April 24, 2025
Large Language Models (LLMs) like Codex and ChatGPT have demonstrated remarkable capabilities in automatic code generation, yet they frequently produce incorrect or suboptimal code, especially for complex tasks. In this talk, Yifeng will introduce Programming with Interactive Grounding (PING), a novel approach that enables bi-directional communication between developers and LLMs through editable inline code comments. By interleaving code generation, fine-grained comment generation, and iterative developer feedback, PING facilitates mutual grounding—establishing a shared understanding that aligns code with developer intent. This approach significantly improves code correctness across multiple LLMs and benchmarks and boosts developer productivity and confidence, as shown in both simulated and real-user studies. This talk will cover the motivation, design, implementation, and evaluation of PING, and demonstrate how it can empower more reliable, human-centered code generation workflows.
About Yifeng Di
Yifeng Di is a third-year Ph.D. student in Computer Science at Purdue University, advised by Professor Tianyi Zhang. His research focuses on code generation, large language models for programming tasks (LLM4Code), and software supply chain security. His recent work, Enhancing Code Generation via Bidirectional Comment-Level Mutual Grounding, will appear at ICSE 2025 and proposes a novel interactive framework to align LLM-generated code with developer intent using editable inline comments.