Early Evidence of Vibe-Proving with Consumer LLMs: A Case Study on Spectral Region Characterization with ChatGPT-5.2 (Thinking)
Researchers used an iterative 'generate, referee, and repair' workflow with a consumer LLM to resolve a conjecture about the nonreal spectral region of specific 4-cycle row-stochastic matrices. The LLM excelled at proposing global structure, while human experts were crucial for verifying correctness. This study shows that AI can assist in high-level proof search but requires human oversight for critical verification.