Shang Tang’s ＂Mid-term Examination＂： The big model was accelerated and upgraded, and the revenue related to generative AI increased by 670.4% year-on-year.

On August 28th, Shang Tang Group released the semi-annual report for 2023, achieving an operating income of 1.433 billion yuan in the first half of the year, up 1.3% year-on-year; The net loss narrowed by 2% compared with the same period of last year, and the adjusted net loss narrowed by 6.7%.

The financial report shows that the Group’s revenue related to generative AI has achieved a strong growth of three digits thanks to the release of the "day-by-day" big model system during the reporting period, and the generative AI product series such as "Consultation SenseChat", "Second Picture SenseMirage" and "Shadow SenseAvatar". According to the latest news, Shang Tang’s "growing day by day" big model has passed the Interim Measures for the Management of Generative Artificial Intelligence Services for the record, and Shang Tang’s "Consultation SenseChat" has been officially opened to the whole society.

Shang Tang said that while increasing investment in R&D of large-scale models, the Group still adhered to the operating goal of continuously reducing losses. Combined with the management requirements of cost reduction implemented since 2022, its AI commercialization efficiency improved, and the current operating cost decreased by 6.5% year-on-year.

Shang Tang’s capital reserve is still sufficient. By the end of the reporting period, Shang Tang held a total of 14.82 billion yuan in bank deposits, cash in hand, structured deposits, bonds and other fixed-income products.

The iteration of large-scale model technology is accelerated to consolidate the core competitiveness of AI.

Since the end of 2022, the generative AI model has been on the rise, and the world’s major technology giants are rushing into the big model track, intending to get a slice of this promising market.

IDC predicts that in 2026, the AI big model market in China will reach 21.1 billion US dollars, and nearly 50% of the terminal devices in China market will have AI engine technology.

In April, 2023, Shang Tang’s "Consultation SenseChat1.0" was released, which is one of the earliest chat robot products based on the large language model with hundreds of billions of parameters in China.Only two months later, Shang Tang, together with a number of top domestic scientific research institutions, released the first basic model InternLM with comprehensive performance exceeding GPT-3.5-turbo in China. On this basis, "Consultation" quickly iterated to version 2.0.

In August, the new model InternLM-123B completed the training, and ranked second in the world in terms of test scores on a total of 300,000 questions in 51 well-known evaluation sets around the world.For example, in five evaluation sets on reading comprehension, the scores of InternLM-123B are all higher than those of GPT-3.5-turbo and the LLaMA2-70B model newly released by Meta, and the scores of individual test sets differ by nearly 20 points.

It is reported that InternLM-123B focuses on upgrading the code interpreter and plug-in calling ability. python interpreter, API calling and search can be used to solve complex tasks and flexibly build AI agent applications. Due to the remarkable improvement of the new model, it is expected that "Consultation SenseChat" will be upgraded to version 3.0 in September.

In addition to the big language model, Shang Tang is also committed to training the world’s top three models of Wensheng Map. As early as January, Shang Tang’s second painting 1.0 started the internal test, and a second-dimensional picture can be generated by inputting prompts; In April, the second painting was upgraded to 2.0 and officially released; In July, version 3.0 of Seconds Painting adopted a brand-new generation model architecture, which can realize professional photography-level picture detail description, and its core algorithm surpassed Google’s Imagen and OpenAI’s DALL·E2 on COCObenchmark.

With the iteration of Shang Tang’s second painting from 1.0 to 3.0, its model parameters increased from the initial 800 million to 7 billion. Version 4.0 is expected to be launched in the third quarter of 2023, and new aesthetic system support will be introduced.

The iteration speed of Shang Tang AI model is extraordinary, which is due to the support of the computing infrastructure behind it. Since 2023, in order to maintain its core competitiveness, Shang Tang has continuously upgraded SenseCore, an intelligent computing infrastructure. The number of online GPUs has increased from 27,000 at the end of March 2023 to about 30,000 recently, and the computing power scale has increased by 20% to 6ExaFLOPS.

Since 2023, more than 1,000 large models with parameters ranging from billions to hundreds of billions have been trained on Shang Tang AI equipment, which has achieved technological iteration leading the industry and supported dozens of generative AI applications.

It should be said that the maturity of any AI technology is by no means achieved overnight. Behind Shang Tang’s big model is the "miracle" of deep learning algorithm, computing power improvement and data accumulation, and deep cultivation for several years.

