AI in Open-Source Software Maintenance

This chapter examines the role of language models in open-source software maintenance, highlighting the ethical concerns surrounding AI training data and biases. It emphasizes the importance of transparency and healthy human-AI interactions while acknowledging the limitations of AI compared to human responses.

Transcript

chevron_right

Play full episode

chevron_right

Transcript

Episode notes

Guest

Devin Stein

Panelist

Richard Littauer

Show Notes

In this episode of Sustain, host Richard Littauer talks with Devin Stein, CEO and Founder of Dosu. Today, they discuss the challenges of sustaining open source software, the role of AI and LLMs (Large Language Model) in automating support and maintenance, and the ethical considerations surrounding AI usage. Devin explains Dosu's approach to creating a living knowledge base to assist engineering teams and open source maintainers. Also, the conversation dives into how Dosu interacts with users, maintains quality control, and addresses the environmental impact of AI. Hit download now to hear more!

[00:01:43] Devin discusses Dosu’s purpose which helps engineering teams’ software by structuring engineering knowledge into a knowledge base, and the tool supports open source by addressing common questions, triaging issues, and identifying project ownership.

[00:02:46] We hear about how Dosu uses LLMs to create a “living knowledge base” that supports open source workflows, such as issue resolution and knowledge sharing.

[00:04:48] Devin explains that Dosu is focused on automating support tasks, not generating code directly, and he fills us in the user base and funding.

[00:06:17] Devin tells us that revenue comes from platform teams and open core companies using Dosu internally and through a per-seat pricing model.

[00:08:03] We learn how Dosu aims to reduce maintainer burnout by handling repetitive inquires, allowing maintainers to focus on unique issues.

[00:10:38] There’s a discussion on users’ positive reception to fast responses via Dosu and how Dosu aims to assist, not replace maintainers, providing first-pass answers or guidance.

[00:12:00] Richard expresses a “net positive” sentiment but admits to initial scepticism about GitHub Actions and automation in open source. Devin shares a similar story of entering open source for community interaction, initially contributing through GitHub, and receiving positive feedback.

[00:14:49] Richard inquires about managing customer expectations for accuracy and Devin acknowledges the challenge and explains that Dosu is designed to adapt by learning from past issues and solutions, and how human-in-the-loop workflows help maintainers refine Dosu’s responses.

[00:18:19] A question on ethical and legal use of LLMs is brought up, as Devin hopes for more transparency and alignment on LLM licensing and legal frameworks in the future.

[00:21:14] Devin explains that Dosu’s knowledge base will soon be accessible, providing transparency for users and maintainers about its data sources.

[00:24:49] Richard questions about how AI companies are ensuring their models don’t reinforce these biases and asks about measures in place to improve AI responses. Devin emphasizes their approach to LLMs, which focuses on treating the AI as a tool rather than imitating human behavior.

[00:26:55] The topic of addressing human elements and consistency is brought up and Devin explains that Dosu’s design keeps responses consistent and supportive, and maintainers and users can provide feedback and adjust responses to align with community needs.

[00:31:23] Devin talks about Dosu’s strategy focusing on helping become contributors without taking over human roles in open source, and maintainers still have the primary role in guiding substantial project changes or complex contributions.

[00:33:34] Devin acknowledges the environmental concerns around AI usage and hopes for more sustainable practices and optimizations in the future.

[00:34:30] Find out where you can follow Devin and Dosu online.