The Mind-Level of Large Language Models: Cognitive and Affective Perspectives
1. Department of Psychology, Nanjing University, Nanjing,210023, China;
2. Department of Radiology, Nanjing Drum Tower Hospital the Affiliated Hospital of Nanjing University Medical School, Nanjing, 210008, China;
3. State Key Laboratory of Media Convergence Production Technology and Systems, Beijing,100083, China
Abstract：Large Language Models (LLMs) have demonstrated remarkable capabilities in language understanding and generation, which has sparked debate regarding their possession of human-like minds or consciousness. However, previous analyses of their mental capacity have often been unidimensional and insufficient. This paper proposes a systematic review and a multi-faceted perspective that integrates reasoning, social cognition, and emotional analysis to assess the performance of LLMs. Our review summarizes that LLMs approach or even surpass human performance in the majority of tasks, but exhibit deficiencies in deductive reasoning, story analogy, commonsense reasoning, and aspects of irony. Future research could benefit from adopting the framework constructed in this paper, combining it with multi-dimensional quantitative metrics to provide a potential evaluation scheme for the development of a comprehensive mind-assessment system.
汪小雪，周仁来. 大语言模型的心智水平：认知和情感的视角[J]. 应用心理学, 0, (): 1-.
WANG Xiaoxue，ZHOU Renlai. The Mind-Level of Large Language Models: Cognitive and Affective Perspectives. 应用心理学, 0, (): 1-.
浙公网安备 33010602002302号 备案号：