According to the AI Big Model Market Research Report (2023) published by Jost Sullivan, Shang Tang ranks first in China in terms of comprehensive competitiveness in product technology, strategic vision and ecological opening-up construction.

Release the commercial potential of big model and accelerate the optimization of business structure

When people talk about the big AI model, most of the time they are feeling the impact and subversion it brings to all walks of life, but few people think that the commercial landing of the big model will come so quickly.

The industry has been growing crazily for half a year, and it has completed the transformation from "Hundred Models Battle" when running into the market to thinking about putting technology into practice. Industry analysts believe that with the rise of basic model and tool layer, the cost and difficulty of building applications will be greatly reduced. For application developers, all downstream applications are worth reconstructing.

The observation of AI trend in the era of big model points out that vertical domain application is the main battlefield of big model. With the rapid development of generative AI technology, it has spawned brand-new commercial value in many fields. Although these models are in the early stage of expansion, they also show excellent capabilities. Like other disruptive technologies, this change will develop slowly at first, and then accelerate rapidly.

A ready-made case is that in Shang Tang’s interim report, this change is gradually emerging.

In the first half of 2023, Shang Tang’s revenue related to generative AI recorded a year-on-year growth of 670.4%, and its contribution to the Group’s business increased rapidly from 10.4% in 2022 to 20.3%.Large-scale model has become an important driving force to upgrade the business of Shang Tang.

It is understood that Shang Tang released a series of generative AI products in April, which is one of the earliest enterprises in China to release related products, including Consultation, Secondhand Painting, Ruying, Qiongyu and Gewu, respectively corresponding to the five mainstream generative AI application landing scenarios: natural language interaction, AI Wen Sheng Tu, digital people, 3D big scene reconstruction and 3D small object generation.

The landing and maturity of AI big model not only accelerates the transformation and optimization of Shang Tang’s own business structure, but also shows the potential of more business application scenarios.

Reflected in the business level, at present, Shang Tang’s main businesses are divided into four major sectors, namely, smart business, smart life, smart cars and smart cities. During the reporting period, Shang Tang actively promoted the commercial application of AI technology and helped customers to quickly iterate new products based on large models, with remarkable results.

Among them,Driven by the big AI model, the revenue of the smart business sector increased by 50.2% year-on-year to 854 million yuan, the revenue proportion increased by 19.4 percentage points to 59.6%, the number of service customers reached 481, and the revenue of single customers increased by 59.9%.For example, in cooperation with domestic head insurance companies, Shang Tang provided cross-validation of crop underwriting data based on AI remote sensing model, which helped to improve the risk assessment and claims service system. Compared with the traditional manual identification scheme, the efficiency was improved by 60 times.

The tentacles of the smart life business extend to the C-end. In cooperation with customers in different fields such as e-commerce, traditional media and community platforms, Shang Tang’s AI products such as consultation, second painting, shadow and grid objects all come in handy. From digital life to high-precision 3D object modeling ability, Shang Tang’s technical strength can be intuitively presented to C-end users. During the reporting period, the revenue of this business increased by 6.7% year-on-year to 312 million yuan, the proportion of revenue increased to 21.8%, and the number of customers increased by 47.5% year-on-year.

Besides,The revenues of smart cities and smart cars are 184 million yuan and 84 million yuan respectively.Among them, the mass production business of Jueying achieved a year-on-year growth of 573%, and the number of mass production deliveries reached 390,000 vehicles, driving the gross profit of bicycles to increase by 29%. In terms of intelligent experience, Shang Tang built an intelligent assistant in the cabin based on "Consultation" and launched a number of interactive products in Chinese in the cabin. In the first half of the year, the number of smart car customers of the company increased by 70% year-on-year to 34. With the continuous transformation of its cutting-edge technology, the business income is expected to show a rapid growth trend.

Guolian Securities Research Report believes that the AI big model opens the second growth curve in Shang Tang, Shang Tang is in the leading position in R&D investment and business scale, and it is also the first manufacturer in China to lay out the AI big model. The company has accumulated rich technology and application experience in the AI field, which is expected to take the lead in the AI 2.0 era and promote the further growth of the company’s performance.

It is worth mentioning that,Previously, Shang Tang United many countries.bestThe model InternLM-7B, with a parameter of 7 billion, released by scientific research institutions, has opened some of its training data, training codes and basic model weights to academia and industry for free, and supports commercialization.

On August 31st, Shang Tang expanded its opening authority again, and announced that the large-scale model application "Consultation SenseChat" will be fully opened to users from now on, which is one of the first domestic large-scale models successfully filed and approved to go online.