Patterns for combining large language models with knowledge bases to improve assurance, performance, and reliability of AI solutions

While Large Language Models (LLMs) have proven themselves useful in many contexts, they also have drawbacks including hallucinations, non-repeatability and high latency inference. In many cases, a decision making based on knowledge bases may be better from an assurance, reliability, performance or repeatability perspective, but suffers from the high overhead required for maintaining and updating the knowledge base. In order to make smart dynamic decisions, the best approach may often consist of using a hybrid approach that combines the strengths of both LLMs and knowledge bases. This combination can be done in many different ways, and results in a few distinct patterns for joint use of LLMs and Knowledge bases. In this paper, we will present and examine some common patterns for joint use of LLMs and knowledge bases and compare their strengths and weaknesses. We discuss the impact on the security and assurance characteristics of AI based business services by combining both LLMs and Knowledge Bases.