Lean and AI caught a bug in my 2-year-old TLA+ spec (medium.com)

A recent exploration highlights the intersection of formal methods and AI assistance in software development, specifically examining TLA+ and Lean. The author reflects on Stephen Diehl's article, which discusses the potential for AI to aid in writing a mathematical library using Lean, enhancing the formal verification process across the software lifecycle. While the Lean code is noted to be more low-level than TLA+, it allows for the translation of high-level specifications into concrete implementations with verification capabilities. This transition emphasizes a shift where AI can assist in generating both high-level specifications and lower-level code, which might require minimal human oversight for refinement proofs. This development is significant for the AI/ML community as it suggests a future where automated processes can address complex verification tasks, facilitating the design of more reliable software. The challenges identified, such as ensuring that the properties proved correspond to the desired specifications, underline the importance of human involvement in crafting accurate specifications. The exploration indicates a duality of strengths and weaknesses between TLA+ and Lean, suggesting a potential framework where a high-level spec is linked to low-level implementation, thus paving the way for a new paradigm in software engineering: formal software creation with AI guidance, directed by human-defined specifications.