Today, the 2023 World Artificial Intelligence Conference opened in Shanghai. Ken Hu, Huawei's Rotating Chairman, delivered a keynote speech entitled "Win-Win AI in a New Era". He believes that the key to the development of artificial intelligence is to "go deeper and real" and empower industrial upgrading. At the current stage, Huawei has two focus points in the field of artificial intelligence: First, build a strong computing power base to support the development of the Chinese industry of industrial intelligence. Second, from general large models to industry large models, let artificial intelligence serve thousands of industries and serve scientific research well.
Full text of the statement
The emergence of ChatGPT has made artificial intelligence the hottest topic at present, and the whole society has unprecedented expectations for the role that artificial intelligence can play. We firmly believe that in the near future, artificial intelligence, especially artificial intelligence in general, will rewrite everything around us.
Promote artificial intelligence to the deepest
For Huawei, we must make every effort to promote artificial intelligence in the next stage. In order to achieve this goal, we have two key measures: First, build a strong computing power foundation to support the development of the Chinese intelligent industry. Second, from general large models to industry large models, truly let artificial intelligence serve thousands of industries and serve scientific research.
Deeply cultivate computing power and create a solid computing power foundation
In the development of artificial intelligence, computing power is the foundation. However, in China's current situation, computing power faces considerable challenges in terms of accessibility and cost. Over the years, Huawei has been deeply engaged in computing power, focusing on the root technologies of Kunpeng and Ascend, and supporting the construction of future computing power foundations through architecture innovation, ecosystem development, and flexible co-construction. We hope that through the joint efforts of everyone, computing power will no longer become a bottleneck in the development of artificial intelligence.
Redefine the computing architecture and break through the bottleneck of computing power
Huawei improves computing efficiency through architecture innovation. At the computing node level, we launched a revolutionary peer-to-peer flat architecture to break through the performance bottleneck caused by traditional CPU-centric heterogeneous computing, thereby improving the bandwidth of the entire computing and reducing latency, resulting in a 30% improvement in node performance.
At the data center level, we launched the Ascend AI cluster in 2019, which leverages the comprehensive advantages of cloud, computing, storage, networking, and energy, which is equivalent to designing the AI data center as a supercomputer, making the Ascend AI cluster more performant and more reliable. At present, the largest AI computing cluster we are building in China is in Shenzhen Pengcheng Cloud Brain Phase II, the current computing power is 1000P scale, according to the plan, by 2024 Phase III, the scale will reach the level of 16000P. At the same time, at Huawei's computing center in Ulanqab, we deployed a scale of several kilocards, and found that through clustering, efficiency improvements of more than 10% can be obtained with the same computing power.
Adhere to open collaboration and activate the computing power ecosystem
The ecosystem is critical to the sustainable development of the computing industry, and it is often a bottleneck. Four years ago, Huawei put forward the strategy of "open hardware, open software, enabling partners, and developing talents" to focus on the development of the computing industry, and over the past four years, with the joint efforts of partners in the entire industry, Huawei has initially built a complete computing industry ecosystem.
In terms of hardware, we insist on further opening up, launching more diversified modules and boards this year, and more than 30 hardware partners have launched hundreds of artificial intelligence hardware products based on Ascend AI to meet the differentiated needs of different industry scenarios.
In terms of software, we insist on strengthening basic software through open source. For the innovation of current large models, it provides a full-process enabling platform, supports scientific research institutions and enterprise customers, natively incubates more than 20 basic large models, and adapts to more than 10 mainstream large models in the industry. Half of the current Chinese models are supported by Ascend AI. We would like to thank the companies and institutions that choose Ascend AI computing power for their trust in Huawei and the confidence they have given us.
Multiple modes to jointly build computing power
In China, customers' computing power needs are diverse, and we adopt a variety of modes to build computing power based on the actual situation.
In terms of urban computing infrastructure, Huawei supports local governments in building AI computing centers to provide inclusive computing infrastructure services. At present, 25 cities in China, such as Shanghai, Wuhan, Xi'an, etc., have built artificial intelligence computing centers based on Ascend AI.
