Recently, the reporter of Science and Technology Daily interviewed Jin Yaochu, an academician of the European Academy of Sciences and a professor of artificial intelligence at Humboldt University in Bielefeld, Germany, around the hot science topic ChatGPT. Professor Jin also served as the chair professor of "Computational Intelligence" in the Department of Computing at the University of Surrey in the United Kingdom, the president-designate of the IEEE Institute of Computational Intelligence, and the chief editor of the IEEE Journal of Cognitive and Developmental Systems. The major development of generative model talks about the core of ChatGPT. Jin Yaochu said that the technology represented by ChatGPT is a major progress made by the third wave of AI technology. It is based on the generative pre-training Transformer model and adopts the "self attention mechanism". Compared with the traditional discriminant model, the biggest feature of the generative model is that it can directly learn the characteristics of the training data itself, so it can more effectively learn a large number of samples, especially massive language and visual information. The traditional language processing model can only find the relationship between words according to their adjacent words, so it has great limitations. After the introduction of the self-attention mechanism, Transformer can flexibly find out the relationship between words not only in a certain sentence, but also in different sentences, even in the whole article or different articles, and its learning ability is greatly enhanced. The birth of ChatGPT is not a day's work. For the birth of ChatGPT, Jin Yaochu stressed that this is just the feeling of the public. From the perspective of scientific research, ChatGPT is also evolved step by step. According to the paper published by OpenAI, ChatGPT took a long time to train and fine-tune with various learning methods. The answer given by the generative model is not the training data that people give it in advance, but the one generated by the model, so it is difficult to ensure that it is 100% correct. ChatGPT uses a lot of the latest learning technologies in the training process. For example, unsupervised learning, supervised learning, multi-task learning, small sample learning, self-supervised learning, etc. In addition, it also adds reinforcement learning based on human feedback to gain the ability to align and empathize with humans. Select the best answer through human evaluation, that is, scoring the answers generated by it. This process is actually a reinforcement learning process, that is, fine-tuning the original model. Other AI models may not be like ChatGPT, which uses a large number of human feedback to make the model produce more responsive responses to human expectations, and the corresponding experience will have a gap. Jin Yaochu, who has no independent "mind", said that as a dialogue system from text to text, ChatGPT's strongest feature is the full learning of natural language. The generated text is more "fluent" and conforms to human experience, which is a major technological innovation. However, it is worth noting that there are many generative machine learning models, such as text to voice information, or text to image information. ChatGPT may not perform as well in these aspects. This is one of its limitations. On the other hand, as of 2021, the current ChatGPT training data has nothing to train,