Test yourself instead

If you take one technique from this module, take this one. It has more evidence behind it than anything else here, and it is the opposite of what most people do.

Re-reading does very little

The standard study method — read it again, highlight, read the highlights — is popular because it is comfortable and because it feels productive.

What it builds is familiarity. The text looks known, so you conclude you know it. That is L1’s illusion, arriving through the most common study habit in the world.

Nothing was retrieved from your memory, so nothing in your memory was strengthened.

Retrieval is the learning

The testing effect is among the most replicated findings in learning research: pulling something out of memory strengthens it substantially more than reviewing it does.

Not testing to measure learning. Testing as the method of learning. The effortful, uncomfortable act of trying to recall is what does the work — and that includes attempts where you fail and then see the answer.

This is why Lrnon’s lessons end with questions rather than a summary, and why the site has review cards at all.

Where AI earns its place

Generating good questions is slow work, which is why most people skip it. That is exactly the bottleneck a model removes.

Below is material I am studying. Write ten questions that test whether I can USE this, not whether I recognise it.

Include: three that require applying it to a new situation, two asking me to explain WHY, two about the boundaries or exceptions, and three I would get wrong if I only skimmed.

Do not give me the answers yet.

That last line matters more than it looks. With answers visible you read them, feel the click of recognition, and skip the retrieval entirely.

Recognition is not recall

The commonest flaw in generated questions.

“Which of these describes X?” can be answered by elimination. You may recall nothing and still be right — which is why the shuffled multiple-choice questions in this course are a check rather than the main event.

Better shapes:

Open recall. “Explain X and give an example.” Application. “Here is a new situation. What happens, and why?” Comparison. “How does X differ from Y, and when does it matter?” Error-finding. “Here is a wrong explanation. Say what is wrong with it.”

That last one is excellent and under-used, because spotting an error requires holding the correct version in mind.

Answer before you check

The discipline that makes the whole thing work.

Write or say your answer first, completely, before revealing anything. Then compare and mark yourself honestly.

Reading a question, thinking “yes, I know that”, and moving on is not retrieval. It is recognition wearing retrieval’s clothes, and it is the most common way people run this technique and get nothing from it.

Then:

Here is my answer. What did I miss, and what did I get subtly wrong?

Subtly is doing work — the near-miss is more useful than the outright error, and it is what you would never catch alone.

Try it now (6 minutes)

Take something you are learning. Ask for ten questions in the shape above, with answers withheld.

Answer three in writing before looking at anything. Then check. The gap between what you thought you knew and what you produced is the lesson.

Check your understanding

1. Re-reading is ineffective mainly because: It takes too long It builds familiarity without retrieving anything from memory People read too fast

2. The testing effect means: Tests measure learning accurately Retrieving from memory strengthens it more than reviewing does Frequent tests reduce anxiety

3. The most common way people get nothing from self-testing: Using questions that are too hard Reading the question, thinking 'I know that', and moving on Testing too frequently

Recap

Re-reading builds familiarity and strengthens nothing. Retrieval — pulling it out of memory — is the learning itself, including the failed attempts. Have AI generate questions that require applying, explaining, comparing and error-finding rather than recognising, with answers withheld. Then answer fully before checking, and ask what you got subtly wrong.