Sure enough, it's the focus again! This AI report states…
From March 26th to March 29th, the Boao Forum for Asia 2024 Annual Conference was held in Boao, Hainan. As one of the most influential forum summits today, the theme of this year’s annual conference is “Asia and the World: Common Challenges, Common Responsibilities.”
01
predict
End-to-end lateralization of large AI models may become a future trend
In the context of accelerating the development of new productivity, what is the development trend of AI large models in China?
The "Report" believes that AI large models are mainly divided into two types: cloud-side large models and device-side large models. Among them, the cloud-side large models are specifically divided into general-purpose large models with strong universality and industry large models with strong professionalism, while the terminal-side large models mainly include mobile phone large models and PC large models.
On the one hand, large cloud-side models represented by Baidu Wenxinyiyan, Alibaba Tongyi Qianwen, etc. provide powerful computing power and massive training databases for C-end individual users and B-end enterprise users;
On the other hand, the client-side large model relies on its advantages such as low cost, strong mobility, and data security to provide C-side users with more thoughtful and exclusive services. In the future, with rich usage scenarios, lower model costs, and secure privacy protection, end-side sizing of large models may become a trend.
Compared with general large models, industry large models are more professional and have higher data security. This means that vertical industries may become the main battlefield for large models.
The report pointed out that——
On the one hand, the industry large model concentrates the resources used by the general large model to form multi-domain capabilities in specific areas, and the model parameters are relatively small, which has significant cost advantages for enterprise implementation;
On the other hand, industry large models combine the internal data of enterprises or institutions to provide services for the actual business scenarios of B-end users, which can better reflect the cost reduction and efficiency improvement effect of the model on the institution.
In addition to predicting the development direction of large models in various categories, the "Report" also proposed that open-source large models will become a trend in the future.
From Tsinghua University’s open-source Chinese-English bilingual pre-training model GLM-130B to Vivo’s open-source 7 billion parameter-level blue heart large model 7B and the corresponding fine-tuning framework, Alibaba Cloud’s open-source Qwen-72B, Qwen-1.8B and Qwen-AudioQwen large models. There are more and more institutions or companies in the industry releasing open-source large models, which continue to promote the innovative development of the AI large model industry.
At the same time, the performance of AI high-performance chips is also the focus of people's attention. Nvidia's A100 chip occupies an important share in the global AI high-performance chip market, and the development speed of domestic AI high-performance chips has also accelerated in recent years. The two main chips launched by Huawei Ascend, 310 and 910, are one of the representatives.
In the future, as AI high-performance chips continue to be iteratively upgraded, the performance and capabilities of large models will continue to be improved.
02
Opportunity
The era of AI mobile phones has begun
If terminal lateralization may become a trend in the future, how will large AI models be used by us in a more convenient form? As the "arms race" in large models intensifies and refreshes people's expectations for AI, the world's major mobile phone s have also tacitly focused their competition on the same focus - AI mobile phones.
In addition to the advantages of the matrix, the operational advantages of the blue heart large model on the terminal side cannot be underestimated. It is understood that the maximum writing speed of the Blue Heart Model 1B model can reach 60 words per second, which is far beyond the human eye's reading requirement of about 10-20 words per second.
As Zhou Zhou, vice president of Vivo and director of vivo AI Global Research Institute, said in the "Report":
"In terms of industry trends, the number of companies investing in basic model training may decrease significantly in the future, and more companies will look for application scenarios and popular applications. However, Vivo combines self-developed large model end-side and matrix technology advantages, and will focus on the application experience of the mobile phone industry, use large models to reconstruct various functions of mobile phones, and find implementation scenarios, to benefit more users."
03
Outlook
The industry wave has arrived
AI large model creates new productivity "acceleration"
Although the industry wave has arrived, the development of China's AI large model industry still faces many challenges. From the perspective of the surroundings, "including insufficient dispersion of computing power, questions about whether the Transformer structure is optimal, scarcity of domain data, and lack of phenomenal applications."
In terms of computing power, due to the dual impact of import restrictions and domestic technology necks, there are gaps in the domestic AI high-performance chip market, and the development of the large model industry has encountered computing power bottlenecks;
In terms of architecture, the Transformer architecture used by mainstream AI large models has limitations such as consuming large computing resources and occupying a large amount of memory. In addition, the Chinese databases that can be used for large model training in my country still have problems such as incomplete data types and low information credibility. The domestic AI large model industry has yet to see blockbuster applications.
However, despite these many challenges, there are still multiple favorable factors that promote the rapid development of large models:
"The first is the strong support from the government level such as the 'Artificial Intelligence +' action. The second is the surge in users' demand for improving life and work efficiency. In addition, technology companies have increased their investment in funds, manpower, and technology research and development in the AI field, and the coordination of all links support the future development of large models.”
The "2024 State Council Government Work Report" proposed that we should vigorously promote the construction of a modern industrial and accelerate the development of new productive forces. Deeply promote the innovative development of the digital economy. Deepen the research and development applications of big data, artificial intelligence, etc., carry out the "artificial intelligence +" action, and create an internationally competitive digital industry cluster.
The "Report" of People's Daily On also pointed out that AI large models, as an important technology in the current field of artificial intelligence, are fertile ground for nurturing new productivity.
In the context of accelerating the development of new productivity, the mobile phone industry, as a benchmark for the real economy, is accelerating new productivity and promoting changes in the real economy.
At the same time, as the official mobile phone designated by the Boao Forum for Asia, the AI functions of the vivo X Fold3 series flagship products also performed well.
It is understood that the large blue heart model carried by this series of mobile phones continues to grow, and the overall data volume has been upgraded to a super combination of "7B on the device side" + "70B on the cloud side". The re-implementation of Vivo's latest AI technology is one of the best examples of new productivity-boosting industrial development and leveraging demand cycles.
Wang Yuntao, the deputy chief engineer of the Artificial Intelligence Research Center of the China Academy of Information and Communications Technology, said that the emergence of large AI models has made the use of artificial intelligence technology to generate content leap from "available" to "easy to use." In the future, artificial intelligence-generated content may undergo one or more technological paradigm subversions from "easy to use" to "efficient."
Since the birth of the industry, mobile phone evolution has been moving at full speed. Currently, in the context of accelerating the development of new productivity, future-oriented trend technologies such as chips, AI, and 6G in the industry are also constantly converging.
As a representative company in the mobile phone industry, vivo has reached a new level in fully embracing new productivity, developing new technologies, and creating new value. It continues to promote the development of the real economy through technological innovation and industrial renewal.
The industry looks forward to leading companies represented by Vivo to tell "new stories" in innovative fields such as AI large models.
Contact us now, you will be more equipped to choose the filling, capping, or labeling machines that can benefit your production output and company. The customized packing machines will provide high-quality processes, eliminate waste, and maintain product quality throughout production. In addition, the packing machines will be able to adapt to your operations changes in the future. Let a design and manufacturing company helps your operations by creating customized filling, capping, and labeling machines that work best with your processes and products.
