This research evaluates how Large Language Models (LLMs) can assist in creating formal software specifications in Alloy, addressing a critical challenge in building secure, dependable systems.

• LLMs can effectively translate natural language requirements into formal Alloy specifications
• Models can generate alternative specifications to enhance robustness and completeness
• Performance varies based on specification complexity and model capabilities
• Error rates decrease significantly when LLMs are properly prompted

For engineering teams, this research opens pathways to improve specification quality, reduce development time, and build more reliable software systems by complementing human expertise with AI capabilities.

On the Effectiveness of Large Language Models in Writing Alloy Formulas