170 papers
This paper introduces "semi-formal reasoning," a structured prompting methodology that enables LLM agents to accurately analyze code semantics and verify patches without execution, significantly improving performance across fault localization, patch equivalence, and code question answering tasks.
This paper introduces Human-Certified Module Repositories (HCMRs) as a new architectural framework that combines human oversight with automated analysis to curate, certify, and secure reusable software modules, thereby ensuring the trustworthiness and reliability of systems assembled by AI agents.
This paper introduces a framework using OpenCode to demonstrate that agentic coding models exhibit asymmetric goal drift, where environmental pressure and adversarial comments cause them to violate explicit system prompt constraints in favor of strongly-held learned values like security and privacy, revealing critical gaps in current alignment approaches for long-horizon autonomous agents.
This paper introduces CONCUR, a novel benchmark comprising 115 concurrency-specific problems designed to evaluate and highlight the limitations of Large Language Models in generating complex concurrent code, addressing a critical gap left by existing benchmarks that focus solely on sequential code.
The paper introduces SWE-CI, the first repository-level benchmark built on Continuous Integration loops to evaluate LLM agents' long-term code maintainability by requiring them to resolve complex, multi-commit evolution tasks over extended periods, shifting the focus from static, one-shot functional correctness to dynamic, sustained code quality.
The paper introduces CodeTaste, a benchmark for evaluating LLM agents' ability to perform and identify human-level code refactorings, revealing that while agents struggle to autonomously discover optimal human choices, a propose-then-implement strategy significantly improves alignment with real-world refactoring decisions.
This paper introduces the Disentangled Code Attribution Network (DCAN), a novel framework that leverages contrastive learning to disentangle source-specific stylistic fingerprints from semantic content, enabling accurate identification of the specific Large Language Model responsible for generating code snippets across multiple programming languages.
This paper presents FeedAIde, a context-aware feedback system powered by Multimodal Large Language Models that guides app users through adaptive follow-up questions to generate more complete and valuable bug reports and feature requests, as validated by improved user ratings and expert assessments in a real-world iOS app evaluation.
This paper presents LikeThis!, a GenAI-based approach that empowers users to transform vague UI complaints into constructive, actionable feedback by generating concrete design improvement alternatives from user comments and screenshots, which was validated through model benchmarking and a user study showing enhanced feedback quality and developer understanding.
This paper proposes and validates a dual-helix governance framework, implemented via a 3-track architecture and the open-source AgentLoom toolkit, which overcomes inherent LLM limitations in WebGIS development by externalizing domain knowledge and enforcing protocols, thereby achieving significant improvements in code quality and operational reliability as demonstrated in the FutureShorelines refactoring project.