New research presented at the 2025 International Conference on Learning Representations reveals that leading artificial intelligence models, including those by Meta, Anthropic, Google, and OpenAI, struggle to read analogue clocks and interpret calendars. In tests, AI models correctly read the time only 38.7% of the time and identified calendar dates correctly just 26.3% of the time. The study highlights fundamental deficits in spatial and logical reasoning as well as insufficient training data for these tasks. Researchers stress that these weaknesses must be addressed for AI to function reliably in real-world, time-sensitive applications, demonstrating that some everyday cognitive tasks remain challenging for even the most advanced AI systems.
Related articles:
OpenAI and DeepSeek AI Models Collapse on Complex Problems, Study Finds
New Study Suggests AI May Interpret Emotions More Accurately Than People
AI Outsmarts Top Mathematicians in Secret Test
Major Differences Found in Human and AI Reasoning, Scientists Report
AI Hallucinates Frequently, But Researchers Offer Solutions
