The AI large model has achieved remarkable results from conception to application implementation. At present, it is gradually moving from the C-side "dialogue and poetry" to all walks of life.
Recently, Tencent Cloud announced the research and development progress of large-scale models in the industry. Based on the demand scenarios of different enterprises, it has built a selected store of large-scale models in the industry based on the Tencent Cloud TI platform, providing customers with one-stop MaaS (Model-as-a-Service) services , to help customers build exclusive large models and intelligent applications.
Coincidentally, at the Huawei Developer Conference 2023 (Cloud), Huawei Cloud Pangu Large Model 3.0 was officially released. This is an industry-oriented large-scale model series, which will focus on "industry reshaping", "technology taking root" and "opening and flying together" Three innovative directions to provide better services for industry customers, partners and developers.
Leading cloud service providers such as Tencent Cloud and Huawei Cloud led the way, followed by leading companies and start-up companies from all walks of life, and the industry model became lively for a while. Not surprisingly, in recent months, the large-scale model industry will usher in a small climax of intensive releases, and the verticalization and corporateization of the large-scale model industry will also be further deepened.
Moved to industry model
Judging from the current market situation, it is the general trend that the large-scale model war will shift from general-purpose large-scale models to industrial large-scale models.
Just as Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industry Business Group, said: "The general large-scale model can solve 70%-80% of the problems in 100 scenarios, but it may not be 100% satisfactory for a certain enterprise. Scenario requirements”, general large models have certain limitations in solving professional problems in specific scenarios.
First, the general-purpose large model has wide applicability but lacks industry depth, and can hardly provide high-value services in professional and more subdivided vertical fields.
As we all know, the general large-scale model needs three pillars of support, namely algorithms, data, and computing power. Among them, the data determines the scope of intelligence and affects the accuracy and comprehensiveness of data analysis. However, most of the general large-scale model data comes from public literature and network information. Professional industries Insufficient data accumulation, in highly specialized industries such as medical care and finance, the value of services provided by general-purpose large models is limited.
Second, the use of general-purpose large models requires uploading data to the server provided by the large-scale model party, and there are risks of data leakage and security issues.
Ordinary general-purpose large models are not locally deployed and do not have user authority control. It is difficult for the financial industry and G-end customers who are sensitive to data security to ensure data security when using large models. Effective control measures should be taken to improve the security and data security of general large-scale models. Protection is imminent.
Thirdly, the general-purpose large model needs to process massive amounts of data and computing resources, which requires high computing power and algorithms, which means higher operating costs. According to the public data on the Internet, taking the GPT-3 AI large model as an example, OpenAI used nearly 1 trillion words and 45TB of data to train it, and the single training cost reached 1.4 million US dollars.
Compared with the general large model, the vertical large model has great advantages in cost control, because the model parameters are less, and the cost of training, debugging, optimization and implementation is lower, which is more friendly to small and medium-sized enterprises pursuing "cost reduction and efficiency increase".
General large-scale models are deficient in data security, computing power cost, and industry depth. Cloud service providers such as Tencent Cloud and Huawei Cloud are actively developing low-threshold, low-cost, high-efficiency and safe industry large-scale models that are suitable for implementation.
Tencent Cloud integrates ecological resources
While other cloud service providers were still fighting in the C-end market, Tencent Cloud targeted the TOB scene early.
On June 19, in response to the high cost and low security of general large models, Tencent Cloud launched the MaaS service panorama to provide customers with one-stop model services, covering complete model tools, mature process methods, and comprehensive supporting facilities services and leading security capabilities.
At the same time, Tencent Cloud developed and expanded the industry's large-scale model business around its own industrial ecological advantages, and achieved certain results and feedback.
It is understood that Tencent Cloud's large-scale industry model capabilities have been implemented in a variety of products such as Tencent Qidian, Tencent Conference, and Tencent Cloud AI Code Assistant. In addition, Tencent Cloud has joined forces with industry leading companies to provide more than 50 large-scale industry solutions for more than 10 industries including cultural tourism, government affairs, and finance.
