Translated data: The latest research indicates that GPT-4 achieves only a 33% accuracy rate in visual reasoning tasks, drawing attention to its capabilities in graphic processing. Researchers tested the model using the ConceptARC dataset, revealing that humans average a 91% accuracy on the same tasks, significantly outperforming GPT-4. The study's methodology has sparked skepticism, including issues with participant recruitment and input methods, highlighting the limitations of large language models in certain tasks and calling for a thorough review of the research methods.