retrieval

Definition

Fetching relevant passages from an external store — documents, a vector index, a database — so the model can condition on them. It is the R in RAG, and the usual motivation is knowledge the weights lack: recent events, private documents, anything that changes faster than you retrain. Unlike a search engine, the results are selected for a model to consume, so chunking and ranking are tuned for a context window rather than for a human reader.

Explainers using this term

• Paper Walkthrough — Beyond Retrieval: LatentStream Turns Retrieved Video Into Latent MemoryBeyond Retrieval: Progressive Latent Memory Evolution for Streaming Video Understanding
• Paper Walkthrough — RoboTok: Mining the Web for Demonstrations That Move Like YoursRoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning
• Paper Walkthrough: DART-SD — Training Tool-Calling Agents Without Flattening the DiamondDART-SD: Diamond-topology Aware Retrieval and Tuning for Self-Distillation of Multi-Turn Tool-Calling Agents