Towards Real-World Industrial-Scale Verification: LLM-Driven Theorem Proving on seL4
Summary
The paper introduces AutoReal, an LLM-driven approach for real-world industrial-scale formal verification, demonstrated on the seL4 project with a compact 7B prover achieving notable proof success. It highlights chain-of-thought training and context augmentation to improve reliability and shows applicability to security-focused verification, suggesting potential reductions in cost and improvements in deployability for critical software.