Prof. Jeremy Bradbury gave a University of St Andrews School of Computer Science Seminar Series this morning on “Democratizing AI in Software Development: From Few-Shot Testing to Trustworthy Benchmarks and Accessible AI Tools.” His talk focused on three key challenges to broader AI adoption in software development and how recent SEER Lab work is addressing them:
🔹 𝗟𝗶𝗺𝗶𝘁𝗲𝗱 𝗰𝗼𝗺𝗽𝘂𝘁𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗮𝗻𝗱 𝗱𝗮𝘁𝗮 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀
💡 Applying few-shot learning to detect flaky tests with minimal training
🔹 𝗧𝗿𝘂𝘀𝘁 𝗶𝗻 𝘁𝗵𝗶𝗿𝗱-𝗽𝗮𝗿𝘁𝘆 𝗺𝗼𝗱𝗲𝗹𝘀
💡 Evaluating data leakage in LLM program-generation benchmarks and designing more reliable alternatives
🔹 𝗕𝗮𝗿𝗿𝗶𝗲𝗿𝘀 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗔𝗜-𝗲𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁
💡 Using block-based programming to make AI tools more accessible
Together, these efforts aim to make AI-driven software development more accessible, trustworthy, and usable for a broader range of developers and organizations.
Slides: [PDF]
Prof. Bradbury’s visit to the University of St Andrews was hosted by his research collaborator, Prof, Miriam Sturdee, and their co-supervised PhD student Stacey Koornneef. During his visit they collaborated on their joint research at the intersection of creativity and computational thinking education. This research is interdisciplinary in nature and includes elements of education, creativity, human-computer interaction and software development.
