Tongyi Qianwen Open Sources 72-Billion-Parameter Version, Bringing Large Model Development Closer to Industries
-
At the Yunqi Conference on October 31, Alibaba Cloud CTO Zhou Jingren announced the open-sourcing of a 72-billion-parameter version by the end of this month or early next month.
Large model technology refers to the use of large-scale neural network models to complete various artificial intelligence tasks through language-based interactions. From the perspective of application types, domestic large models show a development pattern where industry-specific models dominate, indicating that China's large model development is more closely aligned with industrial needs.
After nearly a decade of AI industry development, China has seen the emergence of numerous AI-powered enterprises specializing in vertical industries such as finance, healthcare, education, and manufacturing. By deeply engaging in customer business processes and gaining thorough industry knowledge, these companies have gradually built industry-specific barriers.
AI Large Models Spanning Diverse Industries
This year, tech companies worldwide have been competing to expand their commercial footprints around large models. Statistics show that by the end of July, China had released a total of 130 large models, with 64 of them launched in the first seven months of this year alone. Domestic large models are entering a period of rapid development, accelerating digital and intelligent transformation across industries.
Current State of the AI Market
Artificial intelligence is a technology that simulates human intelligence, encompassing fields such as machine learning, natural language processing, and computer vision. It has broad applications across industries like healthcare, finance, transportation, and manufacturing. AI technology continues to evolve and innovate, bringing new opportunities and challenges to various sectors.
AI refers to the ability of computers or machines to exhibit human-like intelligence, including capabilities such as understanding and processing language, decision-making, and problem-solving.
AI solutions are intelligent systems tailored to specific industry scenarios, effectively integrating multiple AI technologies. They can redefine existing operational models, deliver significant business value, and create market opportunities for AI development.
The AI industry chain is divided into three layers: the foundational layer (data, computing power, and algorithms), the technical layer (building AI systems for perception and analysis, including computer vision, speech recognition, natural language processing, and machine learning), and the application layer (integrating technical capabilities with real-world scenarios to improve efficiency in areas like security, finance, healthcare, and autonomous driving).
Recently, the first batch of large model products approved under China's Interim Measures for the Management of Generative AI Services have been launched, offering services to the public. As of July, over 100 large model products have been released in China.
Meanwhile, many large model products are awaiting regulatory approval. For example, in Beijing—a hub for large model innovation—over 40 models and products have undergone self-assessment for safety and submitted materials for filing, according to Xu Yali, Deputy Director of the Beijing Cyberspace Administration.
Amid the AI large model wave, tech companies are investing heavily in developing their own models. The most pressing task now is figuring out how to recoup these investments.
Not every company can afford to develop general-purpose large models, as they require significant data handling and computing power. From a practical standpoint, what customers truly need are not general models but industry-specific solutions that address real-world problems.
The heating up of the large model market is also driving the upstream and downstream industrial chains, including data, computing power, and algorithms. Professionals in related fields, such as large model engineers, are in short supply domestically, and their income levels continue to rise.
According to statistics from relevant institutions, the average annual salary of artificial intelligence engineers and algorithm scientists in China has exceeded 600,000 yuan. Moreover, there is strong demand for large model technology R&D talent, with significant gaps in some core areas.
The growth of the large model market is also boosting the computing power and algorithm industries, including chips, computing clusters, servers, communication network hardware, as well as software such as cloud computing, databases, and virtualization functions. Computing infrastructure and services are indispensable for large model training, as their performance, quality, and stability determine the effectiveness of model training.
Analysis of the Current State of the Large Model Market
This year, domestic and international tech companies have been competing to expand their commercial applications of large models. At the Hangzhou Asian Games Digital Technology Experience Center, visitors can use the art installation "Imagining the Future Asian Games" to generate Asian Games-themed images with Alibaba Cloud's painting creation model, "Tongyi Wanxiang."
"Alibaba Cloud was one of the earliest tech companies to invest in large model R&D, starting its layout in 2019," said Zhou Jingren, CTO of Alibaba Cloud. This year, Alibaba Cloud has successively launched multiple self-developed Tongyi large models and applications, which can process or generate multimodal content such as text, speech, and images, and are widely used across industries.
On September 21, Huawei Cloud released the Pangu Medical Model at Huawei Connect 2023. This model has clinical assistance capabilities and will effectively support applications such as medical testing, clinical diagnosis, and personal health management. Zhang Ping'an, Executive Director and CEO of Huawei Cloud, introduced that the RunDa Medical Model·Liangyi Xiaohui, developed based on the Pangu model, will be deployed in 4,000 hospitals to assist doctors in report interpretation and diagnosis.
On September 5, the 2023 Baidu AI Cloud Conference announced that nearly 10,000 enterprises are active monthly on Baidu Intelligent Cloud's Qianfan Large Model Platform, covering over 400 scenarios in industries such as finance, manufacturing, energy, government, and transportation.
Forecast and Analysis of the Large Model Market Size
Domestic large model applications primarily focus on industry-specific use cases, shifting from general-purpose models to industry-specific models. In customer service, large models significantly enhance the conversational abilities of service robots and improve user satisfaction. In recruitment, graduates use AIGC to write resumes, simulate interviews, and recommend job positions, while HR professionals leverage large models to post job descriptions, screen resumes, and organize interviews.
In design, fashion designers use AIGC to generate vast numbers of clothing designs in minutes, and home designers provide unlimited inspiration images for homeowners. In smart manufacturing, large models reduce training costs for industrial quality inspection, shorten deployment times, solve cross-model multi-category defect generation and automatic labeling issues, and enable batch replication of AI solutions.
In the first half of 2023, the global AI sector recorded 1,387 financing deals, raising $25.5 billion, with an average funding amount of $26.05 million. In China, about 20 large model companies disclosed investments, with funding typically ranging from tens of millions to hundreds of millions of yuan.
Among them, the startup MiniMax received the most funding, completing a new round of over $250 million on June 1, valuing the company at over $1.2 billion.
Projections indicate that the global large model market will reach $21 billion in 2023 and grow to $109.5 billion by 2028. Meanwhile, China's large model industry is expected to reach 14.7 billion yuan in 2023 and expand to 117.9 billion yuan by 2028.