AI flashcards and spaced repetition: skip the deck-building

Spaced repetition has better evidence behind it than almost any other study technique. It also has a completion rate that would embarrass a gym membership. The gap between those two facts is not motivation — it is deck-building.

Writing good cards is slow, and it happens at exactly the moment you are most tired: after the lecture, after the reading, after the work is already done. AI removes that step, and removing it changes who the technique works for.

Why spacing works

Two effects compound. The testing effect: retrieving something from memory strengthens it far more than re-reading it does, which is why highlighting feels productive and isn't. And the spacing effect: reviewing just as you are about to forget produces more durable retention than reviewing while it is still fresh.

A scheduler exists to find that moment for every card independently, so you are always reviewing at the edge of forgetting rather than on a fixed timetable.

What SM-2 actually does

SM-2 is the algorithm SuperMemo introduced and Anki popularised. After each review you rate how easily you recalled the card. The algorithm keeps an ease factor per card and multiplies the current interval by it: recall it easily and the next review moves weeks out; miss it and the interval collapses back to near zero.

The result is that a mature deck spends almost all of its review time on the small fraction of cards you keep getting wrong. That is the entire value proposition — it is a scheduler for your attention.

Where the technique breaks

• Deck-building fatigue. The most common failure. The deck is never finished, so the reviews never start.
• Borrowed decks. Shared decks skip the work but rarely match your syllabus, and cards written for someone else's course test someone else's emphasis.
• Bad cards. One card should test one fact. A card that tests five is a card you will always rate as failed.
• Backlog collapse. Skip four days, return to six hundred due cards, quit. The scheduler punishes gaps hard.

What AI changes

Generation from your own material fixes the first two problems directly. If the cards come from your notes — your lecture capture, your clipped reading, your meeting write-up — they test the version of the topic you actually need, and they exist within seconds of the note.

A hand-written card is usually better than a generated one, because writing it is itself an act of study. But the comparison that matters is not generated-versus-handwritten. It is generated-versus-nothing, and that is the comparison most people are actually running.

How it works in Aurora

Second Brain holds block-based notes and clipped pages. Any note can generate a flashcard set, which enters an SM-2 review schedule in the same app. Because the notes are also embedded into your Personal Context Graph, the same material is available as grounding when you ask a question in chat — so revision and research draw on one library instead of two.

Notes, semantic search and flashcards are all included on the free tier, alongside the AI study planner in the Learn module.

What it is not: Aurora has no shared-deck marketplace, no add-on ecosystem, and no fine-grained scheduler tuning. If you have a tuned Anki setup you love, keep it — this is for the much larger group who bounced off deck-building and never got to the reviews at all.

A workable routine

• Take notes as you normally would — the point is that no extra step is added.
• Generate cards the same day, while the material is still fresh enough to spot a bad card.
• Delete the cards that test five things at once. It takes seconds and saves months.
• Review daily, briefly. Ten minutes every day beats ninety minutes on Sunday.
• After a gap, do the oldest cards first rather than trying to clear the backlog.