How AI Large Models Are Revolutionizing Enterprises – Key Insights from Huawei Cloud

As AI technology continuously breaks new ground, an increasing number of industries are applying AI to improve efficiency, reduce costs, and enhance user experiences.

In the past year, large‑model AI has evolved at an unprecedented speed. Reviewing its enterprise applications raises questions about the changes it brings, how developers should respond, and the future direction of AI‑driven businesses.

During the Huawei Cloud DTT Annual Closing Ceremony, experts including Xu Yi, Xia Fei, Zuo Wen, Xiao Fei, Xu Weizhao, and Zhou Rulin shared their perspectives on AI large‑model practice.

AI Large‑Model Practice in 2023 saw breakthroughs not only in model scale and performance but also in diverse real‑world uses. Xia Fei highlighted that large models improve the fluency and naturalness of intelligent customer service, enhance multi‑turn context understanding, and are widely used for generating government documents, e‑commerce copy, medical reports, and even code or front‑end UI, greatly boosting productivity.

These examples demonstrate the massive potential of large models to increase enterprise efficiency and productivity. To harness this potential, companies need to understand the underlying technology, select appropriate tools, and explore new capabilities.

Xu Weizhao described the large model as a "brain" that can be optimized through better interaction design.

Precisely describe problems and structure them clearly for the model.

Assign different roles to the model and use retrieval‑augmented generation to enhance memory.

Teach the model to use external tools such as calculators and search engines to improve computation and real‑time data access.

Digital humans are a prime example of large‑model impact. Large models accelerate model construction, enable voice cloning with minimal samples, and improve overall quality. Zuo Wen noted that these advances shorten development time and enhance emotional expression in synthetic voices.

Developer competitions foster innovation. Zhou Rulin’s team used the MindSpore framework to create a sleep‑breathing diagnostic solution, winning national and regional awards.

In healthcare, large‑model hallucinations pose risks. Zhou emphasized the need for rigorous validation, ethical considerations, and safeguards to protect patient interests.

Data challenges such as isolated datasets were addressed through federated learning, reinforcement learning, and unsupervised pre‑adaptation techniques.

Practical concerns for deploying large models include limited compute resources, data quality, high costs, and a shortage of expertise. To mitigate these, Xiao Fei suggested using more interpretable model architectures, combining knowledge graphs with LLMs, and enhancing transparency.

Data security is critical. Measures include encrypted storage and transmission, de‑identification, and anonymization to protect sensitive information.

Compliance for digital humans requires adhering to regulations, obtaining user consent, implementing robust security controls, and establishing content review mechanisms.

When preparing data, Xia Fei noted that a few hundred to a thousand high‑quality examples can be sufficient, as excessive data may cause the model to forget prior knowledge.

Large‑model technology opens vast opportunities for enterprise innovation and developer growth. Companies must build technical and talent foundations, leverage proprietary data for fine‑tuning, and explore domain‑specific scenarios.

AI serves as a productivity catalyst, reshaping human‑machine interaction and creating demand for AI‑agent talent. Platforms like Huawei Cloud ModelArts provide end‑to‑end AI development services, including pre‑installed large models and free compute resources.

While large models drive cost reduction and innovation, they will not fully replace roles such as product managers. Instead, they augment tasks like digital‑human creation, marketing, video production, and live streaming, where human creativity remains essential.

AI developers continue to be indispensable, offering deep technical expertise, solving complex problems, and delivering customized solutions.

In conclusion, large models represent a major technological shift. As more enterprises and developers join this field, collaborative efforts will refine industry‑specific applications and solidify the role of large models in future technology landscapes.