How Large AI Models Are Powering the Next Industrial Revolution

On November 5, the sub‑forum “Artificial Intelligence Empowering New‑type Industrialization” opened at the 7th China International Import Expo in Shanghai, where Baidu Technology Committee Chair Wu Hua delivered a keynote speech.

Wu described the learning process of large models as analogous to human learning, involving both an "innate" stage—building neural‑network parallel algorithms that mimic a language‑cortex—and a "post‑natal" stage where the model refines its understanding through environmental feedback, thereby acquiring capabilities in comprehension, generation, reasoning, and memory.

The speaker highlighted three core advantages of large models: excellent performance, strong cross‑task and cross‑modal generalization, and a high degree of standardization in the research‑development workflow, which together accelerate AI progress and enable new productive forces.

Using Baidu’s Wenxin large model as a case study, Wu explained that the model integrates trillions of data points and billions of knowledge items, employing knowledge internalization and externalization techniques that markedly improve both effectiveness and efficiency. The model has broken through traditional agent architectures by supporting slow‑thinking tool‑calling, allowing it to plan, reflect, and evolve autonomously during interactions.

Since the launch of Wenxin 1.0 in March 2019, the platform has progressed to Wenxin 4.0 Turbo. It now serves over 300 million users, processes more than 7 billion daily calls, and handles over 1 trillion text tokens each day.

Large‑model technology is injecting new momentum into numerous industries. In smart transportation, Wenxin‑based event detection achieves over 95% accuracy, boosting emergency response and information dissemination efficiency by 90%. In intelligent creation, the model powers more than 100 multimodal AI capabilities in Baidu Wenku, accumulating over 2.2 billion uses. In smart finance, an AI agent automates cross‑platform reimbursement tasks, completing them in seconds and raising efficiency by 83% while offering 24/7 service.

The speaker also noted that AI is reshaping scientific research paradigms; for example, deep‑learning‑driven protein‑structure prediction accelerates drug discovery and biological studies.

Concluding, Wu emphasized that large‑model technology is continuously unlocking potential across sectors, fostering a new quality of productivity that contributes significantly to economic and societal development.