하나의 앱 뒤에 있는
50년의 기억 연구.

Intentional cognitive friction — spacing, retrieval, interleaving, varied context — slows performance in the moment but produces dramatically better long-term retention.

Errors made with high confidence, when corrected, are retained more durably than errors made with low confidence. The surprise of being wrong when you were sure is a neurobiological tag.

Pulling information from memory — even when the attempt fails — produces vastly stronger long-term retention than rereading or reviewing notes. Testing does not measure memory; it creates it.

Reviewing material at expanding intervals produces exponentially better retention than reviewing the same amount of material in one concentrated session. One of the oldest and most replicated findings in cognitive science.

Information that you generate yourself — write in your own words, draw, explain — is remembered far better than information you passively consume, even of identical content.

Mixing topics in unpredictable order during practice — rather than blocking all of one type together — forces the brain to re-recognise the problem type at every trial. Harder in the moment, dramatically better for transfer.

Attempting to solve a problem before being taught how to solve it — and failing — produces deeper learning once instruction arrives. The struggle prepares the ground.

The combination of spacing plus retrieval practice applied iteratively at widening intervals. The single strongest evidence-based study strategy in cognitive science — superadditive over either component alone.

Memory duration depends on encoding depth, not storage location. Shallow processing (surface features, phonetic) produces fragile traces. Deep semantic and associative processing produces durable ones.

Thinking about your own thinking. Specifically: knowing what you know, knowing what you don't, and calibrating your confidence against reality. The master skill that multiplies every other study skill.

The experience of fluent recognition being mistaken for the ability to retrieve and apply. Amplified by LLMs, because their articulate answers feel like comprehension even when no encoding has taken place.

Two modes of thinking. System 1 is fast, intuitive, automatic. System 2 is slow, deliberate, effortful. LLMs speak to System 1; durable learning happens in System 2.

The tendency to accept output from automated systems uncritically, especially when it is linguistically fluent. Amplified in LLMs, whose articulate style is mistaken for accuracy.

The gap between what a learner can do alone and what they can do with scaffolded guidance. Optimal learning happens inside this zone — too easy is nothing new; too hard is unreachable.

Temporary support structures that enable a learner to perform beyond their current independent capability, then are progressively withdrawn (*fading*) as the skill internalises.

The local, task-specific belief that you can succeed at this particular thing. The strongest single predictor of academic performance — stronger than IQ, study habits, or background.

The belief that abilities are expandable through effort and strategy, rather than fixed traits. Reshapes how learners respond to difficulty — as information rather than as judgement.

The state of complete absorption in a task when challenge and skill are balanced, goals are clear, feedback is immediate, and distractions are absent. The most productive cognitive state we know how to engineer.

Diagrams of concepts and their labelled relationships. Constructing one produces moderate-to-large learning effects, across disciplines. Studying a pre-built one produces almost nothing.

The human brain's native organisation is spatial, not sequential. Place cells and grid cells in the hippocampal–entorhinal system form an internal GPS that doubles as the substrate of memory.

Thinking is not a purely cerebral phenomenon. Body, gesture, motor action and environmental interaction are constitutive of cognitive processes, not accessories.

Writing by hand produces deeper encoding than typing — even when typed notes are more complete. The slowness of the pen forces compression, paraphrase and selection, which is where learning lives.

Human-plus-AI outperforms human alone and AI alone. The strongest learner is not the one who delegates to AI, nor the one who refuses it, but the one who uses it the way advanced chess uses engines.