The three-day World Artificial Intelligence Conference 2023 has come to an end. At this year's conference, artificial intelligence big models have become the "protagonist", domestic and foreign large model star teams, more than 30 large models all appeared, but also a number of related content forums, chasing the "big model", how will it affect the development of artificial intelligence, and what changes will it bring to daily life and thousands of industries?
The arms raised and fell in turn, dancing to the music, at this year's World Artificial Intelligence Conference, this "Thousand Hands Guanyin" dance performed by robots was eye-catching; With an AI brain, the robot not only learns to dance, but also practices dribbling and shooting skills, becoming a player on the court.
Chen Yuan, vice president of Cloudminds Robotics, said: "With GPT, my robot can have stronger reasoning and analysis capabilities, and walk on a larger stage, which can provide more services to people.
On the other side of the exhibition, the iteration upgraded to version 3.5 of Baidu Wenxin big model, also showed good strength. The dialogue model "Wen Xin Yiyan" has been able to quickly extract the core ideas and abstracts of a 30-page scientific paper; In the "Wen Xin Yi Frame", enter keywords such as happy young people, cycling, blue sky and white clouds, and a set of realistic or sci-fi pictures will immediately jump in front of you.
Wu Hao, a marketing consultant of Baidu's technical middle office marketing department, told reporters: "For cross-modal large models, what is very difficult is that the portrayal of characters' facial expressions and body postures is now more realistic. ”
At this year's World Artificial Intelligence Conference, the big model is the "protagonist". More than 30 large models at home and abroad, such as Baidu Wenxin, Ali Tongyi, iFLYTEK Xinghuo, and SenseTime Ririsin, all appeared, showing their "muscles", demonstrating various tricks such as long text dialogue, AI writing poetry, painting, and music composition. "It can be said that each company has its own characteristics, and now is still a period of blooming." "Artificial intelligence has become a very important topic in our lives, and we are very much looking forward to greater breakthroughs." Visitors said.
When "a hundred flowers bloom", it is inevitable that some people will want to compare the capabilities of these large models. In the past, it was usually only a "simple and rude" competition parameter quantity, but now, in the view of industry experts, the evaluation of the level of large models needs to comprehensively consider multi-dimensional factors such as comprehensiveness, rationality, ease of use, response speed, cost and energy efficiency.
The perspective reporter learned that as of June this year, 79 large models above the scale of 1 billion parameters in the mainland have been released, and a large model technology group that closely follows the world's frontier has been formed; However, behind the "100-model war", where the big model will go, especially how to deepen the application of industry scenarios, is also the focus of the industry.
On July 7, Huawei's Pangu 3.0 model was officially released, with the slogan of "Don't write poetry, only do things". The goal is clear, directly aimed at the industry application, and has been involved in drug research and development, coal mining, railway and finance. In May, their meteorological industry model predicted the path that Typhoon Mawa would turn in the eastern waters of Taiwan Island five days in advance; Compared with traditional weather forecasting, the forecast error is reduced by 10% to 15%, and the speed is increased by more than 10,000 times. The logic behind it is based on the Pangu Grand Model and the training of meteorological data over the past 40 years.
"Traditional weather prediction, mainly relying on super computing power to do simulation, requires a lot of servers to do supercomputing, and our large model is trained in advance with massive data, now only need real-time meteorological data to do small data, small computing power calculation, and then you can obtain more accurate meteorological data." Huawei booth staff introduced.
In the eyes of many insiders, the future of the large model is to become a service platform, which can be used in finance, office, urban management, medical and other industries, through general, open source sharing, to reduce the cost of developers, you can on the platform on-demand "à la carte" processing, do not need each company to start from scratch into development. "On the one hand, we have a basic generic model; The second aspect can be based on the general model, training for a certain industry, or open source for the industry to share. Qian Zhengping, chief architect of Alibaba Cloud Intelligence, said.
At present, Shanghai is also exploring governance research such as large model evaluation and standard system, and at the same time, Shanghai will actively optimize the industrial layout and create a good industrial environment for artificial intelligence innovation and application.
Zhang Ying, deputy director of the Municipal Economic Information Commission, said: "We will further increase the opening of application scenarios, especially key industry scenarios, so that artificial intelligence large model enterprises have more places to use." ”
The "100-model war" is still in battle, and general artificial intelligence is becoming a key track for the key layout of various countries. At present, Shanghai, Beijing, Shenzhen and other places have also introduced support measures to seize the opportunity of large models and carry out innovative algorithms and application research. Of course, with development, will the big model that can learn replace human work? How is privacy and data secured? These ethical and safety issues have always been hotly discussed, and setting up a safety "fence" and making reasonable use of technology and regulations to deal with the risks and impacts brought by artificial intelligence are also mandatory questions that must be overcome.
Reporter: Jin Puqing
Videography: Jiang Wenyue
Edit: Yuyu-chan