Tesla’s second-generation humanoid robot Optimus (Source: Photographed by TMTpost Media app editor)
Under a high temperature of 40 degrees in Shanghai, a more lively world-class AI conference kicked off.
html From July 4th to 6th, the 2024 World Artificial Intelligence Conference and High-level Conference on Artificial Intelligence Global Governance (waic 2024) was held in Shanghai.This conference, with the theme of "promoting sharing through discussion and promoting good wisdom through good governance", gathered representatives from all over the world, top experts and scholars, industry elites, start-up teams, etc. to discuss hot topics in the era of in-depth development of AI.
is an annual AI industry event. The exhibition area of this conference exceeds 52,000 square meters, with more than 500 companies participating, and more than 1,500 exhibits. It focuses on the three major sections of core technology, intelligent terminals, and application empowerment, focusing on large models. , computing power, robots, autonomous driving and other key areas, hundreds of large models from minimax, Baichuan Intelligence, Zhipu AI and other companies were collectively unveiled, showcasing a batch of the latest results of "artificial intelligence +" innovative applications, and the first batch of highly acclaimed Eye-catching innovative products.
hundreds of forums gathered together with stars. The conference will bring together 9 Turing Award, Fields Medal, and Nobel Prize winners and 88 domestic and foreign academicians, covering ten key topics such as AI ethical governance, large models, data, computing power, and embodied intelligence, gathering more than 1,000 people A global leader.
On the first day of the conference (July 4), among the 19 forums, Yao Qizhi, winner of the Turing Award, academician of the Chinese Academy of Sciences, and director of the Interdisciplinary Information Institute of Tsinghua University, director of Shanghai AI Laboratory and founder of Xianyuan Technology , Huiyan Chair Professor of Tsinghua University Zhou Bowen, former executive vice president of Microsoft, foreign academician of the National Academy of Engineering Shen Xiangyang, founder, chairman and CEO of Baidu Robin Li, academician of the Chinese Academy of Engineering, Zhijiang Laboratory Wang Jian, director and founder of Alibaba Cloud, Xu Li, chairman and CEO of SenseTime Technology, Liu Qingfeng, chairman of iFlytek, and many other big names discussed in depth the opportunities and challenges of generative AI technology.
Shun Xiangyang’s conversation with the Turing Award winner: Control AI well but not destroy it
Currently, the new generation of AI technology has injected new momentum into economic and social development and is profoundly changing people’s production and lifestyle. AI new technologies continue to make breakthroughs, new business formats continue to emerge, and new applications accelerate the expansion, and it has become an important driving force for a new round of scientific and technological revolution and industrial transformation. Public data from
shows that in 2023, the number of AI companies in China has exceeded 4,000, and the core AI industry scale has reached 578.4 billion yuan, with a growth rate of 13.9%. The adoption rate of Chinese generative AI companies has reached 15%, and the market size is approximately 144,000 billion.
Behind this, generative AI technology governance has become crucial.
The conference forums of this year's conference are presented according to the "1+3+10+x" structure, including an opening ceremony and plenary session, three main forums on global governance, industrial development, and scientific frontiers, and several industry forums, covering AI ethical governance , large models, data, computing power, embodied intelligence, new industrialization, autonomous driving, investment and financing, education and talent and other key topics, fully embodying the value orientation of AI for good, international cooperation, co-governance and sharing. At the opening ceremony of
, the "Shanghai Declaration on Global Governance of Artificial Intelligence" (hereinafter referred to as the Declaration) was officially released. The declaration proposes to promote the development of AI, maintain AI security, build an AI governance system, strengthen social participation and improve public literacy, and improve the quality of life and social well-being. The declaration calls on governments, science and technology circles, industries and other stakeholders around the world to actively respond and act together to jointly promote AI to benefit all mankind.
At the same time, the opening ceremony also released the "China Wisdom Benefiting the World" case collection, which collects a series of typical practices in China's use of AI to enhance the common welfare of mankind; the United Nations Industrial Development Organization (unido) Global Industrial AI Alliance Center of Excellence was also Officially launched yesterday.
