The team tested if AI systems that process text and images—known as multimodal large language models (MLLMs)—can answer time-related questions by looking at a picture of a clock or a calendar.