Memories gradually losing minor details (leveling) while select remaining details get exaggerated (sharpening) with each retelling.
Imagine you heard a really long bedtime story, and then you tried to tell it to your friend. You'd probably forget a bunch of the boring parts and skip them, but the one really cool or scary part? You'd make that part even bigger and more exciting. That's leveling (dropping the boring stuff) and sharpening (pumping up the exciting stuff).
Leveling and sharpening describe complementary processes that reshape memories over time and through retelling. Leveling is the gradual loss of peripheral or seemingly unimportant details, making a recollection shorter, simpler, and more streamlined. Sharpening is the simultaneous tendency to selectively retain and amplify certain details, inflating their emotional weight or narrative importance relative to the now-absent context. Together, these processes transform rich, complex experiences into compact, schema-consistent narratives that feel coherent but may diverge substantially from what actually occurred.
The same glitch looks different depending on the terrain. Finance, medicine, a relationship, a team — same mechanism, different costume.
Investors tend to remember dramatic market crashes or windfalls in exaggerated detail while forgetting the long stretches of steady, unremarkable returns that constituted most of their portfolio history, leading to distorted risk perceptions.
Patients recounting symptoms to a doctor tend to drop routine or mild complaints (leveling) while amplifying the intensity and frequency of the most alarming symptom (sharpening), which can skew clinical diagnosis.
Frederic Bartlett first described leveling and sharpening as memory distortion processes in 'Remembering' (1932). Gordon Allport and Leo Postman formalized the concepts in their study of rumor transmission in 'The Psychology of Rumor' (1947). Philip Holzman and George Klein extended the framework to individual cognitive styles at the Menninger Foundation in 1954.
In ancestral environments, rapid communication of survival-relevant information — the location of a predator, the identity of a threat — demanded that stories be compressed to their essential elements and that critical details be emphasized. Stripping away irrelevant context (leveling) and amplifying danger signals or actionable information (sharpening) would have allowed groups to transmit high-priority warnings efficiently through oral chains.
Language models trained on internet text inherit leveling and sharpening patterns present in their training data: common, schema-consistent facts and narratives are over-represented (sharpened) while nuanced, rare, or contextual details are underrepresented or lost (leveled). This can cause LLMs to generate oversimplified or dramatized summaries that mirror the distortions of human oral transmission.
This page taught you the name. The deck turns the name into reflex. 1,100+ swipeable scenarios, 1,100+ defenses, 650+ detection prompts — spaced-repetition Swipe Deck, unlimited Spot-the-Bias Quiz, Defense Playbook, Pre-Flight, My Blindspots, Cheat Sheets, Field Guide e-book. $39.53$59.
