Current Status Domestically and Internationally:Globally, China and the United States are leading the development in the field of large models. The United States, with its leading edge in algorithmic model development, has the highest number of large models globally, while China is actively following the global trend of large model development, accelerating output. As of May 2023, the United States has released 100 large models with more than 10 billion parameters, and China has released 79 large models, occupying a pioneering advantage globally. Domestically, large models in China have entered a critical period of performance enhancement, with model applications gradually unfolding, and some models' capabilities in Chinese are nearly on par with international advanced levels.
Competitive Landscape:In the global competition of large models, OpenAI, Anthropic, and Google form the first echelon, with OpenAI leading the industry with the launch of GPT-4. Domestically, there is a lack of distinguishing technical features among models, with internet companies and tech firms starting early in large model development, such as Baidu, 360, Alibaba, and iFlytek. The market size of China's large model industry has grown rapidly from 2020 to 2023, with a compound annual growth rate of about 114%.
Future Development Trends
Diversity in Technical Paths:The path to AGI (Artificial General Intelligence) is diverse, with large models currently being the best implementation method. Large models possess strong generalization, universality, and practicality, which can reduce the development threshold of AI, improve model accuracy, and enhance generalization capabilities.
Development of MaaS Model:MaaS (Model as a Service) will become a new form of AI application and develop rapidly, reconstructing the commercial structure and ecosystem of the AI industry, stimulating new industry chain division of labor and business models.
In-depth Industry Applications:Large models will be deeply applied in user lifestyles and enterprise production models, unleashing creativity and productivity, enlivening creative thinking, reshaping work patterns, aiding corporate organizational changes and operational efficiency, and empowering industrial transformation.
Challenges in Security and Control:Large models face challenges such as high computational demands, high training and inference costs, and poor data quality. The success of a commercially viable large model requires the manufacturer to possess full-stack large model training and development capabilities, business scenario landing experience, AI security governance measures, and ecological openness.
Oligopolistic Competitive Landscape:By 2025-2026, the market is expected to clear out, with dominant companies holding the field while others exit the competition. By 2027-2028, the general large model market is expected to present an oligopolistic competitive landscape, with 3-5 manufacturers becoming the foundational pillars of China's large model ecosystem.
In summary, the development of general large models is in a period of rapid change and fierce competition. Over the next few years, we can expect to witness technological advancements, deeper applications, and further clarification of market dynamics.
--The above article is purely AI-generated. If there is any infringement, please contact us to delete it!