“Write a poem to praise the poet Li Bai.” “Poet Li Bai recites the fairy night. Thousands of feelings, all in the deep poem…” Humans writing a poem may take a long time to do so, but it only takes a few seconds for this auther to write a poem. It is ChatGPT, a popular chat robot on the Internet recently.
If you think the ability of this chat robot is limited to chatting with humans, writing poems, and even writing papers, you really underestimate it.

Not long ago, Anthony, a professor of philosophy at the University of North Michigan in the United States, found that the paper with the highest score in the class was actually written by students using ChatGPT when grading the assignment of the World Religions course.
Because of its powerful functions, ChatGPT has been online for less than a week, and the number of daily users has exceeded one million.
ChatGPT is more like a person by comprehensive use of existing technologies
“Its long-term memory, context-related reasoning ability and logic of organization language are amazing. It feels like communicating with a real person when chatting with it.” This is the comment of a ChatGPT user. In fact, the technical principle of this “human-like” chat robot is not new.
To make a chat robot more like a person, first of all, let it understand human language – natural language. “This requires robots to process natural language. Dialogue intelligence technology based on large models can help them quickly ‘learn’ how to communicate with humans.” Yu Kai, co-founder and chief scientist of Spitzer Technology Co., Ltd. and professor of Shanghai Jiaotong University, told Science and Technology Daily.
The reporter learned that the previous AI models have poor generality. When switching from one application scenario to another, they often become inapplicable and need to be retrained. The large model of ChatGPT contains parameters of more than 100 billion levels, which is a general model. It does not need to collect additional data for small-scale in-depth training, and has the advantage of strong generalization.
For chat robots, large model training is like basic and general undergraduate training. Through learning many “courses”, robots have the ability to deal with general things.
But this “undergraduate” ability is not enough. When outputting text, the robot must also take into account the knowledge information in different fields and the relevance of dialogue, and output high-quality text that is more consistent with language logic and human values. This requires a more professional “postgraduate education” for chat robots, so that they can complete more complex tasks and have a more human language expression habit.
“The core of ChatGPT is big model technology and dialogue intelligence technology, among which big model technology mainly includes three parts.” Yu Kai told reporters, “The first is context-based learning technology, which allows AI to understand the context in which people speak; the second is the thinking chain technology introduced by ChatGPT, which allows AI to make corresponding reasoning according to the context; the third is instruction learning technology, which allows AI to understand the instructions issued by humans and then complete relevant operations.”
Yu Kai concluded: “From the perspective of technological iterative development, ChatGPT is a dialogue model based on in-depth learning and statistics, and its basic technology is our existing technology. But when these technologies are combined with large models with high parameters, they produce such amazing effects as ChatGPT.”
Broad application prospect
The application scenario of ChatGPT is far more than talking and chatting with you. It is reported that Microsoft has integrated ChatGPT into its office software suite and tried to embed it into its browser to help users summarize web content, write email and post on social platforms. “It can also correct the mistakes of articles, and even generate corresponding article templates according to simple instructions,” Yu Kai told reporters.
Because ChatGPT has reasoning function, it can also assist human beings in reasoning work. “For example, it can help lawyers judge the case,” Yu Kai said.
In the future, chat robots may be more and more involved in human daily life and work. Its main application scenario is likely to replace the traditional search engine.
“The current search engine is that users input keywords, and then the system gives search results. This interaction is a single round,” Yu Kai said. “ChatGPT is more interactive, and can continuously interact with users in natural language, infer the answers that users want in the process, and then present them to users.”
In addition, when searching for relevant information, ChatGPT can quickly give accurate and unique answers without giving users a series of information and letting users “search” the information they want. This method is particularly in line with the requirements of human beings to quickly find knowledge and obtain new information.
As long as it can ensure the accuracy of the knowledge it provides and the accuracy of the relevant algorithms, chat robots are likely to replace traditional search engines in the future.
If chat robots replace traditional search engines and become a reality, all industries related to information search may change. “Although the current ChatGPT cannot bring fundamental changes to our production and lifestyle, it represents that the development of AI has entered a new stage.” Zheng Zhizhi, president of Shenzhen Information Service Industry Blockchain Association, said in an interview with the media.
“It should be noted that ChatGPT is only a beginning now. In the future, it will certainly transform from text AI to multimodal AI. At that time, its capabilities will become stronger.” Yu Kai concluded.
The rapid development of AI technology has also caused some people’s concern.
Avoiding technical ethical risks and abiding by the ethics of human society can be said to be the bottom line that chat robots must abide by. Therefore, human beings need to formulate relevant policies to restrict AI such as chat robots.