On July 8, the three-day World Artificial Intelligence WAIC Conference closed in Shanghai. The exhibition area of the conference exceeded 50,000 square meters, a total of 133 forums were held, the conference connected with 210 upstream and downstream enterprises, reached 11 billion intended procurement amount, and promoted the signing of 32 major industrial projects, with a total project investment of 28.8 billion yuan.
It is worth noting that the AI big model has become the absolute focus of this conference. According to incomplete statistics, more than 30 large model platforms were released and unveiled to the public, 60% of the offline booths showed the introduction and application of generative AI technology, and 80% of the participants discussed the content around the large model.
From generative AI technology research and development, the appearance of large model platforms to the layout of scene applications, many followers' questions have arisen: Has the application of large models entered a mature landing period?
More than 30 large models were unveiled
At this exhibition, Alibaba Cloud, SenseTime, Tencent, JD.com and other enterprises have used the WAIC platform to release the progress of their large models.
Among them, Alibaba Cloud announced that the AI painting creation model Tongyi Wanxiang opened a directional invitation. It is understood that Tongyi Wanxiang is based on the combined generation model Composer developed by Ali, which can provide highly controllable and extremely free image generation effects by disassembling and combining image design elements such as color matching, layout, and style, and only one model can support multi-type image generation task capabilities, and the research results have been published at ICML 2023.
At the same time, Alibaba Cloud launched the large model assistant Magic GPT, which can call all large models with one click. At present, Magic GPT can now call the APIs of more than ten core AI models in the Magic Match community, and the capabilities of Magic GPT will continue to be enhanced as more model APIs are added in the future.
Tencent Cloud announced the upgrade of three major areas: technical foundation, innovation scenarios, and AI for Science, which is also another update after the industry's large-model technology solution in June. Among them, the newly upgraded Tencent Cloud's self-developed Xingmai high-performance computing network can improve GPU utilization by 40%, save 30%~60% of model training costs, and bring 10 times the communication performance improvement to AI large models.
SenseTime launched the "SenseTime New SenseNova" large model system, which is a comprehensive upgrade of the multi-faceted system. As a natural language processing model with hundreds of billions of parameters, SenseChat version 2.0 breaks through the limitation of input length of large language models and launches model versions with different parameter magnitudes. The model parameters of SenseMirage 3.0, SenseTime's self-developed large-scale model, SenseMirage 3.0, have increased from 1 billion since its first release in April this year to 7 billion, enabling professional photography-level image detailing.
In addition to the leading enterprises in the field of Internet technology, AI startups such as Fourth Paradigm and Yunzhisheng also competed to show the latest achievements in large models. At the same time, the academic community has also begun to carry out research on large models. The 100-billion-parameter Chinese-English dialogue model ChatGLM-130B of the Knowledge Engineering Laboratory of the Department of Computer Science of Tsinghua University and the MOSS of the Natural Language Processing Laboratory of Fudan University attracted the attention of the participants.
For the development of artificial intelligence technology, Pan Xingang, first author of DragGAN and assistant professor at the School of Computer Science and Engineering, Nanyang Technological University, mentioned at the forum that AI is becoming the basis for broader creative applications.
The landing application of large models has entered an accelerated period
Xu Li, Chairman and CEO of SenseTime, said during the conference: "The breakthrough of large models has set off a new round of technological revolution in artificial intelligence, followed by explosive growth in industrial demand, and new application scenarios and application models are rapidly emerging. ”
This view was also validated at forums and booths.
Ant Group and Tsinghua University released the AI security detection platform "Ant Jian 2.0". AntJian 2.0 can detect generative AI models with generative AI capabilities, identify hundreds of risks in three categories: data security, content security, and technology ethics, covering a variety of data and task types such as tables, texts, and images.
