up questions when you'd benefit from them. The judgmental dimension is what makes a real listener feel costly; remove that and the technique becomes practicable on a daily basis.

Practically: open a doc, write your explanation as if you were teaching Lem, then ask Lem to summarise back what you wrote. Where Lem's summary deviates from what you meant, you've found the hole. The deviation is signal, not failure of the mentor.

Why this pays double for ADHD

Two ADHD-specific forces help here. First, retrieval at the level of full out-loud explanation is more engaging than silent re-reading — there's novelty in trying to phrase it, there's voice, there's a fictional audience. The attention system gets what it needs to stay in the seat. Second, the technique exposes the gap between feeling you know something and actually knowing it. ADHD readers often have the "I get it in the moment" experience repeatedly with material they then can't recall a week later. Explaining out loud collapses the distance between feeling fluent and being fluent — the second test is also the first practice.

Five concrete ways to run the technique this week

• Pick concepts that fit in 90 seconds. Not a chapter; a definition, a mechanism, a single equation. The cognitive load of explaining grows non-linearly with scope. Three 90-second explanations beat one heroic 10-minute one.

• Use a voice memo. Recording yourself is roughly as useful as explaining to an actual audience because the implicit "someone will hear this" pressure produces a similar quality of retrieval. Don't review the recording; it served its purpose during the recording.

• Write the explanation to a thirteen-year-old. Not a four-year-old (the over-simplification trope loses precision); not a peer (you'll lapse into jargon). A bright thirteen-year-old needs you to be clear and accurate at the same time — which is exactly the standard the technique is trying to meet.

• Mark the moment of stumble in real time. Each time you hedge, jargon-up, or skip — make a small mark on paper. After the explanation, those marks are your study plan. The list of stumbles is more useful than the explanation itself.

• Run a second pass twenty minutes later. Don't immediately re-explain — the working memory of the first pass will inflate your perceived competence. A short break, then a second pass, then check whether the stumbles moved. Movement is learning; same stumbles in the same places is the signal to actually re-study the source.

Where it fails (and the repair)

• Explaining with the notes in front of you. If you can see the source material, you're paraphrasing, not retrieving. Move the source out of sight before the explanation begins, even if the explanation is then weaker. Weaker explanation under blind conditions is better data than strong explanation with peek conditions.

• Mistaking fluency for understanding. It is possible to give a smooth-sounding explanation without genuinely understanding the underlying mechanism. The test is whether you can answer two layers of follow-up: "why is that true?" and "what would happen if this part changed?". Smooth at layer one and broken at layer two is the diagnostic for fluency-as-illusion.

• Treating the technique as a final exam. It works best as repeated cheap practice, not as a once-per-chapter audit. The accumulating recall is what wires the learning in — single dramatic attempts are less efficient than three small attempts over three days.

FAQ

Did Feynman actually invent this technique?

Not literally as a named four-step method. The technique was popularised by educators and writers who attached his name to it, and the spirit of "if you can't explain it simply you don't understand it" matches things Feynman said in interviews and in his autobiographical writing. The mechanism is real and well-supported by Karpicke's retrieval research; the branding to Feynman is partly homage.

Can I explain to ChatGPT or Claude instead of Lem?

Yes. Any AI that can paraphrase what you wrote back to you serves the function. The advantage of Lem in the moinaki context is that it sits inside the learning loop with your progress already in view; the technique itself isn't picky about which AI you use to be the stand-in listener.

What if I genuinely have no one to explain to and feel silly talking to myself?

Writing the explanation out as if to an absent reader works almost as well as speaking. The aural component is a slight bonus, not the load-bearing piece. The load-bearing piece is the retrieval — and writing forces the same retrieval as speaking, with the side benefit that the written explanation can be read back to find the stumbles.

How is this different from teaching?

Teaching shares the same mechanism — the "teach to learn" advice is well-grounded. The Feynman technique is a private, low-stakes version of teaching that you can run alone. If you can find a real student to teach (a classmate, a friend, your future self via writing), you get the same benefit. The technique is the structured version when no live student is available.

How often should I do this for a concept?

Pair it with spaced repetition. Day one: explain. Day three: explain again, watch where the stumbles moved. Day seven: again. Three retrievals across a week is more powerful than one dramatic single-pass review on day six. The Feynman technique and the spacing effect are most powerful together.