Context Rot

lastmod 2026-04-14created 2026-04-124 edits

Context Rot refers to the phenomenon in which a large language model (LLM) becomes less intelligent and less accurate at processing information as the context (amount of information) fed into it grows longer. It is caused by the 'Lost in the Middle' phenomenon, in which important information is missed within a long context, or by the dilution of the attention mechanism, and simply supporting long contexts does not guarantee performance.

See also

• LLM
• Prompt engineering
• Context engineering
• everything-claude-code - A hackathon winner's recipe for an AI development team
• Korean: 컨텍스트 부패 (Context Rot)
• Korean: AI 슬롭 (AI Slop)
• Korean: 불쾌한 골짜기 (Uncanny Valley)
• Korean: 할루시네이션 (Hallucination)
• Korean: 모드 붕괴 (Mode Collapse)

Favorite site

• Context Rot: How Increasing Input Tokens Impacts LLM Performance·|·Chroma
• The 'Context Rot' phenomenon in LLMs
• (AI) Context Rot: How Increasing Input Tokens Impacts LLM Performance(Context rot: the impact of increasing input tokens on LLM performance)
• The hidden pitfall of long-context LLMs: the Context Rot phenomenon and practical countermeasures - AI Sparkup