There is a very specific kind of confidence that comes from having read something. You finish a chapter, close the book, and feel like you now know the thing. The feeling is real. The knowledge, mostly, is not.

I have made this mistake enough times to recognize it. I read through a topic - carefully, even - and the next week someone asks me about it and I find myself holding a vague shape where understanding should be. I remember the general territory. I cannot reproduce a single useful detail. I had the experience of reading without the thing reading is supposed to produce.

The problem is that reading is passive and understanding is not. Reading moves information in front of your eyes. Understanding requires you to do something with it - to connect it to something you already know, to notice where it contradicts something you believed, to see what it implies that the author did not explicitly say. Those connections are the thing. The information without the connections is just words.

Build the Web: Solve Problems

What forces those connections is problems. Not exercises in the sense of “apply formula A to input B” - those are not problems, they are rituals. I mean the kind of problems where you have to sit with something that does not yield immediately, where you have to reach for what you know and find that you do not know it as well as you thought, where the gap between understanding and not-understanding becomes visible and uncomfortable.

That discomfort is the mechanism. It is the thing that makes you go back and actually look at what you skimmed past.

When I am learning something seriously, I look for problem sets from university courses - MIT OpenCourseWare, Stanford’s course pages, whatever is available. The problems there are not random. They are assembled by people who have spent years thinking about what trips students up, what concepts tend to coexist as misconceptions, what the minimum set of exercises is to force engagement with the hard parts. They are, in a real sense, curated friction.

Books have problems too, and good ones. But for someone new to a topic, the course sets have an advantage: they are calibrated to a progression that someone has already thought through. You do not have to figure out what order to tackle things in. You just follow the thread.

Maintain the Web: Notes and Repetition

The problem sets build the web. But the web has a maintenance problem.

Even after you have understood something genuinely - after you have worked through the problems and felt the click of a concept landing - you will forget it. This is not a personal failure. It is the default behavior of human memory, and it applies to everyone, including people whose names are on the theorems. Memory is not a filing cabinet. It is closer to a garden, and things that are not tended grow over. The connections you made fade. What was crisp becomes vague. The web becomes sparse.

The two things that slow this down are notes and repetition, and they work together.

Notes in this context does not mean a transcript of what you read. It means a record of your own understanding - in your own words, capturing exactly what you need and nothing more. The test is whether a future version of you could open those notes and reconstruct the thing quickly, without re-reading the source. If your notes are just quotes and summaries, they fail that test. If they are the output of you actually thinking about the material - the key insight stated plainly, the thing that confused you and how it resolved, the connection to something else you know - they pass.

The act of writing notes this way is itself a form of processing. You cannot write a clear explanation of something you do not understand. You find out what you actually know by trying to state it.

Repetition is what keeps the web from dissolving. Not repetition in the sense of re-reading - re-reading gives you the false confidence again. Repetition in the sense of returning to the ideas and reasoning through them: looking at your notes and trying to reconstruct the material before you open them, working a problem from memory and checking the answer, explaining a concept to someone who does not know it.

Each of these is a retrieval act, and retrieval is what strengthens memory in ways that re-exposure does not. You are not putting the information in again. You are practicing finding it.

AI has changed this particular part of learning in a way I did not fully expect.

The repetition problem used to require real infrastructure: flashcard systems, study groups, scheduled review sessions. All of these work, but they all carry friction. You have to remember to return. You have to set aside time. For many people, that friction is exactly where the habit breaks down.

What AI changes is the minimum viable retrieval act. You do not have to sit down with a system. At any moment - waiting for something, commuting, between other things - you can try to recall something, find the edge where your memory breaks down, and immediately probe that edge. Ask a question, get a response that either confirms or corrects your understanding, and repeat. That is active recall, the same mechanism that strengthens memory. The difference is that it now costs almost nothing to initiate.

The exponential decay of memory is not a fixed fate. Every successful retrieval slows the decay. This has always been true, but exploiting it used to require deliberate scheduling and a real commitment to the habit. Now it requires almost nothing except the will to do it. If you genuinely want to learn and retain something, the tools to do it are sitting in your pocket, accessible at the press of a button. That is a remarkable thing to have access to, and most people use it for everything except this.

Solidify the Web: Use It

But the version that sticks most reliably is use.

The things I remember best are the ones I have had to reach for in a real context - not for a test, but because something depended on it. When you use a piece of knowledge to actually do something, it gets connected to the memory of doing that thing, to the problem you were solving, to the moment of reaching for it. All of those associations reinforce the original. The knowledge stops being isolated and starts being part of the story of something you built or figured out.

This is where projects matter. Not toy examples, not “hello world” implementations, but something that actually involves making decisions and running into the cases the tutorial did not cover. Those cases are where understanding separates from the appearance of understanding. You can follow along with an explanation while not understanding it at all, because you can always nod when the explanation nods. You cannot build a working thing without encountering the gap between what you know and what you need to know.

AI makes this easier and harder at the same time. Easier in the obvious sense: you can now build things that would have taken months of infrastructure work in a few afternoons. The barrier to having something real to test your understanding against has never been lower.

But harder in the sense that AI is very good at filling in the gaps in your understanding in a way that prevents you from noticing they exist. If you ask AI to write the part you do not understand and paste it in, the thing works, and you have learned nothing. The version that actually works is: build it yourself, use AI to get unstuck when you are genuinely stuck, and make sure you understand every piece well enough to change it.

The ability to tweak a thing is the real test. If you can modify the behavior and predict what will happen, you know it. If you can only run it and hope, you do not.

The hardest part of all of this is that it takes longer than reading. The shortcut is always available: read the chapter, feel the understanding, move on. The web in your head stays sparse, but it grows fast. The alternative is slower and less comfortable and produces something that actually holds up when you need it.

I do not think there is a way to make this process feel good in the moment. The friction is the feature. What you can do is recognize the feeling of false confidence - that specific warmth of having read something - and treat it as the beginning of the work rather than the end of it.