Evento acontece no dia 1º de julho, das 13h30 às 14h30, no Anfiteatro do CIn
O Centro de Informática (CIn) da UFPE realizará, no dia 1º de julho, o Seminário de Pesquisa “Beyond Code Generation: Reliability, Quality, and Evaluation in AI-Assisted Software Engineering”, que será ministrado por Léuson da Silva, pesquisador de pós-doutorado na Polytechnique Montreal, no Canadá. O evento acontecerá das 13h30 às 14h30, no Anfiteatro do CIn e não é preciso realizar inscrição para participar.
O seminário abordará os impactos da crescente adoção de ferramentas baseadas em Inteligência Artificial na Engenharia de Software, especialmente modelos de linguagem de grande porte (LLMs) e assistentes de programação. A apresentação discutirá como essas tecnologias têm transformado atividades cotidianas de desenvolvimento, como geração de código, testes, revisão de software e suporte a programadores.
Confira abaixo mais informações sobre o seminário e o palestrante:
Abstract:
Since the rise of AI-based tools, especially LLMs and, more recently, emerging agentic coding assistants, we have observed software engineering workflows entering a new era. Practitioners are increasingly relying on these tools for their daily activities to support different tasks, including answering programming questions, generating code, testing, reviewing code, and reasoning about software systems. However, such a growing adoption raises important questions: How reliable are these tools in practice? What kinds of quality issues appear in AI-generated code? And how should we evaluate these tools beyond simple measures of functional correctness? In this talk, I will discuss my recent empirical studies and collaborations investigating AI-assisted software engineering from three main perspectives. First, I will discuss how LLMs are changing developer support ecosystems, including their role in programming Q&A and developer assistance. Second, I will examine the quality of AI-generated code, focusing on inefficiencies, maintainability concerns, and the limitations of correctness-oriented evaluation. Finally, I will discuss studies on AI-assisted software engineering tasks such as testing, repair, and code review. Overall, AI-assisted software engineering should move beyond code generation alone. To effectively integrate AI into software development, we need empirical evidence, rigorous evaluation methods, and tools that account for reliability, quality, and practical usefulness in real development workflows.
Mini-bio:
Léuson da Silva is a Postdoctoral Fellow at Polytechnique Montréal (Canada), where he has been working with Prof. Foutse Khomh since 2023. He holds a Ph.D. in Computer Science from the Federal University of Pernambuco (Brazil), completed under the supervision of Prof. Paulo Borba, and continues to collaborate with the Software Productivity Group at UFPE. His research focuses on supporting software developers in their daily activities, particularly in areas such as code integration conflicts (including build and test conflicts) and code reviews, encompassing both SE4AI and AI4SE perspectives. His main research interests include Large Language Models, Mining Software Repositories, Collaborative Software Development, Empirical Software Engineering, and Software Evolution.
Comentários desativados