HUAWEI CLOUD Pangu Model 3.0 is here! Why doesn't it do the Chinese version of ChatGPT? | Voices of AI

Since the beginning of this year, emerging technologies represented by pre-trained large models are setting off a new round of artificial intelligence boom, and the real usefulness of large models is still hazy under the symbiosis of hundreds of "models", and under the contention of hundreds of schools.

After more than half a year of "frenzy", even the once explosive ChatGPT traffic will not continue. When the popularity of the C-end gradually faded, the noise of the industry began to increase: with functions such as chat and painting, the big model can reconstruct human society? What's next for AI?

"The Pangu model does not write poetry, only does things." On the afternoon of July 7, Zhang Pingan, Executive Director of Huawei and CEO of HUAWEI CLOUD, once again clarified Huawei's positioning in the field of large models at the Huawei Developer Conference 2023 and does not make "ChatGPT" products. He said that Pangu Model 3.0 is a large model system that provides services completely for the industry and is designed based on the needs of the industry.

In HUAWEI CLOUD's view, artificial intelligence has become the focus of strategic competition for many technology companies, and the industry model plays an important role in the integration with the real economy, which will bring greater industry opportunities.

Don't do "ChatGPT"

Since the launch of ChatGPT in December 2022, AI big models have accelerated the development of a new generation of artificial intelligence. When the technology of large models and generative AI continues to make breakthroughs, many people in the industry have begun to think about what kind of utility large model technology will have in commercial operation.

At the above-mentioned developer conference, Huawei believes that large models are leading a new round of human-machine revolution and bringing users a subversive user experience. If we say that in the PC era, "mouse + graphical user interface" opened the first interactive revolution; "Touch screen + gesture" opened the era of mobile Internet, while "dialogue + 5G" opened the era of intelligent interconnection, marked by the large model technology represented by ChatGPT.

In December 2015, Silicon Valley entrepreneurs launched the engine of OpenAI, and the ChatGPT project began to be brewed around this time. The year before, Google had acquired DeepMind, and AlphaGo, developed by the DeepMind team, set off an AI storm around the world.

Subsequently, Google released the Transformer large model in 2017, which became a new watershed for the AI large model. But what Google didn't expect was that Transformer then became the soil for OpenAI to nourish ChatGPT.

After OpenAI's popularity, it also stimulated an arms race for big models among tech giants. In March this year, Baidu released Wen Xin Yi and began to integrate it into all of Baidu's businesses; In April, Alibaba released Tongyi Qianwen, and in June, it expanded the layout of the large model to the AI audio and video track. According to incomplete statistics, at present, only more than 80 large-model products have been released in China, corresponding to different industries and different application scenarios, and the development of "100-model war" is in full swing.

But the longer it goes in the field of large models, the more cautious Huawei's attitude towards the big model track becomes. "In Huawei's view, in the face of the current situation, we must be optimistic and remain calm." Ken Hu, Huawei's rotating chairman, said at the 6th World Artificial Intelligence Conference held on the 6th that the key to the development of AI is to be down-to-earth, promote AI to go deeper and deeper, and truly serve thousands of industries.

From the perspective of Huawei's layout in the field of large models, the project was established in 2020 and the "Pangu Large Model" was released in April 2021.

From the content released on the 7th, after the upgrade of Pangu Industry Big Model 3.0, Pangu will be the collective name of its "large model series", including the basic model including the language model and the visual model, as well as the financial, manufacturing, drug molecule industry model and scene development model service with industry attributes. The first financial reporter learned from Huawei's internal knowledge that in the early stage, HUAWEI CLOUD AI has more than 1,000 projects in various industries, which also paves the way for the landing of Pangu large model in the industry.

Huawei founder Ren Zhengfei once pointed out that there will be a surge in AI models in the future, not just Microsoft. The direct contribution of artificial intelligence software platform companies to human society may be less than 2%, and 98% of it is the promotion of industrial society and agricultural society.

