Reading a clear explanation is not learning
A model will explain almost anything clearly, at whatever level you ask, as many times as you like. This is valuable and it is also the trap, because a clear explanation produces a strong feeling of understanding that is not the same thing as understanding.
The feeling comes from fluency. You followed it, so it seems you have it. Then you try to use it and find nothing there.
The fix is to make yourself produce rather than consume.
The prompts that force production
Be tested, not told.
Ask me one question at a time about [topic] at [level]. After each answer, tell me what is right, what is wrong, and what is missing. Do not move on until I have got it. Do not give me the answer when I am struggling: give me a hint.
That last clause matters. Left alone, models answer their own questions after a moment, which removes the effortful part where learning happens.
Explain it back.
I am going to explain [topic] as I understand it. Find every error, every vagueness, and every place I am using a word without knowing what it means.
This is the strongest single technique available, and it is uncomfortable, which is a good sign. Explaining is where you discover which parts you had memorised as phrases.
Ask for the question rather than the answer.
Do not solve this. Ask me the question I should be asking myself to get unstuck.
Ask for the boundary.
Give me three cases where the rule I just learned does not apply, and one where people commonly apply it wrongly.
Knowing where something breaks is most of knowing what it is.
Getting an explanation pitched right
Explain [topic] to somebody who knows [what you actually know] and does not know [what you do not]. Use an example from [your field]. Then give me a question I should be able to answer if I understood it.
The last clause converts an explanation into a check, at no extra cost.
Three specific cautions
It will confirm a wrong understanding. You already know why: preference training rewards agreement. If you explain something incorrectly and ask “is that right?”, you may get “yes, broadly”. Ask “what is wrong with my explanation?” instead: a question that presupposes an error gets you the error.
Explanations can be confidently wrong. For anything you will rely on, check against a textbook, official documentation, or a course. A wrong explanation absorbed as understanding is more expensive to remove than a gap.
Difficulty is not a bug. There is a body of research on effortful retrieval and spaced practice, and its consistent finding is that the struggle is where the learning happens. A tool that removes the struggle removes the mechanism. Studies on AI tutors show real gains when the tool is used to test and prompt, and much weaker results when it is used to supply answers: the same tool, used two ways, with different outcomes.
A workable weekly shape
- Read or watch the primary material: a textbook, documentation, a course. Not the model.
- Explain it back to the model and let it find your errors.
- Be tested on it, one question at a time, with hints rather than answers.
- Ask for three cases where it does not apply.
- Come back in a week and be tested again cold.
Step five is the one everybody skips and the one that determines whether any of it is still there next month.
The version of the promise
A free, patient, infinitely available tutor that will answer at your level in your language and never sighs is a large thing, particularly for somebody without access to teaching. It is also, used passively, a very efficient way to feel educated.
Which one you get is decided entirely by whether you are producing or consuming.
One more thing worth asking
When you are stuck on something and cannot say why, this question does more than any explanation:
What am I probably misunderstanding, given that I am stuck here specifically?
Being stuck at a particular point is diagnostic. People get stuck at the same places in a subject, for the same reasons, and a model has seen a great deal of text about those places. It will often name the confusion before you can, which is a strange experience the first time and a useful one afterwards.
Every lesson here is open. Your progress is recorded once you pass the AI, Actually Explained exam.
Check your understanding
A student explains a concept to a model and asks “is that right?” and is told it is broadly correct. What is the flaw?
Pick the one you would defend. Your answer stays private.
Stuck on something?Ask Asha, your AI tutor