Core ideas
The 2023 World Artificial Intelligence Conference was held in Shanghai on July 6-8, 2023, bringing together the world's top AI technologies. The theme of the 2023 World Artificial Intelligence Conference is "Intelligent Connected World", the main exhibition covers four major sections of core technology, intelligent terminal, application empowerment and cutting-edge technology, involving large models, chips, robots, intelligent driving and other fields, with more than 400 exhibitors, more than 30 new products for the first exhibition, and the number of exhibitors and exhibition area have reached the highest ever.
Star companies such as Huawei and SenseTime gathered and competed to bring blockbuster releases. 1) Huawei released the Ascend AI large-model training and promotion integrated solution, and joined forces with four partners to reshape the large-scale model development process and provide "out-of-the-box" solutions to improve customer efficiency. The Ascend AI industry has developed rapidly, with more than 30 hardware partners, more than 1,200 ISVs, and more than 2,500 industry AI solutions jointly launched, serving carriers, Internet, finance and other industries on a large scale, and the ecosystem continues to prosper. 2) The "SenseTime SenseNova" large model system has been upgraded in multiple dimensions, and SenseChat 2.0 has broken through the limitation of input length of large-language models and launched model versions with different parameter magnitudes, which can perfectly adapt to the application needs of different terminals and scenarios such as mobile terminal and cloud, and reduce deployment costs. The iteration of new model technology has brought a variety of applications from all walks of life, and has been applied in finance, medical care, online live broadcast, and offline scenarios.
The National Large Model Standardization Task Force was established, and nine "treasures of town halls" appeared in WAIC. 1) On July 7, the National Artificial Intelligence Standardization Group guided by the National Standards Commission announced the first large model standardization task force in mainland China at WAIC, in which Shanghai Artificial Intelligence Innovation Center served as the team leader, and Baidu, Ali, iFLYTEK, 360, Huawei, and Mobile jointly served as the co-leaders. 2) WAIC selected nine innovative exhibits with outstanding contributions and leading technologies in the industry from the exhibitors, from Ant Technology's Antjian AI Model Security Platform 2.0 to Baidu Wenxin Yigewen Sheng Tu AI, from XtalPi's AI drug discovery automation solution to Xijing Technology's intelligent battery swapping driverless commercial vehicle Q-Truck, AI technology is expected to empower all walks of life in the future.
More than 20 robots appeared in WAIC, and Tesla robots debuted. According to the Xinmin Evening News, at the World Artificial Intelligence Conference, more than 20 robots were unveiled, reflecting the vitality of the AI robot industry. 1) At the World Artificial Intelligence Conference, Tesla participated for the first time and brought the Tesla Bot humanoid robot. Tesla said that at present, the humanoid robot can complete walking, going up and down stairs, squatting, taking items and other actions, and has the ability to protect itself and the people around it, and in the future, it can also cook, mow the lawn, help take care of the elderly, or replace humans in the factory to do boring and dangerous work. 2) Robot exhibits involve thousands of industries and can empower various industries to become intelligent. For example, Meituan's fourth-generation drones can be used for delivery services; The intelligent preliminary screening robot displayed in the Jingdong booth can check ECG, heart rate, blood oxygen, and give physical condition index and nutritional supplement suggestions through comprehensive analysis of immunity, respiratory and other systems; The system can detect defective products in time, and can also analyze and compare different batches of product defect pictures, with an output accuracy rate of up to 99.9%.
Investment advice: landing is king, it is recommended to pay attention to 1) Computing power: NVIDIA, optical communication industry chain, Inspur Information, Shuguang of Zhongke Shuguang, Industrial Fulian, Ascend Industry Chain, Haiguang Information, Cambrian, Yihualu, Digital China, Jingjiawei, etc.; 2) Continuous landing applications: Microsoft, Kingsoft Office, Dahua Co., Ltd., Rambler, iFLYTEK, Wondershare Technology, Tolls, Zitian Technology, Kunlun Wanwei, Runda Medical, etc.
Risk warning: the risk that the iteration of AI technology is not as expected; the risk of a weaker-than-expected economic downturn; The risk of increased competition in the industry.
