Seminars

NO.229 A Tale of Two Cities: Large Language Models and Computer Programming

Shonan Village Center

December 8 - 11, 2025 (Check-in: December 7, 2025 )

Organizers

  • Yuekang Li
    • University of New South Wales, Australia
  • Yulei Sui
    • University of New South Wales, Australia
  • Yang Cao
    • Hokkaido University, Japan
  • Ling Shi
    • Nanyang Technological University, Singapore

Overview

Overview

In recent years, the proliferation of Large Language Models (LLMs) has captivated the world with their remarkable language generation, comprehension capabilities, and their transformative impact across various domains. These models, typified by examples like GPT (Generative Pre-trained Transformer) series, Bard, and Llama series, have ushered in a new paradigm in natural language processing, enabling machines to comprehend, generate, and manipulate human-like text at an unprecedented scale and quality.

The emergence of Large Language Models (LLMs) has catalyzed a paradigm shift in software engineering, programming languages, and software security. This seminar aims to explore the intricate interplay between LLMs and computer programming by delving into their pivotal role in advancing software engineering and security practices, addressing programming language problems, and tackling the challenges associated with securing and analyzing LLM applications/frameworks. The meeting serves as a collaborative platform, inviting experts from diverse backgrounds to dissect the multifaceted landscape where LLMs intersect with programming, fostering knowledge exchange and innovation in response to evolving complexities.

Meeting Topics

  1. Leveraging LLMs in Software Engineering/Security: The integration of LLMs in software engineering/security opens avenues for:

Enhanced Static Program Analysis: Harnessing LLMs for refined static taint analysis, value-flow analysis, software component analysis, vulnerability scanning, symbolic execution, and code reviewing/auditing, elevating the precision and effectiveness of software analysis techniques.

Dynamic Program Analysis Enriched by LLMs: Exploring the incorporation of LLMs in dynamic analysis methodologies like fuzz testing, dynamic taint analysis, concolic execution, and penetration testing, empowering dynamic analysis in intricate software systems.

  1. Addressing Programming Language Problems with LLMs: LLMs offer novel solutions to various programming language problems:

LLMs in Code Generation: Unraveling the potential of LLMs in code generation and their influence on compiler optimizations, revolutionizing programming language implementation strategies.

LLMs and Compiler Optimizations: Analyzing the impact of LLMs on compiler optimizations, revolutionizing programming language design paradigms.

  1. Securing and Analyzing LLM Applications/Frameworks: Strategies to secure and analyze LLM applications/frameworks include:

Securing LLM Applications: Investigating methodologies to safeguard LLM applications against emerging threats like jailbreaking and prompt injection, leveraging diverse analysis techniques.

Analyzing LLM Frameworks: Exploring approaches to scrutinize LLM frameworks for vulnerabilities and ensure robustness in developing LLM-based applications.

Meeting Outcomes

The Shonan Meeting aims to foster interdisciplinary collaboration, sharing cutting-edge research findings, and outlining future directions in the domain of program analysis leveraging LLMs. The objectives are:

  • Unveiling and deliberating state-of-the-art methodologies leveraging LLMs in software engineering and addressing programming language challenges.
  • Exploring innovative techniques to fortify LLM applications and frameworks, addressing new complexities introduced by LLMs.
  • Encouraging collaboration among international experts across diverse disciplines, amalgamating theoretical insights with practical applications at the intersection of LLMs and computer programming.

Meeting Schedule

Over the course of four days, Day 1 will feature an introduction and overview session alongside a keynote presentation on the integration of Large Language Models (LLMs) with computer programming. Day 2 will focus on workshops, discussions, and presentations dedicated to exploring the role of LLMs in software engineering, encompassing static and dynamic program analysis techniques. Day 3 will delve into the application of LLMs in addressing programming language problems, concentrating on code generation, compiler optimizations, and their influence on programming language design. Finally, Day 4 will center on securing and analyzing LLM applications/frameworks, featuring workshops, in-depth case studies, and discussions on fortifying LLM-based applications while scrutinizing LLM frameworks for vulnerabilities and robustness.