It has to be said that the huge industrial ecology is an important resource to promote the implementation of large-scale industry models. Tencent invests in large and wide-ranging businesses, and industries such as social networking, entertainment, general health, finance, and automobiles all need to go digital, and large models are standard for the development of industrial digitalization. In terms of application industry large models, Tencent Cloud has Natural advantages.
In addition to industrial ecological resources, the competitiveness of Tencent Cloud's large-scale industry model lies in the accumulation and innovation of technology.
In terms of technology, Tencent Cloud is backed by top laboratories such as Tencent Youtu Lab and Tencent AI Lab. The underlying computing power, algorithm development, AI application and other technical capabilities are self-evident, and the strength of Tencent Cloud computing power, algorithms, etc. Not static but always improving.
It is worth mentioning that Tencent Cloud's self-developed Xingmai high-performance computing network and vector database can increase GPU utilization by 40%, save 30% to 60% of model training costs, and bring 10 times communication to AI large models The performance improvement can provide a more solid computing base for the application of large-scale models in the industry.
According to the financial report, Tencent is also developing technologies such as self-developed database TDSQL, security platform EdgeOne, and big data processing suite TBDS, as well as TI platforms related to AI development. Tencent’s R&D investment in the first quarter of 2023 was 15.181 billion yuan, and since 2018, it has accumulated more than 220 billion yuan in the five years, and its R&D investment is only a lot more.
In terms of ecology, Tencent Cloud adheres to the concept of openness and win-win to open technology to more partners, creating a large-scale ecology of sustainable innovation and development.
In this regard, Wu Yunsheng, vice president of Tencent Cloud, head of Tencent Cloud Intelligence, and head of Youtu Lab, said: "AI large-scale model technology development and industrial exploration are inseparable from industrial chain collaboration and ecological co-construction. Tencent hopes to join hands with industry partners , and jointly promote the innovation and implementation of large models in the industrial field."
All in all, based on its own advantages in ecological resources, a solid technical foundation and an open technology platform, Tencent Cloud is the first to "settle" in the field of large-scale industry models. With the further improvement of technology and the expansion of application scenarios, Tencent Cloud is expected to be in the industry. The modeling domain maintains its leadership.
HUAWEI CLOUD Deeply Cultivates Vertical Industries
The wheels of time are rolling. Since 2019, Huawei has been committed to developing and upgrading the iterative Pangu large model. Through continuous investment of funds, manpower and resources, it has improved its technical strength in the field of large-scale models in the industry.
From 2019 to 2020, Huawei internally invested in the research and development of AI large-scale models, and established a project to make a large-scale model of Pangu; in April 2021, the large-scale model of Pangu on Huawei Cloud officially launched; in 2022, the large-scale model of Pangu was upgraded to version 2.0 and landed in industry applications; Pangu Large Model 3.0 was officially released, continuing to deepen the industrial chain.
The three-layer structure of Pangu Large Model 3.0 can quickly adjust the direction, quickly adapt to the changing needs of customers, and provide reliable solutions for the ever-changing industrial digitalization trend. Behind HUAWEI CLOUD's ability to take into account general large-scale models, industry large-scale models, and more detailed scene large-scale model services, it relies on continuous technology research and development over the years.
On the one hand, HUAWEI CLOUD's strong model technology is strong, which is reflected in the continuous investment and innovation capabilities of artificial intelligence technology. According to the financial report data, Huawei's R&D investment in 2022 will reach 161.5 billion yuan, accounting for 25.1% of the annual revenue, and the accumulated research and development expenses invested in ten years will exceed 977.3 billion yuan.
This level of investment and speed of innovation has enabled HUAWEI CLOUD to make breakthroughs in the field of artificial intelligence, laying a solid foundation for the full implementation of large models. Financial report data shows that in 2022, Huawei submitted 8,440 patent applications, of which more than 90% are invention patents, involving electronic communications, optical technology, Hongmeng operating system, computing storage, smart cars and other fields.
