Source: Xinhuanet
How can the movie special effects screen be more shocking? Why are robots "more and more able to chat"?... In the 2022 China Computer Conference held online from the 8th to the 10th, 118 cutting-edge technology forums were specially set up, focusing on hot topics in computing fields such as cloud supercomputing, meta-universe, and human-computer dialogue. Practitioners of industry, academia, research and application discussed and looked forward to it. What does "cloud" mean on the supercomputer?
The experts at the meeting believe that my country's supercomputing capabilities have generally reached the international advanced level. With the gradual popularization of China, the future operating mode of supercomputing will develop towards cloud services, gradually becoming popular and popularized.
The picture shows the Super Computing Center of Gui'an New District, Guizhou Province. Xinhua News Agency reporter Xiang Dingjie Photo by
supercomputer "flying into the homes of ordinary people"? Tang Zhuo , chief engineer of the National Supercomputing Changsha Center, believes that the computing power between the supercomputing centers and data centers is relatively isolated, and a unified computing power network has not yet been formed. The phenomenon of resource islands is serious, and the construction of computing power networks and multi-cloud management based on computing power networks are the main technical challenges.
Academician of Chinese Academy of Sciences and Professor of Beijing University of Aeronautics and Astronautics Cheng Depei said that supercomputing, as an important part of the national computing power foundation, should further improve the infrastructure form under the background of "east numbers and west calculations". After interconnecting supercomputer centers distributed throughout the country through high-speed networks, there must be a unified operation management layer and operation service layer, and a multi-operator and multi-operation center model should be adopted to encourage competition.
What will the public feel in this process? Experts gave an example that the film and television industry is accelerating its development with the fast train of computing power era. For example, the cool special effects in movies will cost high post-production costs. In the future, these high-demand computing power links will be completed more efficiently and at low cost in the cloud, which will also bring better visual enjoyment to the audience.
digital people "knock open" the door of the meta-universe?
With the popularity of meta-universe concept, digital people represented by virtual anchors, virtual idols, virtual employees, etc. are sought after. With novel appearance and smart interactive effects, the industry is exploring its real value in more fields.
Researcher at the School of Optoelectronics of Beijing Institute of Technology Weng Dongdong is the co-founder and CTO of Digital Xusheng (Beijing) Technology Co., Ltd. His team has launched a digital sign language host and participated in the "Digital Meilanfang" master reproduction project.
In his opinion, the authenticity of digital human modeling, driving and rendering is crucial to the immersion of user experience and communication in the metaverse. "What we want is digital human that can interact."
Academician of the Chinese Academy of Engineering and Professor of Beijing University of Posts and Telecommunications said that in the metaverse world, people can link with the metaverse anytime, anywhere, at low delay, and socialize with immersive sense of socialization with a virtual identity. At the same time, the Metaverse has a large amount of diverse content and an excellent economic system to ensure that people can live in the Metaverse for a long time, and work together to improve or even create digital civilization.
Putting aside the infinite imagination space, the industry generally believes that there are three conditions necessary for large-scale application of digital people in the future: high realistic effect, high production efficiency and controllable cost. Whether it can truly achieve high real-time, highly flexible and highly adaptable to landing scenarios, and effectively interact with humans remains to be verified by time.
How does a chatbot evolve itself?
Recently, a chat robot has become popular quickly. It can write code, write scripts, do math problems, and can also answer many tricky questions from netizens.
Experts said that the reason why it is "excellent in intelligence" is inseparable from large-scale multimodal pre-training. More and more researchers believe that the continued progress of artificial intelligence requires breakthroughs in the research of language, voice, vision, graphics, etc., imitate the learning methods of human children, and integrate information from multiple modalities to understand the world.
The picture shows employees of an artificial intelligence company in Guizhou marking data.Xinhua News Agency reporter Xiang Dingjie photographed
"The difficulty in developing artificial intelligence model is not the training process itself, but how to use a model to adapt to ever-changing scenarios." Lu Zhiwu, professor at the Hillhouse Artificial Intelligence School of Renmin University of China, said that his "Wudao Wenlan" team designed and trained a large-scale Chinese general graphic and text pre-training model in 2021, and achieved outstanding performance in cross-modal retrieval, video editing, graphic and text generation and other tasks. For example, enter " Yellow River far above the white clouds" in the dialog box. In just two or three minutes, a ink painting of the rolling Yellow River water rushing among the mountains will appear on the paper.
He said that after explosive development, large-scale multimodal pre-training is currently in the stage of "biting hard bones". In the future, we should pay more attention to the implementation of the application, and make the artificial intelligence model look like a human, and learn a variety of information such as text, pictures, videos, etc. at the same time, and no matter what scenario we face, we can give the most appropriate feedback. (Reporter Xiang Dingjie)
