Patrick and Bob discuss retrieval-augmented generation (RAG), its purpose, how it works, and its potential security and privacy issues, particularly with vector databases.
Resources:
- Security risks with vector databases: https://ironcorelabs.com/blog/2023/embeddings-not-human-readable/
- Vector database encryption: https://ironcorelabs.com/products/cloaked-ai/
0:00 RAG Architecture Security Risks with Bob and Patrick
00:19 What is RAG?
00:59 Why do we need RAG?
05:00 How RAG works
09:27 Alternatives to RAG
13:06 Vector databases
15:42 Vector DB risks
19:20 Mitigating vector risks
22:06 Risks with sensitive data in prompts
24:26 Hosted vs. local open-source LLMs
27:23 Other issues with RAG
#machinelearning #datasecurity #dataprivacy