On April 21 , the “Large Model Industry Development Seminar” jointly organized by Shanghai Key Laboratory of Data Science, Daguan Data and Shanghai Pudong Software Park came to a successful conclusion . This symposium is another scientific and technological event after Daguan Data successfully held the “ChatGPT and Large-scale Model Symposium” in Beijing and Chengdu. The conference brought together international and domestic top industrial and academic experts and scholars in the large-scale model industry , share their latest progress and future plans in the development of the large-scale model industry, and discuss the future development trends and challenges of the large-scale model industry.
Guo Bin, Director and General Manager of Shanghai Pudong Software Park Venture Capital Management Co., Ltd. delivered a speech
Guo Bin, director and general manager of Shanghai Pudong Software Park Venture Capital Management Co., Ltd., said in his speech that Shanghai Pudong Software Park, as an important part of “one core, three parks and two ports”, has always adhered to the core concept of “scientific and technological innovation, industrial development”, Actively lay out the development of the new generation of information technology industry, strive to build an industrial ecosystem, create a source of data technology and a highland of professional services, and promote the intelligent upgrading of industries. It is believed that through the exchanges and discussions at this conference, the concept and application of large-scale models can be deeply understood, and the development laws and trends of the “big model” + “industry” era can be grasped, so as to inject new impetus into enterprise innovation, industrial upgrading and social progress.
Xiao Yanghua, director of Shanghai Key Laboratory of Data Science, led the speech
Professor Xiao Yanghua, director of the Shanghai Key Laboratory of Data Science, as the representative of the organizer, gave a speech on the topic of “Some Thoughts on the Development of my country’s Large-scale Model Industry”. Professor Xiao Yanghua mentioned that the era of general artificial intelligence has arrived, which will bring about unprecedented industrial changes. The ecology of the large-scale model industry is developing rapidly in the world, but my country is still in its infancy, fragmented, with numerous large-scale models, lack of unified planning, cooperation and coordination, and legislative guarantees, and there are serious homogenization phenomena, heavy dependence on foreign large-scale models, and domestic calculations. The power ecology is not perfect, the quality of Chinese data is poor and the scale is small, the lack of talents for large-scale models, and the high cost of implementation. Professor Xiao Yanghua expressed the hope that everyone can actively participate in thinking about “how the large-scale model industry should develop”.
In the theme sharing session, Chen Yunwen, chairman and CEO of Daguan Data, Dong Xiaofei, deputy director of the Artificial Intelligence Department of the Cloud Computing and Big Data Research Institute of China Academy of Information and Communications Technology, Yang Yu, vice president of Aishu R&D, Wu Hengkui, founder and chief scientist of Supersymmetry, and Zhi Xue Yufei, VP of Spectrum AI Large Model Business Unit, Bao Jie, founder of Wenyin Internet, and many other experts in the field of artificial intelligence expressed their opinions on the questions raised by Professor Xiao Yanghua, and discussed the technical development, application and future prospects of large-scale language models Dimensions are shared.
Chen Yunwen, chairman and CEO of Daguan Data, gave a speech on “Exploring the Vertical Training Technology and Application of Large Language Models”
Chen Yunwen, chairman and CEO of Daguan Data and a Ph.D. in computer science from Fudan University, took the theme of exploring the vertical training technology and application of large language models, and shared in detail the engineering exploration of Daguan Data in language models in vertical fields, including: parameter scale and language Discussion on the parameter scale of the model, research on the pre-training data set of the general large model, prompt engineering in the vertical field, Daguan “Cao Zhi” system, AIGC application of Daguan Data, etc., also introduced the development and application of the financial-specific large model BloombergGPT. He believes that deepening the application of large models and AIGC in vertical fields and truly integrating large models and AIGC into the actual business of enterprises is of great significance for commercialization and large language model research. The vertical field model “Cao Zhi” system and AIGC application that Daguan Data is developing will be applied in various industries in the future, empowering each industry. Among them, the “Cao Zhi” large model is quoted from the allusion of Cao Zhi’s seven-step poem, and it is hoped that it will be used as a vertical, dedicated, and domestically produced GPT model.
Dong Xiaofei, deputy director of the Artificial Intelligence Department of the Institute of Cloud Computing and Big Data, China Academy of Information and Communications Technology, gave a speech on “Introduction to the Large Model Standards and Evaluation of China Academy of Information and Communications Technology”
Dong Xiaofei, deputy director of the Artificial Intelligence Department of the Institute of Cloud Computing and Big Data, China Academy of Information and Communications Technology, introduced in detail the current standard development, evaluation and testing situation of China Academy of Information and Communications Technology, and the next work plan. He shared that the Institute of Information and Communications Technology is establishing a large-scale model standard system 2.0 to adapt to the development trend of the industry, and the compilation work is progressing steadily. A number of standards have been released and finalized. Large-scale pre-training model technology and application evaluation methods” series of standards. The evaluation test was fully promoted, and at the same time, combined with a series of standards such as “Large-scale pre-training model technology and product evaluation method”, “Natural language processing technology and product evaluation method”, “Generative artificial intelligence technology and product evaluation method” and other series of standards, launched a large model special project The evaluation work guides the landing of large segments. The Institute of Information and Communication Technology will consolidate the full-stack evaluation capabilities of large models and build a collaborative and win-win evaluation ecosystem.
