Unconference: Explainable Vector Search
summary
Claim: single-vector underperforms BM25 on complex logical queries, supported by a 50k document dataset with 1k queries where single-vector embedder trails keyword search and multi-vector models. Demonstration builds multi-vector indexing and late interaction similarity that returns token-level matches and multi-vector token-to-token highlights, making results auditable. Notes mention a Google paper arguing single-vector limits and industry debate that sparse encodings or graph structures may be needed for some relational queries.
AI-written summary · about
chapters (8)
- 0:00Vector search introduction
- 3:13Single vector limits
- 6:11Evidence against single vector
- 8:39Multi-vector advantages
- 9:21Demo: explainable multi-vector
- 11:19Hierarchical query relations
- 13:08Graphs and broader tools
- 14:29Closing and Q&A
related talks
41mGetting Started With Vector Databases
54mThe Debate Returns (with more vectors): Which Search Engine?
46mNatural Language & Search, Improve your Search w/ Vector capabilities
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
01h 09mVector Search: Ask Me Anything!
59mContextual search with vector search: exploring your options with open source tools
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
44mVector Search Isn’t Enough