In the Global Governance Turing Roundtable session of the waic 2024 plenary meeting held on July 4, Shen Xiangyang talked with three Turing Award winners Roger Reddy, Manuel Blum, and Yao Qizhi , discussing the coexistence of challenges and opportunities in AI governance in this era.
talked about his views and concerns about AI. Roger Reddy said that he was surprised and happy. First, he saw that everyone was worried about the risks and security issues of AI governance, but what specific needs to be done What studies? What investments are needed? It has not been explicitly done yet. Second, as the spokesperson of the United Nations mentioned "capacity building", we can try our best to let AI help everyone do their job well.
Roger Reddy believes that new technologies will create new opportunities and problems, but we should not stop just thinking about the negative effects. In the future of , if AI can increase everyone’s work efficiency by 10 times, the world will GDP will grow from US$100 trillion to US$1,000 trillion.
Yao Qizhi believes that AI risks come from three aspects: First, the extension and expansion of network risks. Now, we feel that managing data security is already very difficult, and it will be 100 times more difficult with the emergence of AI. The second is the social risks that are not realized. For example, AI is very powerful and can be used in many ways, so it will subvert the current social structure. There is a possibility that it will happen in a short period of time. For example, someone mentioned that AI may bring about large-scale unemployment in the future; the third is the risk of survival or existence, which has been faced before, when the train or the steam engine was invented. , some people have such concerns.
Yao Qizhi said that as a computer scientist, he has seen the most profound problem. On the one hand, we control AI well. After all, it is designed by us; on the other hand, we do not want it to be destroyed by us. This trade-off is Very difficult. As Turing said, this is unpredictable, and it is impossible to predict what the machine will do after it has enough computing power. Therefore, all these risks require a lot of experts to solve, not only scientists, but also governments, lawyers, economists, almost every industry.
In Yao Qizhi's view, AI is equivalent to the amplification of network risks. The risks of AI are also related to computation. The current computing power can solve many problems that we did not know how to solve before, but this is also a very scary place.
Zhou Bowen: Investment in AI safety lags far behind investment in AI performance
html On the morning of July 4, Zhou Bowen, who made his first public appearance since serving as the director of the Shanghai AI Laboratory, delivered a speech with the theme of "Exploring the 45° Balance Law of Artificial Intelligence".Zhou Bowen said that generative AI represented by large models has developed rapidly, but with the continuous improvement of capabilities, the models themselves and their applications have also brought about a series of potential risk concerns, including data leakage, ethical issues, AI loss of control, etc. Challenges. At present, these risks brought by AI have begun to emerge, but more are potential risks. Preventing these risks requires joint efforts from all walks of life and more contributions from the scientific community.
Zhou Bowen said frankly that at present, from the perspective of algorithm research, talent intensity, business driving force, and even investment in computing power, our investment in AI security lags far behind the investment in AI performance. Right now, only 1% of the world's (resources) are devoted to alignment or security considerations.
"AI must ensure controllability, coordinated development and security. There is no doubt that we must avoid such crippled AI development. What we should pursue is: trustworthy AGI (general artificial intelligence), trustworthy AI, and trustworthy universal "ai." Zhou Bowen said that to truly achieve the balance between AI's security and performance, it must be perfect and long-term sustainable. Only by believing that a truly trustworthy AGI can be built can the perfect balance between AI's security and excellent performance be achieved.
"Ultimately, just as safe and controllable nuclear fusion technology brings clean and abundant energy to all mankind, we hope to develop and use this revolutionary technology safely and effectively by deeply understanding the intrinsic mechanism and causal process of AI. Just as controllable nuclear fusion is of common interest to all mankind, we firmly believe that the safety of AI is also a global public welfare." Zhou Bowen said at the end of his speech.
Robin Li: In the fierce competition, the commercial closed-source model is the best
"Today's shocking release, tomorrow's epic update, but I have to ask, where is the application? Who benefits from it?"
7 On the 4th, Baidu founder, chairman and CEO Robin Li said in a speech that many companies focus on basic models, running scores and rankings all day long. Who has surpassed gpt-4? Openai has released sora. , gpt4o, etc. This situation is not good.
In Robin Li's view, the most important thing at present is application. "Without application, just a basic model, whether open source or closed source, is worthless." He appealed, "Don't roll up the model, roll up the application."
