The Feynman technique, explained to an AI
The Feynman technique says explain what you learned to someone else. It always needed a listener. Now an AI listens and points at the gaps. Here is what happened when I explained DNS.
The Feynman technique is simple. Write out what you learned as if explaining it to a twelve-year-old, find where you get stuck, and go back to learn that part. It is named after the physicist Richard Feynman, who studied that way. The problem was always the same: there is no twelve-year-old next to you, and even if there were, they could not point at the places where your explanation is wrong.
If the Socratic method is the asking side, the Feynman technique is the telling side. Both only work once you put things in your own sentences. Both had the same constraint, and for the same reason that constraint is gone.
The four steps
One, pick something you learned. Two, write it out for someone who does not know it, without jargon. Three, find where you get stuck, where you wave your hands, where you hide behind a technical term. That is where you do not actually know. Four, relearn only that part and rewrite it more simply.
Step three is the whole method. Where the explanation breaks is where understanding broke. You cannot see that spot while listening to a lecture or reading a book, because the feeling of understanding covers it.
The missing listener
It is the same constraint as having only one Socrates. Rarer than someone who will listen is someone who will then say "that does not add up". Family gets bored, colleagues are busy, experts are expensive.
So I knew the technique and did not use it. Writing alone and reading alone, I would walk past the gaps without noticing, because I cannot point at what I do not know.
What I explained to the AI
I actually did this while writing the post. The topic was something I learned moving my website: how a domain name finds my site. I wrote it for a twelve-year-old.
Every computer on the internet has a number address. People cannot remember numbers, so we use names, and there is a phone book that turns names into numbers. When someone types my site's name, the phone book gives them the number, and the computer at that number sends the page. If I move the site to another computer, I change the number in the phone book.
Then I asked Claude to point at the places a twelve-year-old would not follow, and the places where I was hand-waving. Three came back.
First, "there is a phone book" says nothing about where it lives or who keeps it. I did not know. I had been using the word name server without realising it is the phone book itself.
Second, "change the number", when in fact I had changed two kinds of record differently. One holds a number, the other holds another name. I left that out not to simplify but because I could not explain why there are two.
Third, nothing about the change taking time to spread. I am the person who changed the name servers and then wandered for a good while because the address with www and the one without went to two different places. That memory was missing from my explanation.
All three were places I thought I knew. Before writing it out I would have said I understood the whole thing.
Relearn, rewrite
I asked about the three spots. Two of the seven questions were enough: "why is that so" and "in my own words, is that right". Then I rewrote the explanation. One sentence about the phone book being copied in many places so changes take time to spread, one sentence about the two kinds of record. Two sentences longer, and simpler.
Two methods, one direction
One asks, the other tells. Both reveal how far you understand only when you produce your own sentences. Both needed a counterpart, and that counterpart now sits beside you around the clock.
The order I use: when I do not know, ask Socratically; when I think I know, explain Feynman-style; when the explanation breaks, go back to asking. Walking between the two is what separates feeling that you understand from understanding.