2023 is undoubtedly the "explosive year" of artificial intelligence, and various new AI-based technologies have sprung up, among which AI large models have become a "competition field" for global technology companies.
A few days ago, China Mobile Research Institute released a domestic AI big model development report, pointing out that Baidu, Alibaba, Tencent, and Huawei are recognized by the industry as the first echelon of domestic large models, with strong R&D investment, technical capabilities and talent teams in large models, and have the strength to catch up with GPT-4.
Among them, from the perspective of application direction, Baidu Wenxin large model layout is the earliest, and internal and external applications are two-way force; Ali Tongyi has taken the lead in deploying internal applications, which have not yet been opened to public customers; Tencent's hybrid model mainly serves internal business.
Domestic large model "dogfight"
Since the beginning of this year, with the popularity of ChatGPT "out of the circle", the domestic large model gods are chaotic, lively, and the popularity even completely exceeds that of the United States.
Today, China has seen a "hundred model war". According to incomplete statistics from Minsheng Securities, up to now, more than 30 large-model products have been released in China this year, and many are in the process of research and development.
In March this year, Baidu launched the big language model Wen Xin Yiyan, and constantly iterated new capabilities. Moreover, Baidu Wenxin Qianfan Big Model Platform, which is still in the internal testing stage, not only supports Wenxin Yiyan big model and third-party big model services, but also has a complete set of tools for large model development and application.
In less than two months, Alibaba, Tencent, Huawei, ByteDance and other major manufacturers have also launched large-model related products.
On April 11, Alibaba Cloud officially released its large-model product "Tongyi Qianwen", and at the same time, Alibaba Cloud Intelligent Group CEO Daniel Zhang said at the summit that all Alibaba products will be connected to the "Tongyi Qianwen" large model in the future.
Huawei and Tencent also announced the new progress of the Pangu large model and the mixed element large model shortly after. In addition, Tencent has also found a new way, starting from the computing power of large models, and launched a new generation of high-performance computing clusters with 3 times higher performance.
In addition, AI companies are also turning their focus to large models, and some emerging AI companies have emerged. For example, the "AI Big Model Entrepreneurship Corps" led by Meituan co-founder Wang Huiwen, Innovation Factory founder Kaifu Lee, Sogou founder Wang Xiaochuan and other corporate executives are also eyeing the AI big model.
Big factory AI technology competition
As we all know, the AI big model is the product of the combination of "big data + big computing power + strong algorithm", and the technical base of science and technology enterprises determines the application value of their AI large model, so the entry qualification of AI large model is scientific and technological strength, and technical competition is inevitable.
According to public information, whether it is Baidu, Alibaba, Tencent, or Huawei, the annual investment in R&D is 10 billion yuan, the number of R&D teams exceeds 10,000, and the strength in technology research and development is strong, and it is currently a well-deserved "first echelon" player.
Taking Huawei as an example, under the pressure of "normalization of sanctions", the company's R&D investment in 2022 will still reach 161.5 billion yuan, accounting for 25.1% of the total revenue.
In terms of technology accumulation related to AI large models, Huawei has a complete ecosystem layout such as chips, computing power, models, and algorithms. As early as 2019 and 2020, Huawei launched the "Pangu Grand Model" project, which learned 40TB of text data in size during the pre-training stage, and the parameter scale reached 100 billion levels.
Coincidentally, Tencent's hybrid model is as deep as the accumulation of technical investment. According to Tencent's 2022 financial report, Tencent's R&D expenditure reached 61.4 billion yuan last year, and since 2019, Tencent's cumulative R&D expenditure has exceeded 182.6 billion yuan.
In the field of artificial intelligence, Tencent has applied 1,063 patent applications in the past five years, ranking first in the global Internet industry. Tencent disclosed the hybrid model in April 2022, and cooperated with Tencent's pre-training R&D strength, covering NLP large model, CV large model, multimodal large model and many industry/domain task models.
Ali Damo Academy has also been deployed in the field of artificial intelligence for many years, forming a full-stack technical capability of "AI + cloud computing" from Feitian cloud operating system, chip to intelligent computing platform, providing a strong computing power, algorithm and data foundation for the training and operation of Alibaba's AI big model.
Big language model AI technology is a hot spot and trend in the field of artificial intelligence today, and for technology giants, mastering this technology means occupying a favorable position in the global market to achieve technological innovation and business breakthroughs. In the face of fierce market competition and many challenges, more technology companies will devote themselves to the research and development and application of large-scale language model AI technology in the future, in order to maintain a leading position in the global technology industry.