A new arXiv paper introduces Eiffel-tools, a Language Server Protocol implementation for Eiffel that uses LLMs to generate code and specifications for statically verified software. The tool directs LLMs with language- and project-specific knowledge, crafts rich programmatic prompts, and verifies output with a static verifier. It matters to practitioners combining LLM code generation with formal methods.
A new arXiv study presents a comprehensive empirical analysis of LLM-based automated program repair (APR), examining how repair performance is affected by bug complexity, fault localization, reasoning settings, and cost. The findings aim to guide practitioners in choosing effective and cost-efficient APR strategies.
A new arXiv study presents a comprehensive empirical analysis of LLM-based automated program repair (APR), examining how repair performance is affected by bug complexity, fault localization, reasoning settings, and cost. The findings aim to guide practitioners in choosing effective and cost-efficient APR strategies.
A new arXiv paper partially replicates the FeedbackEval benchmark, which evaluates how LLMs use different feedback types for Python code repair, and adds an exploratory Java extension. The study targets concerns about reproducibility and generalizability of LLM-based automated program repair results.
A new arXiv paper partially replicates the FeedbackEval benchmark, which evaluates how LLMs use different feedback types for Python code repair, and adds an exploratory Java extension. The study targets concerns about reproducibility and generalizability of LLM-based automated program repair results.
An arXiv paper proposes Program Executability Prediction (PrEx), a task asking LLMs to judge whether a program is semantically valid or invalid. The study probes whether models use explicit programming-language semantics when given them or fall back on pretraining priors, addressing a key limitation in code-related LLM performance.
An arXiv paper proposes Program Executability Prediction (PrEx), a task asking LLMs to judge whether a program is semantically valid or invalid. The study probes whether models use explicit programming-language semantics when given them or fall back on pretraining priors, addressing a key limitation in code-related LLM performance.
Specula is a push-button agentic system that uses LLM-based coding agents to autonomously generate TLA+ specifications and invariants for large system code, enabling effective model checking and bug finding without manual formal methods expertise. It eliminates a key barrier to applying formal verification to real-world systems.
Specula is a push-button agentic system that uses LLM-based coding agents to autonomously generate TLA+ specifications and invariants for large system code, enabling effective model checking and bug finding without manual formal methods expertise. It eliminates a key barrier to applying formal verification to real-world systems.
A tool that verifies whether code behaves as intended, not just passes linting. It uses LLMs to compare code against a natural-language description of intent and outputs evidence of correctness or failure. Helps developers catch semantic bugs early.
A tool that verifies whether code behaves as intended, not just passes linting. It uses LLMs to compare code against a natural-language description of intent and outputs evidence of correctness or failure. Helps developers catch semantic bugs early.