According to Robin Li, the daily calls of the Wenxin model exceeded 200 million more than two months ago, and recently exceeded 500 million. "The changes in the call volume reflect the real demand, and some people really benefit from the large model."
But he also reminded to avoid falling into the "super application trap", "An app with 1 billion dau is called a success." It is already the thinking logic of the mobile Internet era. In the AI era, "super capable" applications are more important than "super applications" that only look at DAU. As long as they can bring great benefits to the industry and application scenarios, the overall value will be greater than that of the mobile Internet.
"In the future, millions of intelligent agents will appear, forming a huge intelligent agent ecosystem." Robin Li said.
once again expressed his views on the hotly debated debate between open source and closed source in the industry, saying that some laymen confuse model open source and code open source. In his view, under the same parameter scale, closed-source models have better capabilities than open-source models. If open-source models want to equal the capabilities of closed-source models, they need larger parameters, which means that the cost of inference will be higher. The speed will be slower.
"The open source model does not allow you to stand on the shoulders of giants to iterate and develop." He said that some companies use open source models to modify models, but only create an orphan model that cannot benefit from the continuous upgrade of the basic model and cannot be shared. computing power. The open source model is only valuable in a few fields such as academic research and teaching, and is not suitable for most application scenarios, especially in a fiercely competitive market environment. To make the company's business efficiency and cost better than those of its peers, commercial closed source Models are the most capable.
Wang Jian: The potential of gpt has not been fully explored
waic In the 2024 roundtable, Wang Jian, academician of the Chinese Academy of Engineering, director of Zhijiang Laboratory, and founder of Alibaba Cloud, said that the potential of gpt has not been fully explored to this day.
"AI has a very long past, but a very short history." Wang Jian recalled that when he was in college, he had already imagined the future of AI. "That was in the early 1980s, and I have been waiting for decades now, and AI has not come yet."
However, decades of development does not mean that AI remains unchanged. "The next ten years will be a very exciting decade." Wang Jian said using neural networks as an example. The neural networks studied in the past had only three layers, and each layer had only two or three nodes; today, the scale is even larger. , and the calculation process is no longer visible.
When talking about the impact of AI technology on the industry, Wang Jian believes that it mainly lies in two aspects. First, as long as it is a new technology, there will definitely be new big companies; there will also be big companies that will be reborn from the ashes. "If new technologies come out and no new big companies emerge, there will be a question mark as to whether it is a disruptive technology."
In Wang Jian's view, considering the computing power, algorithms, and data thresholds required for large models, AI We will be friendly to big companies, but friendliness does not mean tolerance. There will definitely be new big companies emerging, and there will definitely be big companies reborn. Large companies may feel that AI is a revolutionary tool, while small businesses will definitely feel that it is a revolutionary tool. I think big companies must also realize that this is a revolutionary tool, and then this change will come.
Wang Jian also talked about the human factor in new technologies. He believes that AI will have an impact on every department, but it is difficult for all departments and everyone to embrace AI.For companies of different sizes, large companies will definitely feel that AI is a revolutionary tool; small businesses will definitely feel that it is a revolutionary tool. "If big companies also realize that this is a tool for revolution, then changes will come."
Jing Xiandong: AI will bring intergenerational upgrades of services like the Internet
At the main industrial development forum on July 4, Ant Group Chairman and CEO Jing Xiandong said that professional agents can solve the key problems of general large-scale models in rigorous industrial applications. Ant Group is working with industrial partners to build a professional agent ecosystem, accelerate industrial applications, and promote service upgrades.
Jing Xiandong said that in the era of mobile Internet, QR codes have made mobile payment a daily routine for everyone, and "scanning" allows small merchants to enjoy the convenience of payment at the lowest cost. "In the era of AI, we are also exploring how to make AI as convenient as scanning QR code to pay, so that the dividends of AI technology development can benefit more people."
However, Jing Xiandong admitted frankly that the industry generally believes that universal large models Implementing rigorous industries is faced with three "capability shortcomings": relative lack of domain knowledge, difficulty in making complex decisions, and dialogue and interaction do not equal effective collaboration.
