Semantic vs keyword search as context for GPT
summary
Compares semantic/vector search to classic keyword (BM25) search as the step that selects documentation context for GPT and finds accuracy is essentially a draw, but keyword wins for tunability and explainability. Implements keyword queries generated by ChatGPT, takes the top three results with highlights, or uses paragraph-level embeddings and cosine similarity for vector search. Builds an "ask your docs" chatbot on Zeta using OpenAI APIs; it works but hallucinates whenever the docs lack the answer; cost is about one cent per query on GPT-3.5-turbo.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 1:05Demo of chatbot on docs
- 5:47How chatbots on docs work
- 8:00Context size and pricing
- 10:30Keyword vs vector search
- 12:36Accuracy comparison
- 14:04Tunability of keyword search
- 16:31Improving docs with chatbot
- 17:36Q&A session
related talks
30mHow to Decipher User Uncertainty with GenAI and Vector Search
03mContextual understanding using Generative Artificial Intelligence
34mEnter the Brave New World of GenAI with Vector Search
15mSemantic programming by example with pre-trained models
41mGenAI, vector/semantic/hybrid search, RRF, NLP, (...) and other buzzwords
49mEnter the Brave New World of GenAI with Vector Search
01h 01mHow ChatGPT Works
39mExperiments with OpenSearch and AI