Research in the Age of AI
summary
Elasticsearch is presented shifting from BM25-only search to hybrid lexical+semantic retrieval by indexing embeddings and running k-NN over high-dimensional vectors (demo uses 2048‑dim audio embeddings and notes a 99.9999% self-match bug). The talk covers using HNSW with an automatic HNSW-to-brute-force switch for performance, support up to 4096‑dim vectors, commercial text‑expansion via Elser, and Retriever-Augmented Generation patterns that integrate OpenAI/LLMs while warning about memory and ingestion costs.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 3:11Classical vs semantic search
- 10:10Vectors and embeddings
- 24:55Audio demo
- 33:59Data indexing
- 36:47LLMs and RAG explained
- 37:09RAG demo
- 40:01Conclusion
related talks
38mSearch: A New Era
37mLexical, Vector & Hybrid Search with Elasticsearch
41mResearch in the Age of AI
01h 48mInformation Retrieval from the Ground Up
46mNatural Language & Search, Improve your Search w/ Vector capabilities
13mA Beginner’s Guide to Semantic Search
44mVector Search Isn’t Enough
32mRetrieval Augmentation and Semantic Search at Scale