In order to solve these problems, Jing Xiandong pointed out that professional agents are an effective path for general large models to be implemented in rigorous industries. Through the deep connection of professional agents, al will bring intergenerational upgrades of services just like the Internet. In his view, in the mobile Internet era, apps and small programs are service carriers, and in the future, they will be promoted in the direction of professional intelligence.
"The intelligent user experience in the future will definitely not rely on just one big model, but will require in-depth collaboration across the entire industry. It will require the participation of many professional agents, each performing their own duties. Ant insists on taking the road of openness and building professional expertise together with the industry. Intelligent agent ecology." Jing Xiandong said.
Xiao Song: AI reshapes the future of industry, turning imagination into productivity
7 4 July 2024 Waic 2024 afternoon industry development main forum, Dr. Xiao Song, Siemens Global Executive Vice President, Chairman, President and CEO of Siemens China, made a speech " Keynote speech: Reshape the future of industry with AI and turn imagination into productivity.
Xiao Song said that today’s explosive growth of AI technology will definitely accelerate the realization of the “industrial metaverse”, enrich the connotation of the “industrial metaverse”, and ultimately reshape the entire industrial value chain. To this end, Siemens cooperated with Microsoft and developed the world's first generative AI for industrial scenarios based on Microsoft's large language model and Siemens' industry experience. This kind of generative AI allows engineers to generate complex automated codes by inputting natural language, greatly lowering the programming threshold and shortening development time. The tool
is now available on Schaeffler’s production lines.
According to Xiao Song, a large number of AI and scene combinations have appeared in Siemens’ Chengdu digital factory. To date, Siemens has implemented nearly 100 AI projects, most of which came from the innovations of front-line workers and technicians. It has reduced factory manufacturing costs for five consecutive years and the process quality is as high as 99.999%. This is the mutual achievement of human wisdom and AI.
It is understood that in the future, Siemens will join hands with more partners to accelerate the generalization and scale of industrial AI with the new paradigm of "basic model + intelligent application".
Xu Li: The AI industry is very hot but has not yet reached the "super moment", and applications are needed to support
On the afternoon of July 4, Xu Li, chairman and CEO of SenseTime Technology, pointed out in his keynote speech that the current AI industry is indeed very hot, but " It has not yet reached a super moment” - because AI has not yet truly entered industry vertical applications or caused widespread changes.
Xu Li believes that the current large model is just a "memory device" that only memorizes all the knowledge points. The only little bit of intelligence is actually a "high-order logical thinking chain" behind the data from the Internet. .
"But in the past two days, I suddenly felt a little change of mind. Because my retired middle school teacher kept asking me in the group, how to use AI to write copy, generate blessing pictures, and send them to his retirement Waiting in the group. I suddenly thought that in fact, super moments and applications are mutually beneficial.Only the cognitive changes brought about by super moments can finally promote application. Looking back, if we have any application to support, then our current moment is a ‘super moment’. So, application is the key to ‘super moments’. " Xu Li said.
Xu Li pointed out that super moments and applications are mutually beneficial. Only when "super moments" bring about cognitive changes can applications be promoted. If there are applications to support it, then the present is a "super moment." "It's like It’s the same as the iPhone. Because of the platform, there will be an app store ecosystem later. So I think whether this era is a super moment for AI, a key lies in application. "
Xu Li believes that if we want to promote the arrival of the AI super moment, we need large models that can demonstrate excellent in-depth thinking capabilities. Then synthetic artificial data, especially advanced thinking data, is very important. So the more application scenarios there are , can form better, high-quality data; secondly, natural interaction without delay. The terminal side is actually a very important breakthrough point, promoting the terminal side optimization of the model, and real-time interactive changes will be smoother. Only the computer resources of the computer and cloud can become a completely natural interaction mode; third, all generation must be controllable. You don’t need to do it well, but you need to know where you are not doing well, and When it comes to making some modifications, with such a boundary, only truly controllable technology and sustainable technological development can be achieved.
