After the hustle and bustle of the first half of the year, the big model craze triggered by ChatGPT is going through its first cooling-off period. In June, ChatGPT traffic fell for the first time monthly, and the market share of the new version of Microsoft Bing, which connected to ChatGPT's chat function at the beginning of the year, also fell recently, even lower than before the revision.
These signs are that the "watching the excitement" stage of the big model track has passed, and the market is putting forward higher requirements for the practicality of large models. Compared with the warm welcome of the consumer market, people are now more concerned about how the big model is actually applied in the commercial industry, and the large model that can only "chat" can no longer meet the market demand.
A new competitive landscape is emerging. At the just-concluded Artificial Intelligence Conference, a series of large models for different industries and scenarios began to emerge. Huawei, Tencent, Alibaba, iFLYTEK, etc. are all working hard to make the big model land at the commercial level. Compared with the selling point of "writing poetry and painting" in the past, people are now more concerned about how to let the big model help users solve real problems.
It can be said that the participants of the mainstream big model are finally starting to get their feet on the ground and ready to do practical things.
As the earliest technology giant in China to invest in the research and development of large models, Huawei launched Pangu Large Model 1.0 as early as 2021. However, in this year's industry boom, Huawei has not launched its main products. It was not until the World Artificial Intelligence Conference on July 6 that Ken Hu, Huawei's rotating chairman, officially announced the release of Pangu 3.0, and mentioned that the key to the future development of artificial intelligence is to "go deeper and deeper" and empower industrial upgrading. On July 7, at the HUAWEI CLOUD Developer Conference (HDC2023), HUAWEI CLOUD officially released version 3.0 of the Pangu model.
Unlike ChatGPT, Pangu 3.0 is not a large model that focuses on "chat", and Huawei even said that the Pangu large model will not be open to individual users for a period of time, which is not the main direction of the product. Although Huawei did not disclose how long this time was, it could at least confirm that "chat" was not the focus of Pangu's large model development.
"We have never compared to ChatGPT, we have not called Pangu Chat, nor Chat Pangu, we have no time to chat." Zhang Pingan, Executive Director of Huawei and CEO of HUAWEI CLOUD, mentioned in a media briefing on July 7.
Huawei said that the Pangu 3.0 model is not a single large model, but a general term for a series of large model clusters and engineering application platforms, which are divided into three levels, including the general large model of the bottom layer (L0), the industry large model of the second layer (L1), and the subdivision scenario model of the third layer (L2).
It can be said that when the participants in the big model track are competing to compete with who is better at writing poetry and painting, Pangu 3.0 has chosen a new path, and its focus is not only on the iteration of general capabilities, but also on the evolution of professional capabilities to meet the diverse needs of different industries and scenarios.
Huawei has clearly realized that if the big model wants to really land, it must solve the actual needs. Large models must have highly specialized practical capabilities in different industries and scenarios to survive.
Globally, generative AI is evolving at an eye-popping pace. Many tech giants are waiting for this technology to cross a specific intelligent node and thus completely change the way the entire world is produced. In this big model team battle, Huawei, as the last domestic technology giant to enter this track, chose to start from the To B market, which it is best at. They understand that although the To C market may seem lively, the big model must go deep into reality before it can be implemented at the commercial level. Huawei's big model, Pangu 3.0, is not only for entertainment, it is to create real value.
Pangu 3.0 is a major upgrade in Huawei's large-scale model development. Completely different from the route taken by ChatGPT, Pangu 3.0 has been polished and upgraded for three years. In terms of architecture, Pangu 3.0 innovatively adopts a three-layer architecture, including a general large model, an industry large model and a scene model, each of which has its own unique functional characteristics. In terms of training methods, Pangu 3.0 has also upgraded a set of training modes from general to specialized, including pre-training, special training and reinforcement learning.
In addition, Pangu 3.0 is also the industry's first fully hierarchical decoupled large model cluster, and its capabilities can operate independently without interfering with each other to meet the customized needs of different industries and customers. This layered decoupled design allows industry customers to choose different capabilities of large models according to their needs, just like drug selection.
The core positioning of the Pangu model is to empower various industries. To achieve this goal, they have invested heavily in data and computing power. In terms of data, they used more than 3 trillion tokens and over 1000+ terabytes of data for training. In terms of computing power, they trained based on Huawei's Ascend AI computing power cluster, which improved the efficiency of model training.
