A Survey on Domain-Specific Large Language Models

Abstract

Large language model has sparked a new wave in generative AI for its outstanding semantic understanding and text generation capabilities. Although it performs excellently in various general knowledge tasks, it often suffers from factual hallucinations and faithfulness hallucinations when applied to vertical domains such as military, law, healthcare, and education. Constructing domain-specific LLMs for vertical domain applications can effectively address the hallucination issue. This paper first sorts out the typical applications of domain-specific LLMs in various fields, then summarizes existing optimization paradigms such as continual pre-training and prompt engineering in the order of the workflow for constructing domain-specific LLMs, and finally discusses the challenges faced by domain-specific LLMs and future research directions.

Ähnliche Arbeiten