You Only Understand What You Can Explain

The Twelve-Year-Old Test

Look, I’ve got a simple test for whether you really understand something. Try explaining it to a twelve-year-old. Not dumbing it down—explaining it clearly. If you can’t do that, you don’t actually get it. This isn’t theoretical. I spent years teaching undergrads and doing Nobel-level research, and the pattern held everywhere: the people who truly understood their work could make it simple. The ones who hid behind jargon? They were hiding from their own ignorance.

Real learning isn’t about accumulating facts. Learning to learn chess teaches you more than chess itself—it teaches you how to measure progress and find what works. Breaking down neural network math into modular pieces using the chain rule means understanding how rates of change compose. Reframing linear regression as probabilistic inference rather than line-fitting demonstrates real understanding by building from different foundations.

When Fancy Words Hide Empty Thoughts

Watch what happens when you force someone to explain without jargon. “The model exhibits computational irreducibility” sounds impressive. But what’s actually happening? Most people can’t say. They’ve memorized the phrase, not grasped the concept.

This is uncomfortable to discover in yourself. You think you understand something, then try teaching it and hit a wall. That wall is where your real understanding stops and your illusion begins. We all do it—use complexity as camouflage.

Academic culture rewards sounding smart over being clear. Someone talks about “sparse autoencoders extracting polysemantic features.” Sounds sophisticated. But ask them to explain why neural networks have this “dark matter” of hidden knowledge, and watch them struggle. The jargon is a shield. Behind it, understanding might be thin.

The dangerous part? You can fool yourself. You read papers, understand the words, use them correctly. But when forced to rebuild the concept using only simple language—from first principles—the structure collapses. That’s the test.

How to Tell If You Really Get It

Pick something you think you understand. Explain it without jargon. No technical terms, no hand-waving. Just build it from pieces a smart kid would grasp.

You’ll get stuck. Good. Where you get stuck is where you don’t actually understand. You’ve been pattern-matching without building the real mental model.

Go back. Learn that part by working through it until you can rebuild it yourself. Then try again. Keep going until it’s simple.

This is the Feynman Technique. It’s brutal because it’s honest. When AI researchers teach one model from another’s outputs—knowledge distillation—the student can only learn what the teacher clearly expresses. Hidden structure stays hidden. Same with us.

Understanding Means Simplicity

Teaching isn’t about sounding smart. It’s about making others understand. If you truly understand something, you can make it simple. Not simplistic—simple.

Real mastery means clarity, not complexity. When you can build quantum mechanics or Bayesian inference from first principles using everyday concepts, you’ve got it. Not when you cite papers or throw around terminology.

The first principle is not to fool yourself. And you’re the easiest person to fool. Can you explain it simply? If not, you don’t understand it yet.