Under multiple innovations, Pangu Grand Model has achieved industry leadership in multiple capabilities. For example, their NLP large model is the industry's first Chinese model with hundreds of billions of parameters, with powerful text understanding and generation capabilities. Their CV large model takes into account image discrimination and generation capabilities for the first time, and has reached the industry's highest level of small-sample classification accuracy on ImageNet's 1% and 10% datasets.
Overall, Huawei's Pangu 3.0 model is a revolutionary iteration. It's not just a chatbot, but a big model that can dive into reality and provide solutions for all walks of life. In the future, we look forward to seeing more large models like Pangu 3.0, which can better serve industrial and agricultural societies and promote the development of the whole world.
Deepen industry applications and realize scenario implementation
In the global competition for large-scale models, industrialization has become a key focus. Who really understands the needs of the industry and solves the problem will determine who will commercialize the big model first.
Huawei, as the world's largest manufacturer of communications equipment, has decades of accumulation in the government and enterprise market, giving it a huge advantage in industry depth. In recent years, Huawei has established 20 large armies to penetrate into industries such as mines and coal shafts to further serve government and enterprise customers.
Zhang Pingan, CEO of HUAWEI CLOUD, mentioned in a media interview on July 7 that Huawei's biggest advantage is that its scientists and mathematicians can go deep into actual work scenarios and solve practical problems in the industry. They dare to go deep into the field, which is Huawei's most important advantage in large models.
In practical applications, Huawei has realized the application of large models in many industries. For example, in the government affairs market, Huawei's Pangu government affairs model can accurately understand people's consultation intentions by fine-tuning more than 200,000 government affairs data.
In the financial field, the Pangu Finance model can automatically generate processes and operation guidance for counter staff by pre-training various operations, policies and case documents of banks, which greatly improves work efficiency.
In addition, Pangu Model has also launched special industry models in coal mines, railways, drug research and development and other industries to further help the industry improve efficiency. Huawei's goal is to let every industry and everyone have their own "expert assistant".
Zhang Pingan, CEO of HUAWEI CLOUD, said that they have always adhered to the strategy of AI for Industries and continued to move forward on the road of deepening the industry. He firmly believes that big models will reshape thousands of industries, and every developer will be a hero who will change the world.
On top of the industry's large models, Huawei has also developed more segmented and more specific scenario models, which are specially designed to solve specific problems and are "out of the box". At present, the Pangu model has been applied in more than 100 practical scenarios, lowering the threshold of artificial intelligence development and saving more than 80% of R&D costs on average.
Undoubtedly, the development of large models will bring a scientific and technological revolution that will completely change the entire industrial society. As an enterprise, in addition to research and engineering, Huawei needs to explore new large-model business models to ensure the commercial success of large-model models.
At present, Huawei has divided the Pangu large model into a three-layer model from L0 to L2, and on the basis of complete decoupling, it has split and combined it according to the needs of different customers, in order to further explore the boundaries of large model commercialization.
Huawei, as a leader in China's AI industry, has taken an important step in the study of large models. Although the advent of its Pangu 3.0 is later than other competitors, the research accumulation and deep talent reserve behind it have made Huawei more stable on the road of large models.
As early as 2020, Huawei foresaw two major development trends in the AI industry: first, the transformation of small models to large models, and second, the integration of AI and traditional scientific computing. Therefore, Huawei proposed six sub-topics, including the model height plan and the pre-vision plan of everything that is highly related to the large model. Although the driving force of ChatGPT cannot be ignored, Huawei has already begun research on large models before the release of GPT-3.
Huawei's Pangu model team has a deep talent pool, with more than half of its team members being PhD holders. This young, technically skilled, and innovative team is a solid backing for Huawei's large-scale model research.
Huawei's future development direction is not to focus on the size of the parameters of the model, but on its penetration rate in various industries. In addition to the railway, coal mine, finance, government and other industries that have entered, there are more industries waiting for Huawei's large-model services.
Huawei's strength lies in its full-stack R&D capabilities. In the research of large models, Huawei has carried out independent innovation from computing power to operators, frameworks, development platforms, etc., and does not rely on open source technology. All this is due to Huawei's deep accumulation of root technologies such as AI foundation, computing power, and chips.
Huawei's main focus on AI development is: on the one hand, to build a strong computing power base and do a good job in industrial infrastructure; On the other hand, from the general large model to the industry large model, serve all walks of life.
When Huawei's large model Pangu 3.0 was released, a new LOGO was also announced, symbolizing the significance of Pangu opening the world and the birth of all things. This also indicates that Huawei's large model will shoulder an important mission in the future development.