External input is knowledge beyond a model’s embedded training data, such as recent events, proprietary company information, specialized domain material, or real-time updates. section 1.5.1
Use external-input requirements to choose an optimization path:
- Require retrieval-augmented approaches when the task depends on information outside the model’s pretrained knowledge. section 1.5.1
- Avoid retrieval when general language tasks, creative writing, or basic reasoning can rely on pretrained knowledge. section 1.5.1
- Add personalization or fine-tuning when the application requires consistent style, tone, format, or deeper domain reasoning. section 1.5.1


