Non-parametric Knowledge

Non-parametric knowledge refers to explicit, structured knowledge existing outside the AI model. It includes documents, knowledge graphs, databases, or Bodies of Knowledge (BoKs) that can be retrieved as external sources. In RAG (Retrieval-Augmented Generation), the retriever accesses this knowledge to complement the generator’s parametric knowledge.

In SimpleModeling, non-parametric knowledge constitutes the BoK layer, serving as a structured and verifiable knowledge foundation for AI cooperation. In RAG-based knowledge augmentation, entities, relations, and rules defined in the BoK guide the LLM’s generative process.