Yang Yu, vice president of R&D of AISHU, gave a speech “Big language model releases the value of global data”
Yang Yu, vice president of R&D of Aishu, introduced the large domain model and domain knowledge with the topic of big language model releasing the value of global data, and said that the general large model will be split into large domain models of vertical industries, such as chemical industry, securities, government, etc. domain, and the large model can reduce the construction cost of the domain knowledge network and improve the quality.
Wu Hengkui, the founder and chief scientist of Supersymmetry, gave a speech on “Application of Language Models in Scientific Discovery”
Wu Hengkui, the founder and chief scientist of Supersymmetry, introduced the Big Bang Transformer Model (BBT model) in detail in his speech on the application of language models in scientific discovery. He mentioned that the BBT-Science large model is a large model based on the 100-billion-parameter BBT general-purpose large model that continues to be trained on scientific corpus. It can be applied to knowledge questions and answers in different disciplines such as physics, chemistry, biology, and mathematics. It can provide researchers with Fast and accurate knowledge retrieval, providing new Ideas for cutting-edge issues in the research field, and providing interdisciplinary suggestions and insights using the capabilities trained in multidisciplinary knowledge.
Xue Yufei, VP of Zhipu AI Large Model Division, and Bao Jie, founder of Wenyin Internet, participated in the sharing through online connection. Xue Yufei introduced that CodeGeeX is an open source large-scale multilingual code generation model. Currently, there are 23 programming languages in total, covering mainstream languages such as Python, Java, C++, JavaScript, C, Go, HTML, etc., which can better assist programmers in writing code . In his sharing, Bao Jie said that an enterprise’s own large model is essential, because it can help the enterprise understand its own business model and operating mechanism more deeply, so as to better formulate strategies and decision-making, and more effectively improve the enterprise’s profitability. Operational efficiency and competitiveness.
In the roundtable dialogue, Li Zhixu, a researcher at the School of Computer Science of Fudan University and a Ph. Shao Hao, the person in charge, Chen Chengcai, vice president of Xiaoi Robotics and director of the research institute, and Xiao Minglin, co-founder of Yida Technology, conducted in-depth discussions and exchanges on the theme of “China-made “ChatGPT” and large-scale model research status and future development”. The atmosphere is warm.
Round table dialogue: domestic “chatGPT” and large-scale model research status and future development
During the discussion, the experts reached a consensus on “the development direction of domestic ‘ChatGPT’ and large-scale models”: Compared with other fields, the gap between the direction of natural language processing and overseas advanced enterprises is much smaller. To look up to OpenAI rather than myth it, what we are facing is a generation gap, but it is not an insurmountable gap. In the process of catching up and surpassing, we need to give some time to domestic models.
In terms of technology and implementation, the participating experts believe that ChatGPT has driven the overall upstream and downstream of natural language processing and the thinking and development of chips. To some extent, large models may become the next generation of infrastructure. China needs to have its own basic model system , to ensure security, concurrency, stability and other issues. The investment circles, academia, and industry circles should stay calm, stay away from concept hype, and make solid achievements.
Finally , the large-scale model innovation and creative application competition jointly sponsored by Shanghai Key Laboratory of Data Science, Daguan Data, and UKD is now actively soliciting applications . The competition also officially announced the prize pool of this competition, with a total amount of 70,000 yuan. The event aims to stimulate the imagination and creativity of various companies, technical groups, technology enthusiasts and college students, apply the latest large-scale model technology to more field scenarios and tasks, and let some valuable ideas be obtained. Opportunities for realization. At the same time, it is also hoped that through this competition, a platform will be built to allow professionals, technical teams and companies in various fields to have the opportunity to communicate and cooperate to jointly promote the progress of artificial intelligence technology. Come and join the game~
https://www.wjx.top/vm/r3Is5S9.aspx (copy and open the link to register)
This article is reproduced from: https://www.52nlp.cn/%E4%B8%8A%E6%B5%B7%E9%A6%96%E4%B8%AA%E5%A4%A7%E6%A8%A1% E5%9E%8B%E4%BA%A7%E4%B8%9A%E5%8F%91%E5%B1%95%E7%A0%94%E8%AE%A8%E4%BC%9A%E5% 9C%86%E6%BB%A1%E8%90%BD%E5%B9%95%EF%BC%8C%E4%BC%97%E9%A1%B6%E5%B0%96
This site is only for collection, and the copyright belongs to the original author.