Project Grant 2540626

Federal Grant Award Summary

George Mason University received a $674,100 Project Grant awarded July 15, 2026, under the Computer and Information Science and Engineering program (CFDA 47.070) from the National Science Foundation's Division of Computing and Communication Foundations. This CAREER award funds research and educational initiatives focused on advancing artificial intelligence (AI) interpretability for language models (LMs) applied to code generation tasks. The project delivers three primary research outputs: (1) a scalable mechanistic LM interpretation framework that identifies and analyzes only the most relevant LM mechanisms for downstream applications while leveraging LM agents to assist the interpretation process; (2) systematic investigation of LM mechanisms for code generation, including mechanisms for syntactic rule application, semantic knowledge recall, and deliberate reasoning, with examination of how data and learning paradigms impact these mechanisms; and (3) interpretation-informed methods and techniques to improve code generation quality by translating interpretability insights into practical improvements.

The award integrates research outcomes into comprehensive educational initiatives spanning AI literacy outreach programs, undergraduate early research opportunities, new curriculum development, and open-source educational resources. The project runs through June 30, 2031, and will be conducted at George Mason University's Fairfax, Virginia campus. This CAREER-track investment addresses critical gaps in LM scalability and actionability within existing AI interpretability approaches, with particular relevance to structured prediction tasks in data analytics, information acquisition, and software development.