Most of us learned the same way, and no one told us it was inefficient. You read the chapter. You reread it, maybe with a highlighter. By the third pass the material feels familiar, comfortable, almost obvious. So you close the book, satisfied that you have learned it. Then the exam, the interview, or the real bug arrives, and the knowledge that felt so solid is nowhere to be found.
Rereading your notes feels like learning. The research says it mostly is not.
There is a name for what happened. Asher Koriat and Robert Bjork, studying how people judge their own learning, described an “illusion of competence”: when material processes fluently and feels familiar, we mistake that ease of recognition for genuine mastery. The smoother a page reads on the second and third pass, the more confident we become — and that confidence runs ahead of what we can actually recall and use when the material is not sitting in front of us.
Fluent, familiar processing makes us mistake recognition for mastery.
Why the feeling lies to you
The trap is that the signal we naturally trust — how easy something feels — is measuring the wrong thing. When you reread, you are testing whether the words look familiar. They do, because you just read them. But recognition is a much weaker state than recall. Being able to nod along with an explanation is not the same as being able to produce it, apply it, or debug with it when nothing is prompting you.
This matters more for engineers than for almost anyone. Our work does not ask us to recognise a correct answer from a list. It asks us to generate the right move in an open situation: read the datasheet, hold the register layout in your head, reason about the race condition, and write the fix. That is recall and construction under load, not recognition. Study methods that only build recognition leave you feeling ready while your actual capability lags behind.
Recognition is not recall. Being able to follow an explanation is not being able to produce it.
The fix is retrieval, not more rereading
The correction is simple to state and uncomfortable to do: stop feeding yourself the answer and start pulling it out. Close the book and rebuild the idea from memory. When you force retrieval, two useful things happen at once. You strengthen the memory itself, because the act of recalling is what consolidates it. And you get an honest reading of what you actually know, because the gaps announce themselves immediately instead of hiding behind a familiar page.
Concretely, for technical study and work:
- After reading a section, close it and write down the core idea from memory, in your own words. Then reopen and check what you missed.
- Rebuild the mechanism, not the sentence. Redraw the diagram, re-derive the sequence of calls, or re-explain the data flow without looking.
- Teach it out loud to an empty room, or to a colleague. The points where you stall are exactly the points you had only recognised, not learned.
- Space it out. Come back a day later and retrieve again, before rereading. If you cannot reproduce it cold, you had not learned it — you had only met it.
- Treat discomfort as the signal. The struggle to recall feels worse than smooth rereading, and that difficulty is precisely why it works.
The fix is not more rereading. It is retrieval: close the book and rebuild it from memory.
What this has to do with the AI age
The illusion of competence is now available on tap. When an assistant hands you a clean, correct-looking explanation, it reads fluently by construction. That fluency triggers the same false confidence as a third rereading, except faster and more convincing. You can spend an afternoon with a model, feel that you understand a subsystem, and retain almost none of it — because you never once retrieved anything. You only recognised good answers as they scrolled past.
The discipline that protects you is the same one Koriat and Bjork point to. Use the tool to generate explanations and check your work, but do the retrieval yourself first. Attempt the answer before you ask. Rebuild the mechanism before you read the model’s version. Let the assistant test you, not replace the effort that actually builds the knowledge into your head. When answers are free and fluent, the engineers who still practice recall are the ones who can think without the tool open.
When answers are free and fluent, the ability to recall without the tool open is the real skill.
This is why, at TECH VEDA, we push learners to rebuild kernel and driver internals from memory and from source rather than reread notes — because the goal is knowledge you can think with, not knowledge you can merely recognise.
— Raghu Bharadwaj
