LLM—Search Engine Killer?
summary
Claim: large LMs don’t replace search, they augment it. Concrete stack: OpenSearch + k‑NN for vector retrieval, miniLM embeddings, langchain4j to index PDF pages (recursive splitter) and a Llama‑8B CPU runner for streaming generation. Demo shows pure Llama hallucinates about a recent PAL paper, while RAG pulls the correct paper fragment and injects it into the prompt. Result: RAG materially reduces false answers but requires chunking, matching embedding models, and engineering to keep sources current.
AI-written summary · about
chapters (11)
- 0:00Introduction and disclaimer
- 1:58How search engines work
- 5:10Information retrieval basics
- 10:08First ML revolution in search
- 17:36Second ML revolution with LLMs
- 22:10Limitations of large language models
- 27:10Fixing LLM limitations with RAG
- 29:38Live demo setup
- 32:52Retrieval augmented generation in action
- 44:51Future of search with AI
- 45:13Q&A and closing
related talks
49mLLM—Search Engine Killer?
01h 30mUsing LLMs to improve your Search Engine
24mConversational Search with LLMs
01h 19mBuilding an LLM on your Laptop to...
15mSearch Engineering Reimagined: Leveraging LLM in ML Workflows
01h 49mAI engineering with open access LLMs that lie, curse, and steal
57mLLMs gone wild
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch