In 1885 a German psychologist named Hermann Ebbinghaus spent months memorising nonsense syllables and testing himself at intervals. What he produced is still one of the most reliable findings in psychology: the forgetting curve.
Learn something today, and by tomorrow you have lost most of it. Not because you are careless — because that is what memory does with information it has no reason to keep.
The interesting part is not that we forget. It is that each time we recover something just as it is slipping away, it comes back more slowly the next time.
The counter-intuitive part
Reviewing something five times in one evening feels productive. It isn't, or not much. The material never leaves your working memory, so you never actually practise retrieving it — and retrieval is the thing that strengthens the trace.
The same five reviews, spread across three weeks, can hold for years. Same effort, same number of repetitions, radically different outcome. The only variable that changed is when.
How the intervals grow
A working schedule for something you have just learned looks roughly like this:
- First review: one day later
- Second: three days later
- Third: one week later
- Fourth: two weeks later
- Fifth: one month later
Every successful recall pushes the next one further out. Every failure pulls it back sharply, usually to the beginning. That asymmetry is deliberate: forgetting is evidence that the interval was too long, and the schedule should believe the evidence.
Why the timing has to be personal
Those numbers are a starting point, not a law. You will hold a formula you understand deeply for weeks, and lose a date in Roman history by Thursday. A fixed schedule treats both the same way, which means it is wrong twice: too frequent for one, too rare for the other.
This is what algorithms like SM-2 and, more recently, FSRS are for. They keep a difficulty estimate per card and adjust its interval independently. In practice the effect is that your review queue stops being full of things you already know.
Where people go wrong
Making cards that ask too much
“Describe the causes of the First World War” is not a flashcard, it is an essay prompt. You cannot grade yourself honestly on it, so the schedule gets useless feedback. Break it into pieces small enough to be clearly right or wrong: one card per cause.
Reviewing only before the exam
Spaced repetition is not a revision technique, it is a maintenance technique. Starting it three days before the exam is cramming with extra steps — which works, briefly, and is gone the following week.
Skipping the hard cards
The cards you dread are the ones doing the work. Every time you push one to tomorrow, you spend the session on material you already know and leave feeling productive without having learned anything.
The honest cost
The technique is well established. The reason most students don't use it is not scepticism — it is that making the cards is tedious. Turning a ninety-page lecture PDF into two hundred atomic questions takes hours you were going to spend studying.
That gap between “this works” and “I will actually do this” is the entire reason Twixem AI exists: the cards are generated from your document, and the schedule runs itself.
Start tonight
Take one chapter. Not the whole course — one chapter. Turn it into fifteen or twenty questions, and answer them tomorrow, then in three days. That is enough to feel the difference, and small enough that you will actually do it.