Exploring LLMs in Software: A Fresh Look

Large Language Models, or LLMs, have been making waves lately. They're not just standalone tools; they can also be crucial parts of software systems. These LLM-integrated applications use LLMs to do tasks that would be tough or time-consuming without them. As this field grows, we need clear terms and methods to understand it better. This study offers a helpful way to categorize these applications, called a taxonomy. It breaks down how LLMs can be used in software and how to implement these integrations.By examining recent LLM-integrated applications, we found some key points. Many of these apps use LLMs in multiple ways for different jobs. We call these pieces "LLM components. " To get a good picture of an app's architecture, we look at each component separately. We found thirteen ways to describe these components, like the skills used and the output format. This taxonomy helps us see how LLM-integrated applications are built. It can help researchers and developers come up with new ideas and overcome challenges. Integrating LLMs may change how software is made in the future.