Automated Verification with LLMs for Large Programs
Published:Dec 31, 2025 03:31
•1 min read
•ArXiv
Analysis
This paper addresses the challenge of verifying large-scale software by combining static analysis, deductive verification, and LLMs. It introduces Preguss, a framework that uses LLMs to generate and refine formal specifications, guided by potential runtime errors. The key contribution is the modular, fine-grained approach that allows for verification of programs with over a thousand lines of code, significantly reducing human effort compared to existing LLM-based methods.
Key Takeaways
- •Preguss is a framework for automated formal specification generation and refinement.
- •It combines static analysis, deductive verification, and LLMs.
- •It uses potential runtime errors to guide the process.
- •It enables verification of large-scale programs (over 1000 LoC).
- •Significantly reduces human verification effort compared to other LLM-based approaches.
Reference
“Preguss enables highly automated RTE-freeness verification for real-world programs with over a thousand LoC, with a reduction of 80.6%~88.9% human verification effort.”