Productivity Applications of Language Intelligence Modeling in Domain Specific Research

Abstract

Abstract The popularity of large language models (LLMs) has led to their widespread adoption in various domains for diverse tasks. These models, characterized by their vast number of parameters, show advanced capabilities like solving complex problems and multi-step reasoning. Despite potential risks, they are seen as promising for enhancing productivity and workflows in numerous areas. The aim of this project is to integrate LLMs into research workflows in the automotive safety domain to streamline processes. It focuses on assessing and selecting a suitable open-source LLM for this purpose, as well as implementing capabilities such as semantic search and automatic insight generation from the given data. For the complex question answering, a retrieval augmented generation (RAG) pipeline is implemented, which is then shown to be a viable approach to exploiting the capabilities of large language models. The key deliverable is a proof-of-concept demonstrating the practical application of LLMs in processing and analyzing domain-specific data. In this project it is shown that the open-source LLM can have a significant role in enhancing the productivity of domain specific research. Also, by comparing the performance with GPT-3.5 Turbo, which is a proprietary model and has higher costs, it can be seen that the open-source model provides a competitive performance.