RETRIEVAL-AUGMENTED GENERATION (RAG)

CHECK YOUR UNDERSTANDING • In one sentence, what does RAG do? ◦ Answer: It retrieves relevant passages from a document collection and gives them to the model so its answer is based on real sources. • Why can a vector search find an article that shares no words with the query? ◦ Answer: It compares embeddings that represent meaning, so similar meanings are close even when the wording differs. • What goes wrong when chunks are too small or too large? ◦ Answer: Too small and they lack context, too large and they mix topics, blur the embedding, and waste context space. • Define precision and recall for retrieval. ◦ Answer: Precision is the share of retrieved chunks that were relevant, and recall is the share of all relevant chunks that were retrieved. • Name two ways RAG can still fail to prevent hallucination. ◦ Answer: Retrieval may fetch the wrong passages, and the model may ignore or misread the passages or blend in its own guesses.