The body of the report
01
Star companies such as Huawei and SenseTime gathered and competed to bring blockbuster releases
1.1 Huawei: Ascend AI Big Model Training and Promotion Integrated Solution Released, Continuing to Flourish and Open AI Ecosystem
Huawei released the Ascend AI Big Model Training and Push Integrated Solution. On July 6, 2023, the Ascend Artificial Intelligence Industry Summit Forum was held in Shanghai. At the forum, Huawei and four partners jointly released the Ascend AI large-model training and promotion integrated solution, accelerating the rapid application of large models in various industries. With the rapid development of the field of deep learning, more and more application scenarios need to use large models for training and inference. However, in the whole process of large model development, enterprises often face complex problems such as switching hardware clusters, adapting to underlying hardware, and fine-tuning and optimization. In addition, a series of challenges such as long R&D cycle of large models and high deployment thresholds make it difficult for "build-up" development methods to meet the needs.
Reinvent the big model development process to provide "out-of-the-box" solutions to improve customer efficiency. To effectively solve these problems, Huawei and four partners jointly released the Ascend AI large-model training and promotion integrated solution, which integrates training and promotion, and provides an "out-of-the-box" solution that seamlessly switches the training and inference processes of large models in the same cluster, providing customers with an efficient large-model training and inference deployment process. At the same time, the solution has a built-in large model inference engine, tuning tools, and operator acceleration library, which supports customers to use it out of the box and allows customers to directly use large-scale pre-trained models without secondary hardware adaptation, thereby reducing the technical threshold. Combining the basic software and hardware capabilities of Ascend AI and the advantages of partner models and platforms, by embedding the algorithms and model capabilities of the four partners in large model training, fine-tuning, and optimization, we can jointly design, jointly develop, go public, and continuously iterate, so as to provide customers with more comprehensive solutions and better meet the needs of customers in all scenarios and industries.
| Figure 1: Huawei Ascend released an integrated solution for large-model training |
 |
| Source: Huawei China, Guosheng Securities Research Institute |
Ascend AI continues to open up and prosper the industrial ecosystem with partners. Ascend AI industry has developed rapidly, with more than 30 hardware partners, more than 1,200 ISVs, and jointly launched more than 2,500 industry AI solutions, serving carriers, Internet, finance and other industries on a large scale. Huawei continues to open AI modules and accelerator cards, provide rich interface development documents, reference designs, and software and hardware technical support, enabling partners to launch a variety of hardware products and stimulate market innovation. At this conference, 23 companies launched the Ascend AI series of new products, covering inference servers, training servers, edge computing intelligent gateways, industrial computers, developer kits, robots, drones and other rich categories. In addition, Huawei has launched the newly launched "Ascend Partner Network" (APN) partner program, which encourages partners to build their own branded products or solutions based on Ascend AI through sales support provided by the main distributor, technical support between Huawei and the main seller, and a variety of business incentives.
| Figure 2: New product launches by Ascend partners |
 |
| Source: Huawei China, Guosheng Securities Research Institute |
1.2 SenseTime: The "SenseNova" large model system has been upgraded in multiple dimensions, empowering industries such as finance, healthcare, and online live broadcasting
The "SenseNova" large model system has been upgraded in multiple dimensions. 1) As a natural language processing model with hundreds of billions of parameters, SenseChat 2.0 breaks through the limitation of input length of large language models and launches model versions with different parameter magnitudes, which can perfectly adapt to the application requirements of different terminals and scenarios such as mobile and cloud, and reduce deployment costs. 2) The model parameters of SenseMirage 3.0, SenseTime's self-developed generative large model, SenseTime Seconds, have increased from 1 billion to 7 billion since its first release in April this year, enabling professional photography-level image detail portrayal. 3) Compared with version 1.0, SenseAvatar 2.0 digital human generation platform improves the fluency of voice and lip shape by more than 30%, realizes 4K high-definition video effects, and brings AIGC image generation and digital human singing functions. 4) SenseTime Qiongyu SenseSpace 2.0 improves the space reconstruction efficiency by 20%, the rendering performance by 50%, and the mapping time of each 100 square kilometers scene can be completed in only 38 hours (1200 TFLOPS/sec computing power support). SenseThings 2.0 achieves millimeter-level fineness in texture and material restoration of small objects, and breaks through the problem of collecting highly reflective and specular objects.
| Chart 3: "SenseNova" big model system has been upgraded in multiple dimensions |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
The iteration of new model technology has brought a hundred flowers to all walks of life. Relying on the rapid iteration of the "SenseNova" big model system in the underlying technology field, SenseTime is actively empowering industrial upgrading through the multi-modal capability combination of large models and bringing many industry-leading application innovations.
