Empower AI: Building Smarter Solutions with Embeddings
summary
Claims a practical pipeline: use embeddings for retrieval to supply correct context to a large language model and stop hallucinations, with demos on internal wikis, invoices and podcast transcripts. Describes bilingual embedding models that are compact and fast, deployed on AWS sagemaker, and evaluated as top performers "in the wild" though not tuned for standard leaderboards. Notes 8K context length and vector DBs + RAG glue, and that models are open sourced on hugging phase.
AI-written summary · about
chapters (10)
- 0:00AI and embeddings intro
- 1:58LLMs on private data
- 4:26How LLMs process text
- 5:53What embeddings are
- 8:42Embeddings versus large language models
- 9:46Retrieval with embeddings
- 10:50Embedding use cases
- 12:35GI’s open-source embeddings
- 13:58Why GI’s embeddings
- 15:00Closing & Q&A
related talks
03mContextual understanding using Generative Artificial Intelligence
28m"Embeddings: How Computers Learned to Read"
27mData with Purpose: The Power of Embeddings
31mText and Code Embeddings
48mUnderstanding Embedding Vectors and the RAG Pattern in AI Solutions
27mUnlocking the Potential of Large Language Models
30mRAG: Tricks from the trenches.
01h 08mGoing multi-modal