Definition
Knowledge Decay refers to the diminishing utility and accuracy of a RAG system's retrieved context as real-world facts evolve, rendering stored vector embeddings obsolete or factually incorrect. Managing it involves a critical trade-off between high-frequency re-indexing costs and the operational risk of the AI agent providing 'stale' or confidently incorrect information.
Refers to data obsolescence in the retrieval layer, not the neural 'catastrophic forgetting' seen during model training.
"A library where the science section hasn't been updated in a decade, leading students to cite debunked theories as current facts."
- Incremental Indexing(Mitigation Strategy)
- Vector Staleness(Direct Symptom)
- TTL (Time-to-Live)(Control Mechanism)
- Hallucination(Downstream Effect)
Conceptual Overview
Knowledge Decay refers to the diminishing utility and accuracy of a RAG system's retrieved context as real-world facts evolve, rendering stored vector embeddings obsolete or factually incorrect. Managing it involves a critical trade-off between high-frequency re-indexing costs and the operational risk of the AI agent providing 'stale' or confidently incorrect information.
Disambiguation
Refers to data obsolescence in the retrieval layer, not the neural 'catastrophic forgetting' seen during model training.
Visual Analog
A library where the science section hasn't been updated in a decade, leading students to cite debunked theories as current facts.