Overview
The new generation of artificial intelligence represented by large models is a strategic area of technological development, a focal point in international competition, and a new direction for industrial development. Manufacturing is a core sector of the economy and a primary domain for applying large models. Manufacturing digital transformation is progressing toward version 3.0, where industrial large models combined with industrial internet, industrial software, and intelligent equipment are forming a new foundation for digital, networked, and intelligent manufacturing.
Report context
Yuan Xiaoqing, director of the Security Industry Research Institute at the China Electronics and Information Industry Development Research Institute, presented the report "Five Fundamental Issues of Industrial Large Models" at the 2024 CCID Forum subforum on digital empowerment. The report frames its findings in a one-two-three-four-five-six structure.
Key points
1. Industrial large models are a new ontological entity
Industrial large models are an emergent industrial form produced when large models are applied to industrial use cases. They are an important carrier for manufacturing digital transformation 3.0 and represent a new kind of entity in the industry.
2. Two main technical routes
There are two primary technical routes. One route is general-purpose large-model vendors performing fine-tuning using industrial knowledge. The other route involves domain-specialized smaller models developed by industrial internet companies, industrial software vendors, and equipment manufacturers, which continue to evolve through parameter upgrades and other improvements.
3. Three main model categories
Industrial large models can be classified into three categories: general industrial large models, which aim to understand and manage broad industrial knowledge; industry-specific large models, which serve vertical industrial sectors; and scenario-focused large models, which address specific, narrowly defined industrial scenarios.
4. Four-layer technical architecture
The technical architecture for industrial large models mainly includes four layers: data as the foundation, computing power as the driving force, models as the core, and applications as the key for deployment and value realization.
5. Five typical application scenarios
Typical application scenarios span five categories: production-oriented R&D, precision manufacturing, intelligent-assisted management, value-added services, and embodied devices. Deployment is expanding from single process enablement in R&D, production, management, services, and equipment toward multi-scenario coverage across vertical industries.
Additional: Five industrial trends
Five major industry trends associated with industrial large models are emerging: data assetization, integrated computing power, model fusion, advanced application development, and networked ecosystems.
6. Six policy and industry recommendations
Recommendations include optimizing policy, supporting technical development, promoting industry growth, expanding application scenarios, developing standards, and cultivating ecosystems.
