Explain It Like I'm Five

Four hours into something complicated, the conversation has a texture. Terms have been established. Decisions have been made and referred back to. There is a shared vocabulary that did not exist this morning, and the replies have started to arrive in it — dense, confident, and full of nouns that mean something specific now.

That texture is mostly a good sign. It is also the exact condition under which you stop being able to tell whether any of it is right.

So stop and ask for the smallest version.

Explain what we are doing here like I'm five.

The first thing it buys you

You find out whether you are still working on the same problem.

Long sessions drift. Not dramatically — nobody announces a change of subject — but a clarification here and an assumption there, and by hour four the thing being built has moved some distance from the thing you set out to build. Both of you have been present for every step, which is precisely why neither of you noticed.

The five-year-old version collapses four hours into five sentences, and the collapse is the test. If the summary describes something you do not recognize, you have learned something expensive at a very low price. If it describes exactly what you wanted, you have also learned something, and you can get back to work with more confidence than you had a minute ago.

It re-anchors the machine too. A plain restatement pulls the working context back to the thing itself, away from the accumulated scaffolding of terms invented along the way.

The second thing, which is why you actually do it

Complexity is where bad reasoning hides.

This is not a fact about machines. It is why the impenetrable paragraph in a contract is impenetrable, why the consultant's slide has six nested clauses, why nobody asks a question after the dense part of the talk. Technical vocabulary does real work — it compresses ideas that would otherwise take a page — and it also provides cover, because a sentence full of established terms sounds load-tested whether or not anything tested it.

A model produces that kind of sentence readily and for no sinister reason. It is completing a pattern, and the pattern it is completing is what an expert explanation looks like. Expert explanations are dense. So the output is dense, and density is indistinguishable, from the outside, from rigor.

Take the vocabulary away and the cover goes with it. A step that never made sense cannot be smuggled through in short words, because there is nothing to smuggle it inside. Either the plain version holds together — in which case you now understand it better than you did — or it comes apart in your hands, in public, in a paragraph you can point at.

You will also catch a certain amount of showing off. Sentences that were performing sophistication rather than carrying an idea do not survive translation into words a child would follow. They simply have nothing to say once the costume comes off.

Why this feels rude and is not

There is a reflex to resist here, and it is worth naming because it is what stops people doing this.

Asking for the simple version feels like admitting you did not follow the complicated one. Between people, that admission costs something real — status, mostly — which is why rooms full of intelligent adults sit through explanations nobody understood and ask nothing.

None of that applies. There is no one to lose face in front of, no one keeping score, and no cost to asking the same question four times in an hour. Whatever social machinery normally makes this expensive is entirely absent from the exchange, and that is one of the few genuinely free things on offer. Use it more than feels dignified.

The rule underneath

Richard Feynman is usually the name attached to this, and the version that survives him is close enough: if you cannot explain it simply, you do not understand it well enough. He aimed it at physicists, at himself, at anyone inclined to hide behind formalism.

It transfers without modification, because the thing it tests is not a property of minds. It is a property of explanations. A correct account has a small version. A broken one does not, and the reason a broken one does not is that the complexity was holding it up.

So ask for the small version, periodically, out loud. If it arrives and it is clear, you have lost ninety seconds. If it does not arrive, you have just found the crack — before you spent another hour building on top of it.