Harnessing the economic potential of AI for industries with LLMs
Key players insights
By Dafu SHI, Senior AI Specialist at Huawei France
AI is poised to become a game-changer for the global economy and has already started influencing businesses in every domain, with AI adoption expected to reach 20% by 2026 in major industries. According to PwC’s Global Artificial Intelligence Study, AI could contribute up to $15.7 trillion to the global economy by 2030. However, AI is still in its infancy as a business tool and we – as an ecosystem – need to consider how to ensure AI brings benefits to society by ensuring it continues to expand its economic, business and social value.
As part of its research into AI, Huawei has been exploring how to take AI models, more specifically large language models, and make them much more accessible for organizations of all sizes, and across all industries. The company has worked on "AI for Industry", which uses industry-specific large models to create more value. Industries face many challenges when it comes to AI application development. They need to invest a huge amount of manpower to label samples, find it difficult to maintain models, and lack the necessary capabilities in model generalization. Most simply they do not have the resources to do this.
Our answer is called Pangu. Pangu is a set of pre-trained large-scale language learning models, designed and developed to meet the specific needs of individual industries and help them enable progress for us all. Pangu has a three-layer model architecture with different capabilities at different layers. L0 consists of five foundation models, which are NLP model, CV model, multi-modal model, prediction model, and scientific computing model. They provide multiple skills to meet requirements in industry scenarios. L1 contains many (N) industry-specific models. They can be general industry models (healthcare, meteorology, electric-power, etc.) trained on public industry data. Or, users can train their dedicated models with their own data based on L0 models and L1 general industry models. L2 contains scenario-specific models, which are focused on specific application scenarios or services (designing new drugs, automating railway planning, predicting typhoon routes, detecting foreign objects on a conveyor belt, etc.).
This three-layer structure lowers the barrier to AI deployment for organizations. It solves problems caused by fragmented scenarios, and transforms the workshop-scale AI model development to factory-scale development. Pangu also offers a set of automated tools to users, so that they no longer need to manually design and develop new scenario-specific models from scratch. Instead, they can be automatically generated after being fed with relevant data, which greatly simplifies the development. Because a Pangu model trained for one scenario can be applied to another scenario, it reduces the need for repeated training – a time-consuming and costly phase for organizations.
Huawei allows customers to develop, run, and manage their scenario-specific models not only on hybrid, or public clouds, but also a private cloud. It empowers organizations to train AI models while maintaining complete control and ownership of their data inside their organization. This approach opens the door for broader AI adoption, even among companies concerned about data privacy.
This innovation is possible because of the heavy investment that Huawei has put into the space. The company has invested nearly $135 billion in R&D over the last decade, and more than half of its workforce is directly employed in this area.