LLMs and generative AI can be contrasted with applications and architectures that are retrieval-based.
Since the invention of the internet, information retrieval has been a central component for all the major tech products.
Google, Amazon, Facebook, etc are all applications that function as an interface and matchmaker between a user and a particular selection of information that has been retrieved. Websites, retail products, media content, etc.
In contrast, LLMs generate a unique synthesis of diverse sources of information.
RAG (retrieval augmented generation) represents a hybrid of this new generative capability with established retrieval methods.
Source: 2311.18550.pdf (arxiv.org)
(Mental Model)