Large Models + AI-Native Applications Bring New Paradigm of Intelligent Production to Industrial Enterprises
-
Recently, relevant authorities have continuously advocated for advancing new industrialization, which brings new development opportunities to the real economy and assigns new roles and era requirements to technology companies.
Yu Shaohua, an academician of the Chinese Academy of Engineering, stated that the deep integration of the digital economy with traditional industrial entities is a crucial path to achieving new industrialization. The resulting new infrastructure, application models, and industrial ecosystems can create an intelligent manufacturing system covering the entire industrial chain and value chain. Li Bohu, another academician of the Chinese Academy of Engineering, believes that intelligent manufacturing is a key breakthrough and focal point for advancing new industrialization.
Baidu argues that the accelerated integration of the digital economy and the real economy requires new infrastructure like cloud computing and artificial intelligence ("cloud-intelligence integration") as a digital foundation, as well as new technologies such as large models to empower traditional industries with new paradigms of intelligent production.
What Makes New Industrialization "New" and Where Are the Challenges?
Unlike the traditional industrialization characterized by "three mechanizations" (mechanization, electrification, and automation), new industrialization is defined by "three new transformations": intelligence, green development, and high-end advancement.
For example, in manufacturing workshops, workers only need to issue verbal instructions, and large models can automatically schedule and direct multiple systems to complete a large volume of transactional tasks. Intelligence allows workers to free themselves from trivial tasks, giving them more time to explore innovation and focus on solving pain points and challenges on production lines. Baidu Intelligent Cloud shared an example: a leading chemical company required multiple rounds of quality inspections to ensure product quality, but due to minimal fluctuations in product stability, some inspections were unnecessary. They approached Baidu to develop a large model for reducing inspection frequency based on historical data, lowering costs, and improving production efficiency. The continuous enhancement of corporate innovation capabilities also helps improve production processes, upgrade China's manufacturing supply chain, and increase its resilience and security.
Green and low-carbon development is another critical aspect of new industrialization, with significant optimization potential on both the energy supply and consumption sides. For instance, at the Jinneng Guofeng Power Plant in Shanxi, Baidu Intelligent Cloud is exploring the use of AI to reduce coal consumption in power generation, lowering the "carbon intensity" per kilowatt-hour. In the traditionally energy-intensive textile industry, AI can reduce electricity and gas costs for each production line. Baidu Intelligent Cloud collaborated with Zhejiang Meixin Group to optimize industrial processes for energy efficiency, helping the company save over 1 million yuan annually.
These proven intelligent solutions are consolidated by Baidu Intelligent Cloud on the Kaiwu Industrial Internet Platform, covering core enterprise scenarios such as quality control, process optimization, safety production, energy efficiency optimization, and intelligent scheduling. The platform includes 38,000 industrial models across industries like 3C electronics, chemical fiber textiles, equipment, chemicals, mining, metallurgy, and power.
Currently, the transformation and upgrading of traditional industries towards new industrialization still face many challenges. Taking enterprise applications of large models as an example, these models are large in size and difficult to train; they require massive computing power and high performance; and they process vast amounts of data with varying quality. The large-scale computation, parameters, and high costs of these models impose new demands on AI infrastructure in the era of large models.
Moreover, enterprises have even higher expectations for large models. Some clients have expressed to Baidu that they expect large models not only to solve isolated problems but also to serve as the foundation for building intelligent central capabilities within enterprises, thereby evolving into comprehensive intelligent applications across all aspects of business operations. For instance, using Baidu's intelligent customer service as an example, clients hope that large models can not only address customer service issues but also serve as office assistants for employees, automate the summarization of corporate regulations, and improve internal enterprise search functionalities.
Large Model Platform + AI-Native Applications Bring New Paradigm of Intelligent Production to Industrial Enterprises
Regarding how to leverage artificial intelligence and large models to promote new industrialization, Baidu believes that the first step is to establish robust new infrastructure. Shen Dou, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud Business Group, stated: "Cloud computing providers need to standardize the output of intelligent underlying capabilities, turning high-barrier technologies such as chips, deep learning frameworks, and large models into utilities that customers can access on demand, much like water and electricity."
