SmartFAQs.ai
Back to Learn
Intermediate

Recency Bias

The tendency of Large Language Models to disproportionately favor information located at the end of the context window or, in RAG architectures, the prioritization of chronologically recent documents at the expense of historical accuracy. This often requires architectural trade-offs between 'freshness' and 'comprehensive context' to prevent the neglect of foundational data.

Definition

The tendency of Large Language Models to disproportionately favor information located at the end of the context window or, in RAG architectures, the prioritization of chronologically recent documents at the expense of historical accuracy. This often requires architectural trade-offs between 'freshness' and 'comprehensive context' to prevent the neglect of foundational data.

Disambiguation

Not a social psychology concept, but a positional token weighting or metadata filtering priority in AI models.

Visual Metaphor

"A conveyor belt where only the items closest to the exit are picked up, while vital components further back are ignored."

Key Tools
LangChain (LongContextReorder)LlamaIndex (Metadata Filters)PineconeWeaviate (Temporal Reranking)
Related Connections

Conceptual Overview

The tendency of Large Language Models to disproportionately favor information located at the end of the context window or, in RAG architectures, the prioritization of chronologically recent documents at the expense of historical accuracy. This often requires architectural trade-offs between 'freshness' and 'comprehensive context' to prevent the neglect of foundational data.

Disambiguation

Not a social psychology concept, but a positional token weighting or metadata filtering priority in AI models.

Visual Analog

A conveyor belt where only the items closest to the exit are picked up, while vital components further back are ignored.

Related Articles