Let AI Teach Me · Make it stick

Let AI Ask a New Question Every Time

A fixed old question will be memorized. AI’s unique value is infinite new questions. Review belongs right when you are about to forget.

Question six asked what voluntary refund means. Sunday night at the desk, you pasted one page of ticket-change notes to AI. You had already been reviewing for 18 minutes.

The notes said:

If the passenger asks for a refund because their own plans changed, that is a voluntary refund. If the airline cancels, that is an involuntary refund.

AI asked in a row: “What is a voluntary refund?” “What is an involuntary refund?” You answered from the definition. Three rounds, 6 for 6. By round three the question had barely appeared and you could finish the sentence. You closed the chat sure you could use both rules.

Two mornings later at 7:20 your phone said the afternoon flight was canceled by the airline. The order page also said sale fare, no refund, no change. You stared at those words, decided the ticket was dead, and prepared to eat the 780.

The person traveling with you tapped “involuntary refund.” Four minutes later the request was approved — 780 back the original way. When it was your turn you froze. Yesterday you could recite both definitions. On the same order, which one covers an airline cancel, and which covers you changing plans? You were not sure.

You judged that way because the old questions already printed the term on the stem. See “voluntary refund,” and you only had to finish the definition. The order page does not label the category. You have to notice an airline cancel is a carrier reason, then file it as involuntary. A definition you can recite only means you know the tag. Change the conditions, and you still have to pick the rule.

AI can hide the same rule in a new situation

Ask AI to “quiz me again” and it will likely reshuffle those two definitions. The score will climb because all three rounds are hunting the same card. The live-order judgment still was not practiced.

AI is more useful when every round changes the person, the time, and the limits, so you pick the rule again in a situation you have not seen.It can also hold the answer until you reply, so the explanation does not arrive with the question.

Why do new questions drop the score on the spot? And what illusion does same-day review create? Keep the same ticket rules below, and dial both the question type and the days between rounds.

Try it first · Why a fixed old question fools you
Review the same ticket-change notes for three rounds
Days between two review rounds
3 days
A new order two weeks later
1
2
3
Just learned the rules
How you answered on the spot across three rounds
All three rounds answer the same definition. After play, each round’s result shows here.
Two weeks later: the canceled ticket from the opening
When the three rounds end, an unseen order question appears here.
Watch three rounds of the fixed old question, then switch the button to a new question each round.
Why the high score on the left misleadsCram three rounds into that day and repeat the same definition, and what you drill is the pairing of stem and answer.
Which result you should actually watchA new question each round makes you think longer on the spot, and two weeks later you can still drop the rule into a new order.
Teaching demo: questions and bar lengths show direction of change, not a real measurement.
Paste this whole block to AI

Swap in your topic and notes. Keep the other constraints. Paste the notes after this block so AI has a range to write from.

Copy this whole block

I just finished ticket change-and-refund rules. Notes are below. Write questions only from my notes. No definitions. Give me two scenario questions. Hold the answers.

After you answer, add: “First name which rule I used and which condition I missed, then give the answer. Next round, change the person, time, or fare limit, and test the same judgment.”

The three constraints in that sentence each block a common false “I get it”:

  1. “No definitions” blocks tag-memorizing. The stem no longer names the concept. You have to classify what just happened.
  2. “Give me scenarios” blocks paraphrase drills. AI has to put in a person, an action, and a limit, or the question will not force a rule choice.
  3. “Hold the answer” blocks the leak. If the explanation arrives early, you only have to nod along. Once the answer is hidden, you have to retrieve the rule with the notes closed — let AI quiz you once.

“Write only from my notes” matters too. It locks the range to the page you just learned, so AI does not pad in ticket lore you have not studied. The opening question can be: a sale fare says no refund, no change, and the airline canceled — can you still get the money back the original way?

When the next new question should come

Cram three rounds into that day and the last answer is still in front of you, so it feels smooth. Wait, then do a new question: retrieval slows and misses rise. That is the gap showing. Review is for being able to judge again after a gap. A perfect score on the spot is only a stretch of the process.

The spacing effect has a long research trail. Late in the nineteenth century Ebbinghaus recorded how forgetting moves with time. Cepeda and colleagues published a 2006 review of distributed practice. They support spreading study across time. They do not compute one exact day-count for everyone.

In practice, use the feel of answering: if it comes with almost no thought, push the next round farther; if you cannot retrieve it, pull the next round closer. After a miss, write which rule you used, then let AI compare — see Let AI find the step I got wrong. To keep practicing retrieval: Close the Notes and Let AI Quiz Me.

Using a rule means it still works after the conditions change.