1) In the financial field, SenseTime cooperates with banks, insurance, securities firms and other customers, uses digital humans for intelligent customer service, smart marketing, etc., and provides new functions such as investment research analysis and research report writing by accessing the ability of large language models to reduce costs and increase efficiency. In addition, after mounting the financial knowledge base, it can also output content Q&A based on the customer's product description 100%, and realize timely update of information.
| Figure 4: Application of the SenseTime model in the financial sector |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
2) In the medical scenario, SenseTime has built a Chinese medical language model "Big Doctor" based on massive medical knowledge and clinical data, providing multi-scenario and multi-round conversation capabilities such as guidance, health consultation, and decision-making assistance, and will soon support multi-modal comprehensive analysis of medical images, text, structured data, etc., and continuously improve medical language understanding and reasoning capabilities, and continue to empower hospital diagnosis efficacy rate and patient service improvement.
| Figure 5: Application of the SenseTime model in the medical field |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
3) Combining the comprehensive capabilities of Discussion 2.0 and Miaohua 3.0, SenseTime also brings a variety of intelligent interaction solutions to mobile terminal customers, including Q&A interaction for information acquisition, knowledge interaction for life scenarios, content interaction for language and image generation, etc., relying on the lightweight version of SenseTime's model, which can be easily deployed and operated on mobile terminals.
| Figure 6: SenseTime's mobile intelligent interaction solution |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
4) For offline scenarios, SenseTime brings intelligent solutions such as long-tail fault identification and complex defect judgment to power grid inspection through large model capabilities. Based on the spatial reconstruction of Qiongyu 2.0, SenseTime created a digital twin of the real space for the regional development of Mashan Town in Jinan, China Vision Park in Hefei, and Shanghai Ruijin Hospital to improve the efficiency of operation and management. In the jewelry industry, SenseTime relies on Gewu 2.0 to reproduce jewelry for jewelry brands, meticulously display the characteristics of product craftsmanship, and enhance customer shopping experience.
| Figure 7: Qiongyu 2.0 commercial implementation |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
5) At the same time, the digital humans generated by SenseTime 2.0 are being widely used in online short video and live broadcast platforms, and SenseTime has also reached channel strategic cooperation with a number of leading enterprises to jointly build a "cloud + AIGC+ short video live broadcast" ecology, bringing more efficient, low-cost, convenient and easy-to-use AI video and marketing tools to the industry.
| Figure 8: Application of SenseTime's virtual human technology in live streaming and short video |
| |
| Source: SenseTime, Guosheng Securities Research Institute |
WAIC SenseTime's live booth showcased large models and AIGC capabilities, including AI painting, intelligent live streaming and other functions. The paintings displayed at SenseTime's booth are all from SenseTime's "Sensemirage", a creation platform that includes SenseTime's self-developed AIGC large model and convenient LoRa training capabilities, and provides third-party community open source models to accelerate reasoning, aiming to provide creators with convenient and perfect content production and creation tools. The paintings displayed at SenseTime's booth are all from SenseTime's "Sensemirage", a creation platform that includes SenseTime's self-developed AIGC large model and convenient LoRa training capabilities, and provides third-party community open source models to accelerate reasoning, aiming to provide creators with convenient and perfect content production and creation tools.
| Figure 9: SenseTime AI painting display |
| |
| Source: WAIC official WeChat public account, Guosheng Securities Research Institute |
| Figure 10: SenseTime Smart Stadium |
| |
| Source: WAIC official WeChat public account, Guosheng Securities Research Institute |
01
The National Large Model Standardization Task Force was established, and nine "treasures of town halls" appeared in WAIC
2.1 The National Large Model Standardization Task Force was established, with Baidu, Alibaba, iFLYTEK, 360, Huawei, and Mobile jointly taking the lead
On July 7, the 2023 World Artificial Intelligence Conference "Talking Standards, Driving Industry - Generative Artificial Intelligence Standardization Sub-forum" was held in Shanghai. At the meeting, the National Artificial Intelligence Standardization Group guided by the National Standards Committee announced the leader of the mainland's first large model standardization task force, which was jointly served by Shanghai Artificial Intelligence Laboratory and iFLYTEK, Huawei, Ali and other enterprises, and issued certificates on site and officially launched the formulation of national standards for large model testing.
