A study from The Ohio State University reveals that large language models (LLMs) like ChatGPT struggle to grasp the concept of flowers as humans do due to their lack of sensory experiences. Lead researcher Qihui Xu explained that LLMs, trained solely on text and images, cannot smell, touch, or interact with flowers, affecting their ability to fully represent them. The research compared human understanding with LLMs (OpenAI’s GPT-3.5, GPT-4, and Google’s PaLM and Gemini) through two measures: Glasgow Norms, focusing on emotional and visual ratings of words, and Lancaster Norms, assessing sensory and motor experiences. While LLMs correlated well with human concepts overall, they faltered in areas linked to sensory information, emphasizing the richness of human experience. Despite these limitations, Xu noted that LLMs are improving and could become more adept at capturing human emotions when integrated with sensor data and robotics. The study was published in Nature Human Behavior.
Study Finds ChatGPT and Gemini AI Struggle to Comprehend the Concept of Flowers
