On the Construction of Database Interfaces Based on Large Language Models

Abstract

: This paper argues that Large Language Models (LLMs) can be profitably used to construct natural language (NL) database interfaces, including conversational interfaces. Such interfaces will be simply called LLM-based database (conversational) interfaces. It discusses three problems: how to use an LLM to create an NL database interface; how to fine-tune an LLM to follow instructions over a particular database; and how to simplify the construction of LLM-based database (conversational) interfaces. The paper covers the first two problems with the help of examples based on two well-known LLM families, GPT and LLaMA, developed by OpenAI and Meta, respectively. Likewise, it discusses the third problem, with the help of examples based on two frameworks, LangChain and LlamaIndex.