In Ren Zhengfei's view, the application of the model is sometimes more promising than the model itself, Huawei will be the underlying computing power platform of AI, but the application platform is not Huawei's option, "In the 2% of the platform contribution, we can account for a little bit." What are the opportunities for ChatGPT for us? It will stretch up the calculation, stretch the pipeline flow, so that our products have market demand. ”

How does a large model go from concept to landing?

In terms of the current competitive landscape of the global market, there are more than 3,000 generative AI applications available, with thousands of technology companies around the world participating. Under the "fury" of large models, enterprises need to be more clear about the direction of application landing to break through in this tough battle.

Zhang Pingan said, "At present, most of the applications of large models are concentrated in the 2C field, and when facing industry applications, due to the difficulty of obtaining industry data and the difficulty of combining technology and industry know-how, the landing of large models in the industry is slow. ”

The breakthrough point chosen by Huawei is to amplify the computing power advantage from artificial intelligence chips, and the other is to cultivate large models in multiple scenarios.

Zhang Dixuan, president of Huawei's Ascend computing business, said in an interview on July 6 that Huawei has helped incubate more than 20 basic big models, such as iFLYTEK's Spark model, "about half of China's big models are supported by Ascend AI."

In August 2019, Huawei announced the commercial use of its self-developed AI training chip Ascend 910, 7nm process, saying that the computing power at the same power consumption is twice that of the NVIDIA V100 chip (the previous generation of NVIDIA A100). According to Huawei's previously disclosed information, each cluster needs 1,000 Ascend 910. Under the sanctions, Huawei has increased the size of the Ascend computing cluster from a maximum of 4,000 cards to 16,000 cards.

This means that Huawei has become an "alternative choice" to NVIDIA, providing large-model computing power for companies in other industries and driving the shipment of its own products, forming a positive business cycle. In the official introduction of Pangu Big Model 3.0, the model can already provide customers with a series of basic large model training with 10 billion parameters, 38 billion parameters, 710 parameters and 100 billion parameters.

In addition, Zhang Pingan mentioned Pangu's "5+N+X" three-layer architecture in his speech.

The L0 layer includes five basic big models: natural language, vision, multimodality, prediction, and scientific computing, providing a variety of skills in industry scenarios, and the L1 layer is N industry large models, and HUAWEI CLOUD can provide industry-wide models trained using industry open data, including government affairs, finance, manufacturing, mining, and meteorology.

The L2 layer provides customers with more detailed scenario models, focusing more on specific industry applications or specific business scenarios such as government hotlines, network assistants, lead drug screening, conveyor belt foreign body detection, typhoon path prediction, etc., and provides customers with "out-of-the-box" model services.

In other words, Huawei's large model is not only for zero-based industry customers, but also for enterprises that have applied large models.

Zhang Pingan said that the Pangu model has covered many industries such as finance, finance, manufacturing, pharmaceutical research and development, coal mining, and railways. "For example, in the field of drug development, it turns out that the development of a new drug takes an average of 10 years and costs $1 billion. The Pangu drug molecular model helped the team of Professor Liu Bing of the First Affiliated Hospital of Xi'an Jiaotong University discover the world's first new target and new class of antibiotics in 40 years, shorten the lead drug development cycle to 1 month, and reduce the R&D cost by 70%. ”

It is worth noting that in addition to HUAWEI CLOUD, technology companies such as Tencent, Alibaba, Byte, and 360 have also seen opportunities in the industry and are aiming at the layout of large industry models.

"Dialogue, writing poetry, and painting are by no means the be-all of the big model. We need to think deeply about the application direction of large models. Wu Hequan, an academician of the Chinese Academy of Engineering, believes that in order to effectively put the big model into urban development, financial technology, biomedicine, industrial manufacturing, scientific research and other fields, it is also necessary for professional enterprises and organizations to accelerate their landing in the real industry, bring real value to the industrial demand, and truly serve the society on a large scale.