The rapid advancement of artificial intelligence (AI) has spurred a surge of interest in large language models such as ChatGPT. However, a recent paper published on the preprint platform arXiv by a team of researchers from Arizona State University challenges our understanding of these AI models. The study suggests that these models do not actually engage in real thinking or reasoning, but merely seek correlations.

The researchers argue that although AI models often generate a series of seemingly logical intermediate processes before providing answers, this does not equate to reasoning. They emphasize that anthropomorphizing AI model behavior can lead to misunderstandings about how they work. According to the study, the "thinking" of large models is actually about calculating correlations between data, not understanding causal relationships.

To support their argument, the researchers cited reasoning models like DeepSeek R1, which perform well in certain tasks but do not possess human thinking abilities. The study indicates that there is no real reasoning process in AI outputs. Therefore, if users view the intermediate inputs generated by AI models as reasoning processes, they may develop misleading confidence in their problem-solving capabilities.

This research serves as a reminder to approach AI technologies with caution in an increasingly AI-reliant era. As our understanding of large model capabilities deepens, future AI research may shift towards being more interpretable, helping users better understand the actual working mechanisms of AI.