Retrieval Augmentation
Also called Retrieval-Augmented Generation
The general move of feeding a model fetched facts at answer time instead of relying on its memory.
Think of it like
Handing a contestant the reference sheet before the quiz rather than trusting what they crammed.
Example
Instead of fine-tuning a model on your policies, you retrieve the relevant policy text and paste it into the prompt each time.
How it actually works
Retrieval augmentation is the umbrella idea: rather than baking knowledge into weights, you fetch relevant information at inference and give it to the model as context. It keeps answers current and grounded, sidesteps expensive retraining, and localizes updates to the data. The trade is that quality now depends on retrieval, and every answer carries the token cost of its evidence.
For product teams
Update knowledge by editing docs, not retraining models — cheaper and faster to keep current.
For engineers
Inject fetched context at inference vs. parameterizing knowledge; shifts the quality bottleneck to retrieval.
Related
- RAG — RAG is the best-known instance.
- Hallucination — Reduces hallucination by grounding answers.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome