Frost & Sullivan releases the 'White Paper on Excellent Cases of Large Model Industry Applications in China 2024'

The demand for digital transformation and intelligent upgrading in various industries has surged, and national and local governments have successively introduced supportive policies and action plans to vigorously promote the development of AI large models.

As an innovative technology application, large AI models receive strong support from the national level. On one hand, national policies provide support and guiding suggestions for the development of related technologies and industrial applications of large models; on the other hand, they gradually improve and standardize regulations regarding the security and compliance of large models and industry standards. At the same time, local governments have successively introduced supportive policies and action plans in response to national policies to accelerate the regional and sustainable development of the large model industry.

In order to actively respond to policies, cope with technological changes, and achieve intelligent upgrades, enterprises across various industries in China have begun to explore and practice the empowerment of AI for their businesses, actively promoting the deep application and implementation of large AI models in various scenarios. At the same time, enterprise users have put forward higher requirements for the accuracy, implementation effect, development and deployment efficiency of AI large model applications. In multiple fields such as finance, manufacturing, and healthcare, traditional industries are actively seeking cooperation with new technology companies, investing a large amount of resources to jointly develop specialized AI large models for industry-specific applications.

The AI large model industry faces both potential application opportunities and challenges. How to balance supply and demand and solve technical and application challenges is key to the industry's future development.

The AI large model industry application market has great potential. After 2020, with the explosive growth of internet data, larger-scale pre-trained models began to be explored, and domestic large models began to emerge at this stage. Despite the huge market potential, the challenges of technology and application cannot be ignored. Technological progress on the supply side provides strong support for the development of the industry, while the demand-side's digital transformation needs provide momentum for market growth. How to balance supply and demand, solve the challenges of technology and application, is an important factor in ensuring the successful application and continuous development of large models in actual business.

Supply Side: Technological Progress and Innovation-driven Development

• The rapid development of AI large models benefits from the comprehensive upgrade of technical architecture. The infrastructure layer has achieved a reconstruction of the new generation of infrastructure around high availability, high scalability, and other needs, providing stronger hardware support and more efficient distributed training algorithms for large models.

   

• The development of model layers includes innovative optimizations in model architecture, data processing, model optimization and compression, such as multimodal fusion and Transformer architecture optimization. These optimizations further improve the basic performance of models and reduce inference costs. The functions at the application layer have significantly enhanced the performance of large models in professional fields, achieving more efficient and accurate processing.

   

• At the AI application layer, the emergence of Agent, RAG, large model fine-tuning, and prompt engineering has significantly improved the performance of large models in professional domain tasks. On one hand, through meticulous model adjustments or the use of external tools, these technologies enable large models to better adapt to specific task requirements and achieve notable performance improvements. On the other hand, by optimizing prompts, they guide large models to generate content that more closely aligns with user expectations, thereby enhancing the quality of generated text.

Demand Side: There is an urgent need for industry digital transformation

• The demand for digital and intelligent transformation across industries has driven the growth of the market scale of industry large models in China. In 2023, the market scale reached 10.5 billion yuan, and it is expected that in 2024, the market scale will reach 16.5 billion yuan, a year-on-year increase of 57%. Artificial intelligence empowers various fields of economic and social development, and the industrial upgrading of downstream industries drives a continuous rise in demand for large models. Large models are expected to continue to penetrate downstream fields and achieve large-scale implementation applications.

   

• Currently, the adoption of AI large models in various industries is concentrated in the exploration and incubation phase, as well as the experimental acceleration phase. The adaptability of demand and the availability of data are the core influencing factors for progress. Digital-native industries such as internet applications have been pioneers, while traditional industries like financial services have seen relatively rapid progress. Heavy asset industries such as construction have progressed more slowly. As industries further integrate large models, there is significant room for market demand release.

Multi-modal and agent development has advanced significantly, with AI large models shifting from technical competitions to industry demand-driven development.

• Multimodal modalities endow large models with emotional value. Multimodal large models combine different types of data (such as images, text, videos, etc.) for analysis and processing. By correlating and integrating different data types, they significantly improve the model's accuracy and robustness, further expanding application scenarios. Currently, multimodal large models have become a cutting-edge direction in the development of large models. After 2023, large models will gradually develop from single-modal tasks such as text to support multimodal tasks, which is more in line with how humans perceive the world. Therefore, large models will inevitably move towards multimodal capabilities to achieve intelligent decision-making in complex scenarios.

   

• Agents inject productivity into various industries. With the development of technology, AI agents are redefining the way users interact with digital systems. Compared to traditional AI applications, agents have the ability to gradually achieve given goals through independent thinking and tool invocation, and are expected to become an indispensable new form of productivity in various industries. In the future, agents will develop towards a new paradigm of actively observing the environment and inferring tasks to execute.

   

• The focus of technology competitions is shifting towards business application demand-driven development. In the future, AI large models will adopt a development model that combines general-purpose and specialized capabilities. The powerful generalization and adaptability of general-purpose models, combined with the precision and specialization of specialized models, will present a symbiotic relationship that supports each other and collaborates with each other. They complement each other in terms of application accuracy and breadth, empowering various industries to achieve optimal business results.

The core technology of AI large models and enterprise-level high-standard requirements jointly determine the competitiveness of large model enterprise-level product applications.

At the core capability level of AI large models, with the deep integration of AI models into various industry scenarios, in addition to model security compliance, user interaction, and basic performance, scenario adaptability has become one of the indispensable capabilities for industry large models. Scenario adaptability requires large models to have sufficient knowledge information in their professional fields to complete professional tasks with high quality.

In terms of enterprise-level user needs, users mainly face challenges in four aspects during the implementation of large model products: economy, professionalism, security, and sustainability. Therefore, comprehensive implementation guidance, advanced product architecture, comprehensive security governance, and open ecological support have become the main considerations for users when choosing large model products.