Among them, Shanghai Artificial Intelligence Innovation Center serves as the team leader, and Baidu, Alibaba, iFLYTEK, 360, Huawei, and Mobile jointly serve as the co-leaders.
| Figure 11: The leader of the large model task force of the National AI Standardization Overall Group was announced |
| |
| Source: iFLYTEK official WeChat public account, Guosheng Securities Research Institute |
2.2 Nine "treasures of the town hall" appeared at WAIC, and AI technology empowers thousands of industries
From the industry perspective of professionals, this year's WAIC selected innovative exhibits with outstanding contributions and leading technologies in the industry from exhibitors, encouraged innovative achievements in the field of artificial intelligence worldwide, guided and encouraged artificial intelligence enterprises to pursue excellent quality, and continuously improved the core competitiveness of technology, products and enterprises.
| Exhibit 12: 2023 WAIC Nine "Treasures of the Town Hall" | ||
| Product name | Exhibitors | Introduction to the feature |
| Antjian AI Security Detection Platform 2.0 | Ant Technology | Identify mining large model vulnerabilities |
| Amazon Bedrock | Amazon Pass Technology Services | Generative AI basic model application |
| Wen Xin Yi style | Baidu | Wen Sheng diagram |
| Graphcore C600 IPU processor PCIe card | Proposed technology | Inference/training cards |
| Ascend AI "Big Model Gigafactory" | Huawei | Large model development |
| AI drug discovery automation solutions | XtalPi | AI drug research and development |
| Cloud Fire Intelligent Computing Cluster | Flintstones Technology | AI-accelerated clustering |
| Intelligent battery swapping driverless commercial vehicle Q-Truck | Nishii Technology | Driverless commercial vehicles |
| Multimedia Lab XMusic | Tencent | AIGC Music Composition |
| Source: WAIC website, Guosheng Securities Research Institute |
·Antjian AI Security Detection Platform 2.0: AntJian 2.0 can detect generative AI models with generative AI capabilities, identify hundreds of risks in three categories: data security, content security, and scientific and technological ethics, covering a variety of data and task types such as tables, texts, and images, and is the industry's first all-data AI security detection platform that realizes industrial-level applications. Developers can connect AI models to Ant Discovery through API interfaces to identify and mine model vulnerabilities with one click.
| Figure 13: Schematic diagram of AntJian AI Security Detection Platform 2.0 |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
· Amazon Bedrock: A fully managed, generative AI service from Amazon Web Services. Help customers build and scale generative AI applications through APIs that provide foundational models from leading AI startups and Amazon's own development. Customers can choose from a variety of base models and get started right with Amazon Bedrock's serverless experience, or fine-tune with their own data while ensuring data security and privacy.
| Figure 14: Base models supported by Amazon Bedrock |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
·Wen Xin Yi Ge: At the opening ceremony, Baidu AI digital human Xi Jiaja used Wen Xin Yi Ge to automatically create a variety of styles of art works with Shanghai style, such as the unique charm of "Ink Alley", Van Gogh-style "Bund Cruise", and Monet-style "Impression Raw Frying".
| Picture 15: Wen Xin Yige painting the opening ceremony |
| |
| Source: Phoenix Technology, Guosheng Securities Research Institute |
| Figure 16: Schematic diagram of the Wen Xin Yi Ge Wensheng diagram system |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
· Graphcore® C600 IPU processor PCIe card: The PCle card that is not yet launched focuses on inference and training, which can support various mainstream AI applications, and has unique advantages in search and recommendation services. The C600 delivers low latency and low power consumption, with a thermal design that consumes only 185 watts when running typical workloads, significantly reducing data center operating expenses. In addition, the Bow IPU is the world's first 3D Wafer-on-Wafer processor to be launched. Each chip provides AI computing power of up to 1.4 petaFLOPS for AI applications and significant power efficiency gains.
