Home▸Education & Learning Tools▸Generative AI Tutoring: Adaptive Difficulty Engine

Spaced Retrieval and the Forgetting Curve: Why Review Timing Beats Review Count

Twenty flash cards forget along an Ebbinghaus-type exponential curve, and every review that succeeds makes a memory more stable, more so the harder the recall was. Compare three schedules with the same number of reviews: adaptive (review when retention falls to a target), evenly spread, and crammed into the last day and a half before an exam on day 60. See exam-day recall, recall 30 days later, study minutes and the growth of memory stability. Blender-built study desk, with exact-tree, Monte Carlo and ODE checks built in; the memory-update rule is illustrative and not tied to any product.

Education & Learning Tools3DEasy60 FPS📱 Mobile-adapted🟠 Blender⇄ 2D version
Generative AI Tutoring: Adaptive Difficulty Engine ↗ Open standalone

A generative-AI tutor doesn't hand every student the same worksheet — it watches each learner's success rate and continuously re-targets task difficulty to keep them challenged but not overwhelmed. This simulation visualizes that personalization loop for a whole class at once: students spiral toward the glowing AI core as their mastery grows, while color shows whether the AI's current difficulty choice is landing in their productive zone, running too hard, or too easy. Adjust the target success rate and adaptation speed to see how a mistuned tutor stalls learners in frustration or boredom instead of steady progress.

⚙ Under the hood

Twenty flash cards forget along an Ebbinghaus-type exponential curve, and every review that succeeds makes a memory more stable, more so the harder the recall was. Compare three schedules with the same number of reviews: adaptive (review when retention falls to a target), evenly spread, and crammed into the last day and a half before an exam on day 60. See exam-day recall, recall 30 days later, study minutes and the growth of memory stability. Blender-built study desk, with exact-tree, Monte Carlo and ODE checks built in; the memory-update rule is illustrative and not tied to any product.

Three.jsgenerative AIadaptive learningeducationagent-basedpersonalizationSpaced RepetitionForgetting CurveRetrieval PracticeLearning ScienceEbbinghaus

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install

What did you find?

Add reproduction steps (optional)