Open-source Large Language Models for Secure System Transformation

This technical report discusses the application of open-source Large Language Models (LLMs) in the context of secure system transformation, particularly focusing on their use in interpreting legacy software source code. The authors, Gerhardt Scriven, Tony Moenicke, and Sebastian Ehrig from Capco, present an evaluation of open-source LLMs as alternatives to closed-source models like ChatGPT. The report emphasizes the importance of these models for financial institutions aiming to modernize their legacy systems while ensuring data security. It outlines how these models can enhance workflow automation, improve data analysis capabilities, and facilitate internal communication. Additionally, the report highlights the significance of the decision between using third-party hosted services versus local hosting of LLMs, noting that local hosting offers better control over data privacy and customization to meet specific business needs. The findings aim to guide organizations in leveraging LLMs effectively to drive efficiency and productivity across various business functions.