MoHoBench: Assessing Honesty of Multimodal Large Language Models via Unanswerable Visual Questions Paper • 2507.21503 • Published Jul 29, 2025 • 3