Consider two language model systems. System A is provided with a set of relevant documents from a database at the same time it receives a user's query, before it begins generating a response. System B, in the middle of generating a response, pauses to execute a function that fetches real-time data, then uses that data to complete its answer. From a modeling perspective, what is the most accurate description of the fundamental goal shared by both systems?

Architectural Solutions for Context-Aware AI

The fact that Retrieval-Augmented Generation (RAG) gathers information before inference begins, while tool use accesses functions during inference, means that from a modeling perspective, they serve fundamentally different purposes in enhancing language model outputs.

The Agentic Perspective on Context Generation