Separating Coherence from Knowledge with Retrieval-Augmented Generation
14 Jun 2023
Generative models like ChatGPT excel in coherence but model knowledge “implicitly” and struggle in out-of-domain settings. Retrieval Augmented Generation (RAG) tackles this by anchoring generative LLMs in domain-adapted semantic search, which also solves for privacy and sensitivity. Additionally, domain-adapted verification and entailment models reduce hallucinations to improve user experience. These best practices are essential to integrating a generative model that users can trust to better inform and supercharge their workflows in today’s complex enterprise environments.