Towards Leveraging LLMs for Reducing Open Source Onboarding Information Overload

Consistent, diverse, and quality contributions are essential to the sustainability of the open source community. Therefore, it is important that there is infrastructure for effectively onboarding and retaining diverse newcomers to open source software projects. Most often, open source projects rely on onboarding documentation to support newcomers in making their first contributions. Unfortunately, prior studies suggest that information overload from available documentation, along with the predominantly monolingual nature of repositories, can have negative effects on the newcomer experiences and onboarding process. This, coupled with the effort involved in creating and maintaining onboarding documentation, suggest a need for support in creating more accessible documentation. Large language models (LLMs) have shown great potential in providing text transformation support in other domains, and even shown promise in simplifying or generating other kinds of computing artifacts, such as source code and technical documentation. We contend that LLMs can also help make software onboarding documentation more accessible, thereby reducing the potential for information overload. Using ChatGPT (GPT-3.5 Turbo) and Gemini Pro as case studies, we assessed the effectiveness of LLMs for simplifying software onboarding documentation, one method for reducing information overload. We discuss a broader vision for using LLMs to support the creation of more accessible documentation and outline future research directions toward this vision.CCS CONCEPTS• Generalandreference→Experimentation; • Human-centered computing → Accessibility systems and tools; • Software and its engineering → Documentation.