Large Language Models as a Service: Optimisation Strategies via Knowledge Space Reduction

In Large Language Models (LLMs), “reducing the domain space” for text generation refers to limiting the range of content to be utilized for generating responses. This work explores Sequential Language Model Integration (SLMI), which mirrors the organization and distribution of knowledge across different fields of expertise. SLMI is the technique of linking multiple LLMs in a systematic manner and creating LLM-chains. In this paper, we refer to a process of choosing, linking and connecting LLMs with other services (invoke external machine learning processes or rule based applications) as “Large Language Models as a Service” to describe a complete system with specific agency. We outline the development and evaluation process of an SLMI methodology to refine response accuracy and we evaluate the methodology in the domain of dermatology using medical knowledge paths, where gains were found regarding diagnostic accuracy.