In a study by researchers at the University of Berkeley, young children outperformed large language models like ChatGPT in problem-solving tasks requiring innovation with tools. The experiment revealed that children aged three to seven could successfully choose the correct tool for a task about 85% of the time, while AI models struggled to match their success rates. The authors emphasized that large language models excel at imitation but lack the ability to introduce novel ideas or concepts, showcasing a significant gap between human and artificial intelligence capabilities. To enhance AI systems' innovation potential, researchers suggest emulating how children learn through active exploration and curiosity-driven interventions.
翻譯成其他語言
從原文內容
bigthink.com
從以下內容提煉的關鍵洞見
by 於 bigthink.com 02-26-2024
https://bigthink.com/the-present/large-language-models-lose-to-children/深入探究