Xu Li emphasized, “I hope to welcome AI with many of the ecosystems here. Super moment. "
Liu Qingfeng: Education needs to cultivate new human beings standing on the shoulders of AI
In the AI and Education Forum with the theme of "Education Reform and Talent Cultivation in the Intelligent Era" held on July 4, iFlytek Chairman Liu Qingfeng focused on the latest Spark model progress and empowering educational practice.
Liu Qingfeng said that AGI, represented by large models, is setting off a wave of educational reform globally.
Internationally, since the advent of chatgpt, 89% of college students in the United States have used it to write homework. It was resisted by many schools; in September 2023, UNESCO released the "Guidelines for Generative Artificial Intelligence in Education and Research"; in October, the White House launched the first set of regulatory regulations on generative AI to support educators in deploying AI Supported educational tools stimulate the potential of AI to transform education. At the same time, China attaches great importance to the application of large models in the field of education. In March 2024, China will be committed to cultivating a large number of digitally literate teachers to integrate AI into education and teaching. and management of the entire process and link; the Ministry of Education released 4 actions, proposing to "promote large models from classrooms to applications" to boost AI-enabled education
As early as June 27, iFlytek was the first company based on national production computing power. iFlytek Spark v4.0, a large open model trained by the platform "Flying Star One", is benchmarked against gpt-4 turbo. Especially in the field of education, it has multiple capabilities such as complex instructions, complex reasoning, spatial reasoning, and multi-modal understanding. The improvement has allowed Spark Model to rank among the best in third-party high school and college entrance examination evaluations. At the same time, Spark Model's code capabilities have continued to improve, with the code adoption rate increasing from 30% to 52%, and the overall efficiency improvement exceeding 15%, ranking first in the domestic echelon. . In addition, the Xinghuo speech model and the Xinghuo image and text recognition model have been fully upgraded, and the recognition capabilities of complex scenes such as multi-lingual speech, smear correction, and handwritten formulas have been further enhanced.
benefits from the improvement of the Xinghuo model base. iFlytek's education series intelligent hardware iFlytek AI learning machine has significantly improved in semantic understanding, multi-round interaction, graphic and text capabilities, etc., and can realize super-humanoid question-answering and interactive "ai 1-to-1". According to statistics, ai. In the Q&A pilot, compared with the original problem-solving video learning, the student learning completion rate increased from 67% to 90%, and the problem solving rate increased from 72% to 93%. At the end of the
keynote speech, Liu Qingfeng said: "ai. It will go down in history for solving the urgent needs of mankind. Education needs to cultivate new human beings who stand on the shoulders of AI."
Meng Pu: 5g+ai is a booster for the large-scale expansion of generative AI technology
At the waic 2024 meeting on July 4, Meng Pu, chairman of Qualcomm China, said that although the current research and development and development of generative AI Applications are mainly concentrated in the cloud, and cloud computing will still play an important role, but if 20% of the generative AI workload is moved to the terminal side, it is expected to save $16 billion in computing resource costs by 2028.
He pointed out that this. The close integration of such terminals and the cloud will become the key to promoting the large-scale expansion of generative AI and accelerating digital transformation. At the same time, in order to promote the widespread application of generative AI, we also need to extend its capabilities to smart devices used in daily life. , such as smartphones, mobile PCs and intelligent connected cars.
To realize the implementation of generative AI on the terminal, it is necessary to build a high-performance AI processor on the terminal side; it is also necessary to train and optimize the generative AI model to make it The volume is getting smaller and the efficiency is getting higher.
idc predicts that China's new generation of AI mobile phone shipments will reach 150 million units in 2027, and the market share will exceed 50%. In terms of PCs, consulting companies predict that the number of AI PCs will increase. The penetration rate will increase from 2% in 2024 to 65% in 2028.
Zhipu Zhang Peng and minimax Yan Junjie are on the same stage: It is a good thing to reduce the price of large models, but it cannot last long.
July 4, at the waic 2024 industry development main forum. , focusing on the theme of "Construction of a New Value Chain Driven by Large Models", Minimax founder and CEO Yan Junjie, Zhipu AI CEO Zhang Peng, National and Local Co-constructed Humanoid Robot Innovation Center General Manager Xu Bin, Yufeng Future founder and CEO Xie Ling, the four of them had a collision of ideas.
