Conference: Pushing Towards Open-Source AI

Award ID: 2335774

This workshop will bring together researchers and practitioners to understand the key challenges in building a robust, open-source ecosystem for generative AI. The technological systems that underlie generative AI differ from other open-source systems and present a unique set of issues. This workshop serves a niche that is not being fulfilled in either the open-source or academic machine learning community alone, and will connect researchers and open-source developers to specifically target core shared challenges. The outcomes of the workshop will serve as a roadmap to foster open-source AI that is safe and equitable and can be deployed to increase American economic growth and worker productivity. Open-source software development contributes to enormous growth in diverse industries across the world. The goal of this workshop proposal is to study how to foster a robust open-source ecosystem for generative AI that is comparable to the general open-source software ecosystem. The technological systems underlying generative AI present novel and complex issues that make it non-trivial to adapt current open-source best practices. Successes of generative AI are also not primarily due to code; they are the product of several factors, including: carefully coordinated data curation, strategically coordinated training runs, tuning with large-amounts of human feedback, and rigorous evaluation on realistic use-cases. The workshop will focus on the following four themes to define and address the core challenges of open-source generative AI: model adaptation for a broader range of users; open ecosystems for human feedback; evaluation of ethical, safe, and accurate systems; and supporting decentralized AI development. The workshop will strive to identify the challenges and opportunities in open-source models for AI development that will serve as a roadmap for the coming years. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

NSF Program Director: Peter Atherton