Spaced Repetition and AI Flashcards: Memory That Survives Exam Week

A week before exams, a S2 Maths teacher walks in and hears: "I revised the whole chapter three times last night — half of it is gone already." She does not hand out another set of notes. She splits "the discriminant of quadratic equations" into twelve flashcards for the class with one rule: each day, only the cards due today, then self-rate. Two weeks later a quiz shows pass rate on discriminant items rising from 54% to 71%. She says, "Not more revision — revision on the right days."

Massed cramming works for the short term; research consistently favours spaced repetition for retention. AI splits cards, schedules reviews and returns data; teachers review card quality and adjust teaching from the numbers.

Why cramming does not stick

Ebbinghaus's forgetting curve suggests more than half of new material can vanish within twenty-four hours. Hong Kong exam culture pushes massed revision — revision weeks, overnight essay drills, whole-topic re-reads. Spaced repetition asks for a different contract with pupils: less heroism the night before, more steady five-minute sessions for three weeks. That shift needs teacher modelling in the first lesson, or pupils treat the deck like another homework pile and cram it the night before the quiz anyway. Re-reading a whole chapter three times the night before an exam feels like mastery; a week later much is gone — because all repetition was massed, never timed at the edge of forgetting. Spaced repetition is review at the right time, not more passes. Each successful retrieval just before forgetting lengthens the next interval; only then does the curve flatten and memory survive exam week.

How SM-2 scheduling works

SM-2 is a classic, transparent algorithm. Each card has an ease factor and a next-review date. After answering, pupils self-rate 0–5: 5 easy and correct — interval multiplies by ease; 4 correct after thought — slightly longer; 3 barely correct — slightly longer with ease reduced; 0–2 wrong — reset to one day. First correct review in about one day, second about six days, then ease-driven multiples. Pupils open "due today" — usually five to ten minutes, not two hundred cards in one panic session.

Teach honest self-rating

P6 pupils often rate a lucky guess as 5. Spend five minutes in the first lesson modelling: 5 only if no effort, 3 if you guessed, 0 if wrong then read the back. Honest ratings keep the schedule honest; teachers need not teach the formula, but should say "gaming the system games yourself".

Splitting cards with AI

Four steps: Choose source — lesson text, teacher notes or unit objectives. Generate — ask AI for one concept per card: definition, cloze, true/false, image ID, short answer. Review — delete vague, off-syllabus or multi-concept cards; General Studies benefits from image cards. Assign — save to the panel deck, assign to classes; pupils can build cards from their own mistakes.

For reflection of light: "What is it called when light changes direction on a smooth surface?" → Reflection. "How are incidence and reflection angles related?" → Equal. Image card: "Which line is the normal?" → Dashed line perpendicular to the mirror. Three cards, three concepts — far better than one card pasted with a whole page.

One concept per card is the line between spaced repetition that works and a deck nobody finishes.

Ten minutes on the pupil side

Each day: due cards, answer, self-rate. In Edor.ai the pupil app shows only today's queue — not the whole deck — which protects the habit. A parent who asks "how much revision tonight?" gets a concrete answer: "Five cards, about eight minutes," not "the whole chapter again". Streaks and subject leaderboards give moderate motivation but should not be the only goal. Read-aloud and large text support SEN pupils; the mobile PWA allows review on the bus or at recess. The habit is small, daily, no debt — if someone skips three days, due cards pile up; teachers nudge early, not one week before exams.

Teachers turn data into lessons

Before spaced repetition, Ms Wong guessed that "discriminant" was the weak point because pupils looked confused in class. After two weeks of card data, she saw that cards on "setting a equal to zero to find k" failed three times as often as the discriminant formula itself — a different re-teach target entirely. Review data answers three practical questions: most-missed class cards — re-teach or re-example next lesson; persistently behind pupils — individual follow-up, check dishonest ratings or non-use; pre-exam mastery — which unit decks have heavy due loads and high error rates for revision lessons. One teacher projects "weakest five cards" for the first three minutes every Monday; pupils see they are not alone — more solid than guessing blind spots.

Common misreadings

Card design also matters for bilingual schools: front-of-card language should match what pupils must produce in the exam — English Science terms on the front if the test is English, even if the lesson was taught bilingually. AI can generate both sides; the teacher chooses which deck to assign.

• The claim: More cards are always better — generate the whole book at once. What actually happens: Oversized decks get abandoned; release unit by unit, ten to twenty core cards each, teacher-reviewed before the next batch.

• The claim: AI cards are ready to assign unchanged. What actually happens: AI splits concepts well but produces off-syllabus, vague or wrong backs; a five-minute skim beats writing from scratch.

• The claim: Spaced repetition is only for rote subjects. What actually happens: Vocabulary, General Studies concepts, formulae and procedure names all fit; extended argument still needs class discussion — cards secure basic retrieval.

Three things to try on Monday

1. Pick one unit before an upcoming test, generate ten to fifteen one-concept AI cards, delete vague entries and pilot with one class.
2. Spend five minutes in the first lesson modelling SM-2 self-ratings 0–5; stress honesty over speed.
3. Block three minutes every Monday to view "weakest five class cards" and note re-teach targets for the next lesson.