How spaced repetition with AI flashcards beats cramming

Cramming is a rational response to a bad incentive structure. If the only thing that matters is the exam on Friday, and you're measured once, cramming works. You can get a B+ on a biochem midterm with twelve hours of concentrated memorization on Thursday. The problem is that by the time the final rolls around six weeks later, you remember roughly 15% of what you crammed.

Spaced repetition is the opposite bet. It assumes you care about remembering the material months or years later — which is what happens when the content is prerequisite for the next course, or when you'll use it at work, or when the exam itself tests cumulative understanding. This guide explains why spacing works, how the underlying algorithm schedules your reviews, and why AI-generated flashcards have turned SRS from a power-user tool into something every student can use without pain.

The forgetting curve is real

Ebbinghaus's forgetting curve is one of the most replicated findings in psychology. When you learn something new, your retention decays roughly exponentially — fast at first, then slowing down. By 24 hours you've lost about 70%. By a week you've lost about 90%, if you do nothing. The curve bends a little differently for different types of content, but the shape is universal.

The insight that launched SRS: every time you successfully retrieve a memory, the decay curve resets and gets flatter. The second time you recall a fact, you forget it more slowly than the first time. The third time, slower still. This means the optimal review schedule is non-linear — closely spaced at the start, then exponentially further apart as the memory strengthens.

The SM-2 algorithm, explained simply

SuperMemo's SM-2 algorithm, published by Piotr Woźniak in 1985 and still the basis of Anki's scheduler, works like this. Every flashcard has two properties: an interval (how many days until the next review) and an ease factor (a multiplier between 1.3 and 2.5 that determines how fast the interval grows).

When you grade a card from 0–5 based on how easy recall was:

• If you got it right and easily (4–5), the interval multiplies by the ease factor. Ease may tick up slightly.
• If you got it right but with effort (3), the interval multiplies by the ease factor, but ease ticks down.
• If you got it wrong (0–2), the interval resets to 1 day and ease drops. The card goes back into daily rotation until you've re-learned it.

That's it. Two numbers per card, a handful of branches, and you get an algorithm that automatically figures out which cards you know well and which cards need more attention — with no effort from you beyond honest grading.

Why cramming loses to spacing, measured

Cepeda et al. (2008) ran a massive study comparing spaced vs. massed study across 1,354 participants. At 10 days after learning, students who spaced their study with a 1-day gap scored 8 percentage points higher than students who massed the same total study time. At 1 year, the gap was 22 percentage points.

In other words: same total hours of study, dramatically different retention. The mechanism is the same one SM-2 exploits — every time you successfully retrieve, you strengthen the memory, and spacing forces you to retrieve from a weaker state, which creates a stronger retrieval cue next time.

The flashcard-making bottleneck

SRS works. Everyone in learning science agrees. Yet most students who try Anki quit within two weeks. Why?

Making flashcards is brutal. For a single 60-minute lecture, you need to read the material, identify the atomic claims, phrase each as a question, phrase the answer, tag the card, and drop it into a deck. 20–25 cards per hour of lecture is a reasonable target. That's two hours of card-making on top of attending the lecture.

Serious medical students do it anyway — the USMLE makes it worth it. But undergraduate chemistry students? Law students reviewing cases? They mostly don't. The friction is too high, and so they fall back on cramming, which is what we were trying to replace.

What AI flashcard generation actually fixes

AI doesn't improve the algorithm. SM-2 has been fine since 1985 and a few newer algorithms (FSRS, 2023) squeeze out another 10%. What AI fixes is the friction. A modern pipeline takes a transcript, finds the 20–25 atomic claims, generates question-answer pairs for each, and drops them into your SRS — in about 60 seconds.

This flips the economics entirely. The time cost of running SRS goes from 2+ hours per lecture to 20 minutes of actually doing the review. That's the difference between "I tried Anki for two weeks" and "I've been running SRS every day since February."

A minimum-viable spaced-repetition routine

1. After each lecture, auto-generate 20–25 flashcards using any AI tool that exports to Anki (or uses a built-in SRS).
2. Skim the generated cards for 60 seconds. Delete anything obviously wrong or redundant. Don't obsess — 80% of value comes from 80% coverage.
3. Every day, do a single SRS review session. On a typical semester workload, this is 15–25 minutes.
4. When you get a card wrong, read the linked transcript context (good tools show you this automatically) before moving on. This is the single most important habit — understanding why you got it wrong doubles the retention boost of the retrieval attempt.
5. Never skip more than two days in a row. The backlog grows quadratically and becomes a wall.

The habit is the whole game

People who win at SRS aren't smarter than people who don't. They've just made it boringly mechanical. Open the app, do the review, close the app. Twenty minutes. Same time every day. The AI removed the last excuse — you don't have to make the cards — and the only remaining barrier is showing up.

Here's the uncomfortable truth about spaced repetition: it makes studying less interesting, not more. The cards are short. The sessions are short. Nothing feels like a breakthrough. But six weeks later, when your classmates are cramming and you already know the content cold, the boring discipline pays off.