Stanford's free RAG lecture goes beyond chunking: BM25, HNSW, and agents

stanfordnlp · x · 2026-11-23

Stanford NLP reshares Diyi Yang's Stanford Online lecture 'RAG Explained: From BM25 to Agent' (48k views), arguing most RAG tutorials stop at chunking and cosine similarity.

The lecture covers classical retrieval like BM25, HNSW hierarchical graph indexes (greedy top-down search, log N hops for high recall), RAG benchmark comparisons on open-domain QA, human evaluations showing RAG's factuality and specificity gains, and agentic retrieval.

Original post →

More from coding & agent

coding & agent channel →