On the other hand, for large enterprises with the demand for self-built artificial intelligence computing power centers, we can help them build independent computing power centers. At present, enterprises such as China Mobile, iFLYTEK, and China Southern Power Grid are planning and building large-scale computing power clusters, and Huawei is actively participating in them.
At the same time, more small and medium-sized enterprises also have a strong demand for AI. We can also provide AI computing services on HUAWEI CLOUD, so that these enterprises can quickly and agilely develop and apply AI, which can be obtained directly on the cloud and used as they go.
From a general-purpose large model to an industry large model
While deeply cultivating computing power, we must also truly bring artificial intelligence into thousands of industries and serve scientific research. We believe that, on the one hand, we must lay a good foundation and continuously improve the capabilities of general large models, on the other hand, on this basis, we must build a good industry model, fully combine industry know-how, experience and large model capabilities, and provide more professional and accurate solutions for end customers.
For example, in response to the question "I live in Futian District, a 78-year-old man in my family, can the government provide any subsidies?" "When it comes to the same problem, the performance of the general large model and the industry large model is different. The general big model will give some correct but general information, but the industry big model can give more accurate and valuable answers, which is also the direction we want to work towards.
AI for Industry, from reading 10,000 books to traveling 10,000 miles
For the first time, Huawei proposed a three-layer large-model architecture to build different capabilities at different levels.
The lowest basic model, do a good job of learning massive basic knowledge, equivalent to "reading ten thousand books" and laying a good foundation;
On top of this, for different industries and different scenarios, special knowledge and experience training is carried out to create easy-to-use and easy-to-use industry models and scene models, which are equivalent to "traveling thousands of miles".
From "reading 10,000 books to traveling 10,000 miles", there are still many challenges to overcome. The key is to match the industry knowledge of customers and industry partners with the big model, solve the integration of scenarios, technologies, algorithms and data, and let the big model play an important role in the value creation of the industry.
At present, HUAWEI CLOUD Pangu models have penetrated into more than 10 industries, including finance, manufacturing, government affairs, electric power, coal mining, healthcare, and railways, supporting the implementation of AI applications in more than 400 service scenarios. In the future, we hope to join hands with more industry partners, land big models in more industries, go deeper, and truly go deeper and practical.
AI for Science, powering scientific research
Huawei hopes that the big model can not only be used in factories, ports, banks, and other industry scenarios, but also in laboratories and research institutes to help scientific research.
By learning massive historical data, scientific knowledge, and encoding mathematical equations into large models, AI can promote the combination of AI and basic disciplines, such as molecular dynamics, fluid mechanics, heat transfer, biology, etc., so as to discover more scientific laws and unlock more codes in nature.
Huawei's Pangu Scientific Computing Model currently includes the Drug Molecule Model, the Pangea Meteorological Model, and the Wave Model. We work with scientists and make great progress. For example, in the field of weather forecasting, the prediction of the Pangu model can complete the weather forecast of the future global weather from one hour to seven days in seconds, and it is fast and accurate. There are two keys to this, one is to fully study the world's 40 years of meteorological data and lay a solid foundation, that is, "read ten thousand books"; The second is to train on this basis, and then to the industry, that is, "travel thousands of miles". For example, to predict the path of a typhoon, the results of the large model deduced by the equation must be calibrated with the real historical typhoon path, and each calibration requires the participation of algorithms and scientific experts to find the best match through continuous tuning.
We hope that AI for Science can bring more new ideas, new methods, and new tools to scientists and scientists, and also inject new impetus into our industry.
Pangu Grand Model 3.0 is coming soon
Tomorrow afternoon at the Huawei Developer Conference, HUAWEI CLOUD will release Pangu Model 3.0, providing unlimited possibilities for innovation in thousands of industries.
We are very fortunate to witness several waves of scientific and technological revolution, from the Internet, to mobility, from cloud computing, to intelligence, each round of change has a far-reaching impact. There is no doubt that General Artificial Intelligence (AGI) is ushering in the next golden decade, let us join hands to innovate together, serve thousands of industries, serve scientific research, and win a new era of artificial intelligence.