Baidu Intelligent Cloud has developed a "Cloud-Intelligence Integration" new infrastructure that provides cost-effective heterogeneous computing services and efficient AI development and operational capabilities. This infrastructure not only helps enterprises build a complete digital foundation but also offers AI computing services with clusters of tens of thousands of cards, enabling enterprises to stably and efficiently complete large model training and inference.
Baidu Intelligent Cloud has also encapsulated complex processes such as model development, training, optimization, and operation to create the world's first one-stop enterprise-level large model platform, "Qianfan." This platform provides low-barrier, high-efficiency large model services and a complete toolchain to various industries. Enterprises can select a general large model on Qianfan and quickly optimize its performance for their specific business scenarios using the platform's pre-installed 41 datasets and 103 high-quality prompt templates. The Qianfan platform also manages 42 mainstream large models from both domestic and international sources, offering capabilities such as Chinese language enhancement, performance enhancement, and context enhancement.
Currently, nearly 10,000 enterprises are active monthly on Baidu Intelligent Cloud's Qianfan platform, covering over 400 scenarios in industries such as manufacturing, energy, and transportation.
Another strategy Baidu is advancing is the vigorous development of "AI-native applications," which involves using large models to "redo" all previous "solutions."
What constitutes an AI-native application? Robin Li, Founder, Chairman, and CEO of Baidu, outlines three essential criteria: First, it must enable natural language interaction—the most fundamental shift. Second, it should fully leverage previously unavailable technical capabilities such as comprehension, generation, reasoning, and memory. Third, every application's interaction flow should not exceed two menu levels.
For industrial enterprises, Shen Dou highlights that large language models (LLMs) will usher in a new paradigm for intelligent production. Employees no longer need to spend excessive time memorizing routine content; they can query the LLM for rapid, high-quality solutions. More importantly, LLMs' powerful comprehension and generation capabilities can integrate knowledge across domains, creating unprecedented "AI-native innovations." With LLMs, innovation is no longer the exclusive domain of geniuses—every employee gains the potential for disruptive innovation.
Recently, China Southern Power Grid demonstrated its AI-native application "Grid Dispatch Assistant," built on Baidu's ERNIE model and powered by their proprietary AI innovation platform for the power industry. During outages, the assistant generates response plans within seconds, meeting the 15-minute power market adjustment requirement. Employees simply issue voice commands to have the assistant categorize safety alerts, generate work orders, and compile shift reports—enabling management to grasp operational status instantly. Another example is "Du An'an," an AI safety assistant adopted by CNOOC's refining subsidiary Dahua Petrochemical, achieving second-level hazard alerts and improving incident response speed by 50%.
Internally, Baidu Smart Cloud has upgraded its Kaiwu Industrial Internet Platform using ERNIE, enhancing knowledge augmentation, semantic understanding, and interaction capabilities. Previously focused on vertical industry scenarios, the new Kaiwu now spans "production line intelligence" to "enterprise intelligence" and "supply chain intelligence," assisting local governments in strategic "chain strengthening and gap-filling" decisions for global optimization.
LLMs have also amplified Baidu Smart Cloud's solution replication capacity. Traditional AI development resembled "artisanal workshops"—rigidly focused on single tasks with minimal reusability. Now, LLMs enable a "factory model," where fine-tuning base models yields tailored industry solutions for multiple scenarios.
Baidu Smart Cloud Kaiwu has been deployed across 18 regions including Suzhou and Guangzhou, serving 220,000+ enterprises by rapidly replicating standardized solutions. Beyond these examples, Baidu will unveil more industrial innovations at "Baidu World 2023" on October 17th, where Robin Li will demonstrate AI-native app development and showcase updates to the Qianfan LLM platform and cloud-AI integration architecture.