Data
Model Drift
Model drift is the degradation of an AI model's accuracy over time as the real-world data distribution it is applied to shifts away from the distribution it was trained on. Drift is detected through statistical monitoring of model outputs and business KPIs, and corrected by retraining or retrieval-layer updates.
Data drift (input distribution changes) and concept drift (the relationship between inputs and correct outputs changes) require different remediation. LLM deployments experience drift primarily in the retrieval layer as organizational knowledge changes — regular knowledge-base updates are the first line of defense.
Related terms
- MLOps — MLOps (Machine Learning Operations) is the set of practices that operationalize ML and AI models in production — covering CI/CD pipelines for model updates, automated testing, performance monitoring, data versioning, and rollback procedures.
- Fine-Tuning — Fine-tuning is the process of training an existing AI model on additional task-specific data so its weights adapt to a narrower domain.
- Hallucination — A hallucination is when an AI model generates text that is fluent, confident, and factually wrong.