aitutors.me vs SuperMemo: the original, and what we owe it

SuperMemo invented spaced-repetition software. Both the algorithm in our product and the rule we use to judge whether a card is well made came from it. It is also the least suitable of the major tools for a twelve-year-old, and it has never pretended otherwise.

At a glance

What we took from SuperMemo

Two things, and they are not small.

The algorithm. Our scheduler is SM-2 — the algorithm Woźniak published in 1990, described in Optimization of repetition spacing in the practice of learning. Ease factors, quality grades, the 1-day then 6-day opening, the multiply-by-ease progression: that is all SuperMemo's work. Anki's default scheduler is the same algorithm. An enormous share of the world's spaced repetition traces to that one paper.

The rule for judging a card. SuperMemo's Twenty rules of formulating knowledge contains the minimum information principle: keep each item as small as it usefully can be. That single rule is the difference between a deck that works and a deck that gets abandoned, and it is the rule our card checks enforce automatically — a question asking for several things at once is rejected before it is saved.

We did not improve on either. We implemented them and built a service around them.

Where SuperMemo has gone since

This is worth being accurate about, because outdated claims circulate.

SuperMemo did not stop at SM-2. It has iterated for three decades: SM-17 introduced a substantially different model, SM-18 arrived with SuperMemo 18 in 2019 with a new approach to computing item difficulty, and Algorithm SM-19 now ships with SuperMemo 19. These later algorithms model memory as a two-component system — retrievability and stability — rather than adjusting a single ease number, and they are meaningfully more accurate than SM-2.

So when anyone describes SM-2 as "the SuperMemo algorithm", that is thirty-five years out of date. It is the SuperMemo algorithm we use, which is a different statement.

Why we still use a 1990 algorithm

Three honest reasons.

It is published and exact. SM-2 can be implemented precisely, tested to the day, and explained to a parent in a table. We publish our actual ladder — 1, 6, 15, 38, 90 days — because we can. A modern adaptive model cannot be summarised that way.

Modern models need history. SM-17 onward, and FSRS in the Anki world, work by fitting parameters to a large volume of review data. A child two months into using flashcards does not have that data, and an adaptive model without it performs no better than good defaults.

Our binding constraint is elsewhere. The gap between SM-2 and a modern scheduler is efficiency — roughly the same retention for fewer reviews. That matters enormously at 10,000 cards. At a few hundred cards inside a 20-a-day cap, it is not what stands between a child and a study habit. Card creation is.

Our review log was deliberately built so a better algorithm can be fitted later. That is the upgrade path, and it becomes worth taking when there is history to fit against.

Incremental reading, and why we don't have it

SuperMemo's most distinctive feature has nothing to do with flashcards. Incremental reading lets you work through long articles in pieces over months, extracting fragments into questions as you go, with the whole reading queue itself under spaced-repetition control.

It is a genuinely original idea and, for a self-directed adult building deep knowledge of a field, possibly the most powerful learning tool ever shipped. It is also completely unsuited to a KS3 child following a fixed national curriculum, who does not need to construct their own syllabus from source material — they need to learn the one they have been given.

The interface gap

SuperMemo has never optimised for approachability, and its users tend to treat the learning curve as an initiation rather than a defect. It is Windows-first, dense with options, and expects you to author your own collection.

That is a coherent position for a tool aimed at people who will use it for thirty years. It is the opposite of what a twelve-year-old needs on a Tuesday evening.

Who should use which

SuperMemo, if you are an adult, serious about long-term knowledge, and interested in incremental reading. Nothing else does what it does.

Us, if the learner is a KS3 child with a tutor. We use SuperMemo's algorithm and SuperMemo's card-quality rule, wrapped in a service that writes the cards and keeps the daily queue small enough to finish.

FAQ

What is SuperMemo?

The original spaced-repetition software, created by Piotr Woźniak, and the source of the SM-2 algorithm that Anki and we both use. Its own algorithm has since advanced well past SM-2 — SuperMemo 19 ships Algorithm SM-19.

Why do you use SM-2 when SuperMemo is on SM-19?

SM-2 is published, simple enough to implement exactly and to explain honestly to a parent. The later algorithms are more accurate but need substantial review history to work well, and a child's collection does not have it yet.

Is SuperMemo suitable for a child?

Realistically, no. It is a Windows-first tool with a steep interface built for serious adult learners, and it expects you to author your own material.