Getting Through Material You Do Not Understand

The paragraph you have read four times

Everyone who has studied anything knows this feeling. You are on page three of a chapter, or halfway down the second column of a paper, and you realise you have taken in nothing since the top of the page. The words are English. The sentences parse. Something in the middle is missing and you cannot name what it is.

Traditionally you had three options: reread it and hope, find someone to ask, or give up and hope it does not come up. The first rarely works, the second depends on somebody being awake, and the third is how gaps become permanent.

This is the job AI is best at, and using it well is mostly about being specific about who you are and what you already know.

Tell it where you are starting from

The single biggest improvement you can make to an explanation is to say what you already understand. Most people ask "explain gradient descent" and get an explanation aimed at nobody in particular: too technical to help, too vague to be wrong.

Give it three things. What you are, in study terms. What you already know that it can build on. And what specifically is not landing. That last one is the most valuable and the one people skip, because working out what you do not understand feels like it should be the answer rather than the question. It is not. "I do not understand why the second term is negative" gets you somewhere. "I do not get it" gets you the textbook again.

Explain [the concept] to me.
Where I am: [e.g. second year undergraduate, studied X but not Y].
What I already understand: [the nearest thing you do get].
What is not landing: [as specific as you can manage, even if it is
"I follow every step but I do not see why we do the third one"].
Start from what I already understand and build to the new thing.
Use no terms I have not already told you I know, or define them
in one line the first time.
Do not summarise the concept back at me in more formal language.

Ask for the same thing three ways

Different explanations fail for different people, and there is no way to predict in advance which one will land. So do not settle for one.

Ask for a plain description, then an analogy from ordinary life, then a worked example with actual numbers or a concrete case. You will usually find that one of the three does something the others did not, and often understanding arrives from the collision of two of them rather than from any single one.

The analogy is worth a specific note. Analogies are how most people actually learn abstract things, and they are also where explanations quietly go wrong, because every analogy breaks somewhere. So always ask the follow-up: where does this analogy stop being true? A good answer to that question teaches you as much as the analogy did.

Explain [the concept] to me three separate times:

Plainly, in about four sentences, no jargon.
As an analogy from everyday life.
As a worked example with concrete numbers or a specific case,
showing each step.
Then tell me where the analogy in 2 breaks down and would mislead me
if I pushed it too far.
My level: [where you are]. What I already know: [nearest thing].

Now the harder half: is the explanation any good?

Here is the problem with a tool that explains things beautifully. A wrong explanation and a right one arrive in the same voice, at the same speed, with the same easy confidence. Fluency is not evidence. In fact fluency is what the system is optimised for, so it is the one signal you should trust least.

Four checks, none of which take long.

Does it match the source? This is the big one, and it is why explanation works best when you have the actual material in front of you. Paste the paragraph you did not understand and ask about that paragraph, rather than asking about the concept in the abstract. Then read the paragraph again with the explanation in mind. If it now makes sense, that is real evidence. If the explanation makes sense but the paragraph still does not, the explanation is probably about something adjacent.

Can you use it? An explanation you cannot apply is a story about a concept, not the concept. Try a problem. Try predicting what happens in a case the explanation did not cover. If you cannot, you have not got it yet, whatever the warm feeling says.

Does it survive a challenge? Ask what a specialist would object to in that explanation, or what it simplifies away. Genuinely useful answers come back. Watch what happens if you push back on something that was actually correct: many models will fold and agree with you. That tells you agreement is not confirmation, in either direction.

Can you say it without looking? Close the window and explain it out loud, or write four sentences from memory. This is the only check that tests you rather than the text, and it is the one that transfers to Lesson 4.

What good and bad look like side by side