On the other hand, HUAWEI CLOUD continues to deepen the digitalization of vertical industries such as government affairs, railways, manufacturing, and pharmaceuticals. By providing safe, reliable, and efficient AI large-scale model solutions in these vertical industries, Huawei Cloud has established a good image of itself and also highlighted the Pangea large-scale model. the value of.
In the field of coal mines, the large model of Pangu Mine can cover more than 1,000 subdivided scenarios under business processes such as mining, excavation, machinery, transportation, transportation, and washing of coal mines, and has been used on a large scale in 8 mines across the country; in the field of railways, Pangu The large railway model can accurately identify 67 kinds of freight cars and more than 430 kinds of faults running on the current network, and the screening rate of non-faulty pictures is as high as 95%...
It has to be said that the highly customizable capability of Huawei Cloud Pangu Large Model 3.0 enables it to be customized according to the needs of different industries, enterprises, and users, and can meet various complex scenarios in thousands of industries. This flexibility and scalability The uniqueness has allowed the Huawei Cloud Pangu large model to gain "a lot of achievements" in the vertical field.
Regarding the future, Hu Houkun, Huawei's rotating chairman, said that Huawei has two focus points in the development of artificial intelligence: "First, build a strong computing power base to support the development of China's artificial intelligence industry. Second, from general-purpose large models to The large-scale industry model allows artificial intelligence to serve thousands of industries and scientific research and innovation."
This means that HUAWEI CLOUD will continue to deepen the industry's large-scale model technology research and ecological expansion, and accelerate its occupation of a larger market share, and this process will also encounter more technical challenges and scene competition.
Accelerate to seize the B-side "landing point"
In addition to Tencent Cloud and Huawei Cloud, more and more major manufacturers are investing in the development, training and commercialization of large-scale industry models. They hope to use more accurate industry data and lower costs to solve the "stuck neck" of certain industrial core technologies. "problem, which also means that a "landing battle" for large industry models is coming quietly.
First of all, large-scale industry models in the fields of autonomous driving, medical care, education, and finance are emerging one after another. The concentrated explosion of a series of large-scale industry models will promote the development and application of large-scale models in the industry, and will inevitably intensify market competition.
In March, Netease Youdao launched the first domestic ChatGPT-like model "Ziyue" in the education scene; in April, Momo Zhixing released the world's first autonomous driving generative large-scale model DriveGPT; in May, I Love My Home launched the industry's first real estate brokerage Large-scale model version 1.0; in June, Neusoft launched Tianyi’s large-scale model in the medical field for the medical field; in July, China Mobile announced the release of the nine-day artificial intelligence industry large-scale model...
Secondly, Baidu Cloud, Alibaba Cloud, Tencent Cloud, and Huawei Cloud chanted the slogan of large-scale model development, and launched related industry large-scale models. The landing in different industries and industries has become a new battlefield for leading cloud service providers.
Some people say that AI big models can only be played by large companies and companies with sufficient funds, and it is true. The research and development and training of AI large models require massive amounts of data, high costs, and excellent scientific and technological talents, and companies that meet these conditions are basically Internet giants or industry leaders. The development consensus of them going deep into the industry is also the industry Competitive benchmark.
Finally, the digital transformation needs of all walks of life in China are diverse and rich, and industry model manufacturers are exploring the feasibility of implementing large-scale models in various industries. IDC has predicted the scale of the AI market based on large models. It is estimated that the scale of China's artificial intelligence market will exceed US$14.7 billion in 2023, and will exceed US$26.3 billion by 2026.
To sum up, large-scale industry models are an outlet that cloud service providers must not miss. Tencent Cloud and Huawei Cloud have already provided differentiated large-scale industry model services by virtue of their industrial ecological genes and AI technology. The implementation of large-scale industry models will soon Become the next home field for cloud vendors to compete.