At the conference forum, ZhongAn Insurance released the first AIGC application map of the insurance industry, and brought ZhongAn AIGC Middle Platform Lingxi and the first batch of AIGC application tools for insurance vertical scenarios - Yichuang content operation platform and Jizhi operation analysis platform. Compared with the traditional "AI + insurance" application, Atlas conducts feasibility studies on the enabling value and future application potential of AIGC in 19 major categories and 78 sub-categories of business scenarios from six dimensions: product design and innovation, marketing and exhibition, underwriting claims, customer service, smart office and R&D management, covering property insurance, life insurance and other businesses, involving agents, self-operated and other marketing channels, bringing reference and reference for the digital and intelligent transformation of the insurance industry.
At the NetEase booth, NetEase's many artificial intelligence large model landing products made their debut, focusing on the technical achievements in the artificial intelligence infrastructure layer, engine layer, platform layer, model layer and application layer, including the unmanned loading robot jointly developed by NetEase Fuxi and China Construction Eighth Bureau, and the Chinese model Danqing developed by NetEase Fuxi. Among them, NetEase Youdao's self-developed domestic first education model "Ziyue" latest application results - virtual population language coach also made its debut. NetEase staff revealed that the application results of the "Ziyue" large model series, including virtual population language coaches, will be disclosed in detail at the end of July.
In terms of industry applications, different companies have similar views: the launch of large models must enter the industry application. Zhou Jingren, CTO of Alibaba Cloud Intelligent Group, said that the release of Tongyi Wanxiang is a key step for Alibaba Cloud to fully grasp the multi-modal capability, which will gradually be opened to industry customers. He said: "Multi-modality is an inevitable path for the evolution of large models, and we hope to serve thousands of industries with the capabilities of different modalities to help enterprises fully embrace the intelligent era." ”
SenseTime hopes to continue to promote the leap forward and improve AI infrastructure capabilities through 'big model + big device', not only to create a basic model with more powerful general capabilities, but also to further efficiently integrate expertise in different vertical fields to build a professional big model with better understanding of the industry and more expertise."
On the basis of the big model, how to serve the development of vertical industries, Tencent also gave its own answer. Tang Daosheng, senior executive vice president of Tencent Group and CEO of the Cloud and Smart Industry Business Group, said at the conference that building your own exclusive model based on the big industry model may be a better option for enterprises. At present, Tencent's own enterprise-level applications have been connected to the industry's large-scale model capabilities.
How far is the real application?
According to the latest data from KPMG, with China's vigorous promotion of computing power infrastructure construction, the scale of China's computing power-related industries can reach 24 trillion yuan.
The application of large models in the industry is expected to bring more development opportunities, but the problem is how to make large models truly realize the application. Yang Zhilin, founder of AI start-up Moonshot AI and assistant professor at Tsinghua University, talked about his concept of innovation and achievement transformation. He said that there are still many problems in large models that have not been solved, and when thinking about them, it is more important not to think based on a single point of thinking, but to more systematically abstract the underlying commonality of these problems and return to a more essential level to solve.
In this regard, Jiang Jiyun, chief technology officer of ZhongAn Insurance, said that with the landing of more vertical industry large models and self-developed large models, AIGC will show the potential of integrated migration, through the combination of scenarios and data, to achieve personalized customization, risk control and other advantages, bringing more efficient, intelligent and more humanized services.
Ant Group believes that the in-depth application of artificial intelligence not only puts forward higher requirements for data, algorithms and computing power, but also poses more challenges to security, privacy and ethics. Under the premise of ensuring data security and privacy protection, and improving the ethics and security of artificial intelligence, AI technology can truly release the application value.
Wu Yunsheng, vice president of Tencent Cloud, head of Tencent Cloud Intelligence, and head of Youtu Lab, believes that the development of AI big model technology and industrial exploration are inseparable from industrial chain collaboration and ecological co-construction.
Wu Yunsheng, Vice President of Tencent Cloud, Head of Tencent Cloud Intelligence, and Head of Youtu Lab
According to KPMG's research, the development of China's manufacturing sector is worth looking forward to. Qingjie Zhang, Managing Partner of Digital Empowerment, KPMG China, says: "China's manufacturing industry has the characteristics of a complete industrial system, long industrial chain, and multiple industrial clusters, which makes the manufacturing industry digital scenarios very rich, and it is expected to fully realize economies of scale and scope in terms of digital transformation. ”
Source: Global Network