Beijing: Despite initial setbacks in developing artificial intelligence (AI) large models, experts believe China has the potential to surpass the United States and establish global leadership in this crucial technology. Recent discussions at the 2024 Large Model Technology and Application Innovation Forum in Beijing highlighted the significant progress and future opportunities for China in the AI landscape.
Key Trends in AI Development
Zheng Weimin, an academician at the Chinese Academy of Engineering, emphasized two major trends shaping AI development this year: the evolution of foundational large models into a multimodal phase and the growing applications of “large models+” across various industries, including finance, healthcare, automotive, and intelligent manufacturing.
He noted that these foundational large models are now integrating multiple data types, such as text, images, and videos, paving the way for more sophisticated AI applications.
The Importance of High-Performance Computing
Zheng underscored the necessity of building domestically produced large-scale systems equipped with specialized AI acceleration chips, such as GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units). Establishing a high-performance computing system is essential for supporting the development and deployment of these advanced AI models.
Zheng outlined the five stages of large model development:
- Data Acquisition: Gathering the necessary data.
- Data Preprocessing: Cleaning and organizing data for use.
- Model Training: Creating the foundational models using the prepared data.
- Model Fine-Tuning: Customizing the models for specific applications.
- Model Inference: Deploying the models in real-world scenarios.
Foundational models are developed during the first three stages, while fine-tuning is crucial for adapting these models to specific domains. For instance, medical applications often require models trained with specific hospital data to ensure accuracy and relevance.
Adapting Models for Industry-Specific Uses
Zheng illustrated this iterative process by highlighting the need to refine foundational models for specialized industries. For example, to create a model suitable for medical scenarios, it is necessary to fine-tune it with hospital data. Further training with ultrasound data can enhance its capabilities for ultrasound applications.
This process is a defining feature of the “large models+” concept, where models are tailored to meet the needs of various industries.
A Shift Towards Practical Applications
Yu Youping, president of Beijing ZKJ Technology Co., a company specializing in large model technologies, remarked that the large model industry is transitioning from a phase of rapid advancement to one focused on “fine-grained implementation.”
He stressed that the market increasingly demands large model applications that provide real solutions to practical problems. The ideal approach involves a combination of platforms, applications, and services, which are essential for effectively implementing enterprise-level large models.
Conclusion
China’s advancements in AI large models reflect a broader shift toward integrating these technologies into various sectors. As the country continues to refine its capabilities and adapt foundational models for specific industries, it is well-positioned to compete on the global stage. With a focus on practical applications and high-performance computing, China aims to solidify its leadership in the rapidly evolving field of artificial intelligence.