| Figure 17: PCIe card with C600 IPU processor |
| |
| Source: WAIC public account, Guosheng Securities Research Institute |
Ascend AI "Big Model Super Factory": Huawei has incubated the industry's first 200 billion parameters Chinese NLP large model Pengcheng based on Ascend AI. Pangu, the industry's first multi-modal large-scale model Zidong. More than 20 domestic models, including Taichu and HUAWEI CLOUD Pangu series. At the same time, Ascend AI has also adapted and supported dozens of mainstream open source models such as ChatGLM, LLaMA, GPT-3, and BLOOM.
| Figure 18: Ascend AI "Big Model Gigafactory" |
| |
| Source: WAIC public account, Guosheng Securities Research Institute |
· AI Drug R&D Automation Solution: The AI drug R&D automation solution launched by XtalPi is highly automated and intelligent, which not only breaks through the key rate-limiting bottleneck of chemical synthesis in the DMTA cycle of drug discovery, but also further solves the problem of missing standardized process data in AI drug R&D experiments, and can record standardized and traceable process data and positive and negative result data throughout the process for AI model learning and iteration to achieve a positive cycle.
| Figure 19: XtalPi's AI Automated Solution for Drug Discovery |
| |
| Source: XtalPi website, Guosheng Securities Research Institute |
Cloud intelligent computing cluster: This cluster product can support over-kilocard-scale clusters and break through E-level computing power; It is the industry's first liquid cooling solution based on OAM open computing standards, with PUE as low as 1.1. It has landed a large-scale training cluster in the national key laboratory, providing more than 100P of advanced AI computing power, supporting multiple AIGC applications including media fusion, text generation PPT applications, cross-modal image generation, as well as the development and cutting-edge exploration of a variety of AI4S scientific computing applications such as computational genes, computational pharmaceuticals, and computational materials.
| Figure 20: Cloud tunnel liquid cold open acceleration computing server |
| |
| Source: The Paper, Guosheng Securities Research Institute |
Intelligent battery swap unmanned commercial vehicle Q-Truck: The world's first intelligent battery swap driverless commercial vehicle Q-Truck, supporting the smart energy service PowerOnair, can complete unmanned battery replenishment in 5 minutes, suitable for multiple logistics scenarios such as sea, land, air, rail, and factories. As a driverless brand designed and made in China in China, its products have been commercially landed in Thailand, the United Arab Emirates, the United Kingdom, Malaysia, Mexico and other countries.
| Figure 21: Cloud tunnel liquid cold open accelerated computing server |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
XMusic: XMusic, a generative universal composition framework based on AIGC technology, supports multi-modal content such as videos, pictures, text, tags, and humming as input prompt words to generate high-quality music with controllable mood, genre, and rhythm. XMusic's commercial-grade music generation capabilities have broad application prospects in many fields such as video scores, interactive entertainment, auxiliary creation, and music education.
| Figure 22: Schematic diagram of Tencent XMusic |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
03
More than 20 robots landed on WAIC, and Tesla robots were unveiled
More than 20 robots appeared in WAIC, and the AI robot industry is developing vigorously. With the development of AI large model technology and embodied intelligence, software technology breakthroughs have brought new possibilities to the application of robots, catalyzing rapid changes in the robot industry. According to the Xinmin Evening News, at the World Artificial Intelligence Conference, more than 20 robots were unveiled, reflecting the vitality of the AI robot industry.
