Systems engineering issues for industry applications of large language model

Abstract

Large language model (LLM) is an important direction in the development of AGI, but its technology is still in rapid change, and its capabilities still have obvious deficiencies and imbalances, with persistent problems such as hallucination, value non-alignment, weak specialization, and black-box effect. In this case, how to apply LLM to different professional fields and develop high-quality AIGC industry applications has become a great challenge for ISVs. Building AIGC industry applications based on LLM is not simply a matter of functional realization. Although researchers and open-source communities have proposed numerous application development frameworks or tool components, there is a lack of overall architecture design for systems engineering and a lack of discussion on theories and methods of LLM application development in large-scale industry domains, such as healthcare, government affairs, finance, and media. This paper analyzes the basic ideas of LLM industry applications development, defines the functional requirements and feature requirements of LLM industry applications, puts forward the concept of Large Language Model Systems Engineering (LLM-SE), and develops an AI assisted clinical risk prediction system for amyloidosis disease based on the architecture of LLM-SE, which adopt knowledge engineering, quality engineering, etc., and verifies the LLM-SE development architecture and methodology.