Glossary · term54 of 202
context rot
Measured accuracy decline as a model's input grows longer, even when the extra tokens are irrelevant, showing up well before the advertised context limit is reached.
also written as context degradation
Related terms
F1 scoreA single accuracy metric combining precision and recall, used in some of the studies cited to show how model performance falls as context length increases.view termNoLiMaA long-context benchmark that removes lexical overlap between questions and answers, measuring how far model accuracy falls as input length grows past a short baseline.view termcontext poisoningA failed attempt or wrong answer left sitting in a conversation keeps competing for attention on every later turn, dragging output back toward the same mistake even after correction.view termdistractor sensitivityA model's tendency to lose accuracy as irrelevant or off-task tokens accumulate in its context, even when the genuinely relevant information is still present.view termeffective lengthThe context length at which a model's benchmark accuracy actually holds up, often far shorter than its advertised maximum window size.view termlost in the middleA pattern where models recall information best when it sits at the start or end of the context window and recall it worst when it sits in the middle.view termmulti-turn driftMeasured accuracy loss that occurs when the same task is spread across several conversational turns instead of arriving in one clean prompt, distinct from length-driven degradation alone.view term
At a glance
- cited by
- 1 post
- categories
- 1
- first used
- jul 2026