Tesla exhibited for the first time, and humanoid robots appeared. At the World Artificial Intelligence Conference, Tesla participated in the exhibition for the first time, not only brought two star models of Model Y/Model X, showing the audience the latest progress in intelligence such as Tesla's full self-driving capability (FSD) and Dojo supercomputer, but also displayed the Tesla Bot humanoid robot. Musk said that in the future, we will have a large number of robots, and in the long run, the ratio of robots to humans will exceed 1 to 1. "In a future with a large number of robots, global productivity will increase to incredible levels." Tesla said that at present, this humanoid robot can complete walking, going up and down stairs, squatting, taking items and other actions, and has the ability to protect itself and the people around it, and in the future, it can also cook, mow the lawn, help take care of the elderly, or replace humans in the factory to do boring and dangerous work. Because he is equipped with the same fully autonomous driving ability computer and visual neural network system as Tesla cars, with the blessing of Dojo supercomputer, he can also use motion capture to "learn" humans.
| Figure 23: Tesla bot at the World Artificial Intelligence Conference |
| |
| Source: Xinmin Evening News, Guosheng Securities Research Institute |
Meituan released the fourth generation of drones, which can be used for delivery services. In addition to Tesla BOT, other robots also participated in this World AI Conference. On July 5, Meituan UAV released the fourth generation of new models on WAIC, which took more than two years to develop and is a new multi-rotor model focusing on urban low-altitude logistics and distribution scenarios. Compared with previous generations of models, the new model has upgraded environmental adaptability, can fly stably in "minus 20 degrees to 50 degrees" in moderate rain, moderate snow, level 6 wind, night and other environments, and can adapt to the natural environment requirements of more than 97% of domestic cities. Mao Yiyian, head of Meituan's drone business, said that the fourth-generation drone will allow users to enjoy more efficient and convenient delivery services. Wang Puzhong, senior vice president of Meituan and president of Daojia Business Group, said that in the past two years, more and more users tend to "buy everything online", and the requirements for delivery time are becoming more and more stringent, and Meituan can provide consumers with a better service experience by building a three-dimensional urban distribution network integrating space and land. Up to now, Meituan drones have been operated in many business districts, scenic spots, residential areas and other cities such as Shenzhen and Shanghai, with a total of nearly 170,000 orders and nearly 20,000 categories that can be delivered.
| Exhibit 24: Meituan drone officially released |
| |
| Source: WAIC website, Guosheng Securities Research Institute |
Robot exhibits involve thousands of industries, empowering various industries to become intelligent. For example, the intelligent preliminary screening robot displayed in the Jingdong booth can check people's electrocardiogram, heart rate, blood oxygen, and give overall physical condition index and nutritional supplement suggestions through comprehensive analysis of immunity, respiratory and other systems, and comprehensively analyze a person's physical state, traditional Chinese medicine physique and health risks. The system can detect defective products in time, and can statistically analyze and compare different batches of product defect pictures, with an output accuracy rate of up to 99.9%, which is equivalent to equipping the machine with eyes and brains. The Sound Intelligence booth also showcased the robot brush writing function.
| Figure 25: JD intelligent initial screening robot |
| |
| Source: Guangming Network, Guosheng Securities Research Institute |
| Figure 26: Micro Intelligent Manufacturing Intelligent AI Appearance Defect Inspection Equipment |
| |
| Source: Guangming Network, Guosheng Securities Research Institute |
| Exhibit 27: Live demonstration of the robot writing brush word function |
| |
| Source: Guangming Network, Guosheng Securities Research Institute |
04
Investment advice
Landing is king, it is recommended to pay attention to:
1) Computing power: NVIDIA, optical communication industry chain, Inspur Information, Shuguang, Industrial Fulian, Ascend Industry Chain, Haiguang Information, Cambrian, Yihualu, Digital China, Jingjiawei, etc.;
2) Continuous landing applications: Microsoft, Kingsoft Office, Dahua Co., Ltd., Rambler, iFLYTEK, Wondershare Technology, Tolls, Zitian Technology, Kunlun Wanwei, Runda Medical, etc.
05
Risk Warning
Risk of AI technology iteration not being as expected: If AI technology iteration is not as expected, it will have a certain adverse impact on companies related to the industry chain.
Economic downturn exceeds expectations: If the macroeconomic boom declines, the amount of fixed asset investment slows down, which will affect the willingness of enterprises to reinvest, thereby affecting consumers' willingness to consume and industrial chain production, which will adversely affect the entire industry.
Risk of intensified industry competition: If relevant enterprises accelerate technology iteration and application layout, the overall industry competition will intensify, which will threaten the growth of enterprises in the current industry.