Full-stack verification is the new trend (wiki.alcidesfonseca.com)

Recent advancements in full-stack verification have emerged, significantly impacting the AI and machine learning communities. Companies like Harmonic, SF-Tensor, &Truth, Theorem, and AutoResearch.inc have announced groundbreaking formal verification initiatives, leveraging AI for both software and hardware systems. For instance, Harmonic's Aristotle agent effectively identified and rectified bugs in a C library, while SF-Tensor formalized NVIDIA's PTX language in Lean to confirm the accuracy of GPU kernels. The collaborative efforts of Theorem and Anthropic culminated in rigorous verification processes for open-source software on RISC-V processors, demonstrating complex properties like idempotence and equivalence to reference versions. The trend towards full-stack formalization—encompassing comprehensive verification across software and hardware stacks—marks a paradigm shift from previous practices that were limited to core components or simplified models. This broader scope enhances robustness and reliability in AI-generated software, proving critical concepts such as correctness and security. However, challenges remain with using languages like Lean for production applications, including a steep learning curve for developers and potential security vulnerabilities associated with memory-unsafe languages like C. As the community pushes forward with these formal verification techniques, the integration of AI capabilities promises to further streamline the verification process and bolster software reliability in complex AI systems.