Alignment of human values in the big language model
From: HIT SCIR
Enter the NLP group —> join the NLP exchange group
Bing Qin Harbin Institute of Technology Social Computing and Information Retrieval Research Center
Title: Alignment of Human Values in Large Language Models
Abstract: At present, the fourth wave of artificial intelligence technology represented by ChatGPT is sweeping all aspects of social and economic life. Relying on its astonishing ability, this kind of technology is rapidly affecting the development of society. However, due to its technical reasons, the use of large language models represented by ChatGPT may still cause harm to humans and society, such as generating inaccurate information to mislead humans, outputting harmful information to harm humans, etc. How to align the values of large language models with humans is a major problem that needs to be solved urgently. This report will propose to align the big language model with human values from four levels, which are: factual (telling facts and logic), legal (conforming to laws and regulations), reasonable (conforming to morality) and cultural (conforming to national regional culture). Based on this framework, we sort out the existing work in detail, not only summarizing the existing tasks and methods, but more importantly, finding valuable research directions and possible development trends in the future. We firmly believe that the goal of artificial intelligence is to provide theoretical support for the improvement of human well-being and the harmonious coexistence of all things! Therefore, through this report, it is hoped that more researchers will be able to improve the fidelity and value alignment of the generated content of the large language model model, so that artificial intelligence technology will be more beneficial to human society.
Lecturer: Qin Bing, professor of Computing Department of Harbin Institute of Technology, doctoral supervisor, deputy director of Institute of Natural Language Processing of Harbin Institute of Technology, director of Social Computing and Information Retrieval Research Center. National key research and development projects, the person in charge of key projects of the National Natural Science Foundation of China. Expert of the Science and Technology Innovation 2030-"New Generation Artificial Intelligence" major project management expert group of the Ministry of Science and Technology, executive director of the Chinese Information Society of China/deputy director of the Language and Knowledge Computing Committee/director of the Emotional Computing Committee, natural language processing of the Heilongjiang Computer Society Director of the special committee. Main research directions: natural language processing, knowledge graph, affective computing, text generation. Won the first prize of Qian Weichang Chinese Information Processing Science and Technology Award of Chinese Information Society, the first prize of Heilongjiang Science and Technology Award, the second prize of Heilongjiang Science and Technology Award and the second prize of Heilongjiang Technology Invention Award. Selected into the "2020 Global Women in Artificial Intelligence and AI 2000 Most Influential Scholars List" and "Forbes China 2020 Technology Women List".
PPT sharing
Harbin Institute of Technology Social Computing and Information Retrieval Research Center
Qin Bing
Editor in charge of this issue: Zhao Yanyan
Editor of this issue: Yang Xin
Enter the NLP group —> join the NLP exchange group