Zhang Peng said that the essential characteristic of a large model is that it can provide general capabilities on the basis of a model and can solve a series of scenarios and application needs. Zhang Peng believes that if generative AI is to be used to empower the real economy, it is necessary to build more general and basic capabilities.
In the future, Zhang Peng believes that in addition to accuracy, the other capability breakthrough point is multiplicity. Modal. Why does
need to be multi-modal? Zhang Peng explained that because people are multi-modal when solving problems in the real world: in addition to natural language, there are also vision, hearing, touch and common sense, so if a large model can Breakthroughs in this area will benefit AI.
Zhang Peng also believes that the current "price war" among large AI models will not continue, because in his opinion, this is not a normal business logic. Major models have reduced their prices one after another, mainly because the technology itself is getting better and better and the cost is getting lower and lower, but too much is not enough. The real value empowerment should be realized step by step.
Zhang Peng said, “We provide everyone with better high-quality services and hope that everyone will use this service to create greater value, and then this part of the value will be passed back in reverse. This is a normal and reasonable market value logic. .” Therefore, Zhang Peng believes that the current price war does benefit users, but it will certainly not last long.
Yan Junjie believes that overall, the continued decline in prices of large models is a very positive thing. Because it should be lowered. At the same time, as it decreases, the effect should be better. But for large-scale enterprises, the advantage of lower prices is that you can have more users, more online users, more traffic, and generate greater value, which is more based on this traffic value. to find a good business model.
Yan Junjie said frankly that the core problem of the current large model of is that the error rate (illusion) is still relatively high. For example, gpt-4 has an accuracy rate of only 60% and 70% on many test indicators, which means it has an error rate of 30% and 40%. Domestic models overall have an error rate of 60% to 70%. And why are all large model products in the form of dialogue? Because the error tolerance rate of dialogue is relatively high. Why can't it be a standalone agent? This is because if it is an agent, it requires multiple steps and the error rate is higher, so there is no way to use it.
“So I think the core issue is how to reduce the error rate of large models from 30% or 40% to 3%, 4%, or 2%.The core sign that AI has evolved from a tool that assists humans to being able to complete work independently is the overall reduction in error rates. This is critical to greater social value. " Yan Junjie said.
After the meeting, Yan Junjie revealed in a conversation with TMTpost agi and others that a "killer" AI super application (killer app) similar to WeChat, Douyin, and Toutiao will have to be at least three years old. It can only be realized after 20 years.
Zhang Peng finally said that in the future, the new value chain brought by AI will show the characteristics of "single point explosion and comprehensive breakthrough", that is, the changes caused by AI will be comprehensive and involve various industries. , various fields, you can expect an "exponential" value creation.
Hundreds of large-scale model competitions, 25 humanoid robots show technological innovation
The exhibition scale, number of exhibitors, number of highlight exhibits, and number of new products launched at this conference all reach The highest in history. In the exhibition part of the conference, more than 500 companies including Tesla, Microsoft, Dell, Lenovo, Alibaba, Tencent, Baidu, iFlytek, and SenseTime participated in the exhibition, with more than 1,500 exhibits, including hundreds of domestic and foreign products. Large models made a collective appearance. Not only leading companies such as Baidu, Alibaba, Tencent, SenseTime, iFlytek, etc. competed on the same stage, but newcomers such as Minimax, Baichuan Intelligence, and Zhipu AI also appeared confidently with large model-related products and the latest applications. . Microsoft, Dell Technologies, Lenovo, ZTE, Kingsoft Office, and Ant Group are all ready with end-side applications.
Tesla’s second-generation humanoid robot Optimus made its debut at waic this year. Tesla said that this exhibition will "witness the further evolution of humanoid robots."
It is understood that the second-generation Optimus robot made its debut on December 13 last year. There have been significant improvements in terms of weight, flexibility, etc. Some analysts believe that if it continues to iterate at the current speed, the Optimus robot may soon replace humans in many fields. Compared with the previous generation products, its improvements include: using all robots. Actuators and sensors independently designed and manufactured by Tesla; the overall appearance design is more refined; walking speed is increased by 30%; weight is reduced by 10 kilograms, while balance and body control are improved; equipped with new hands that can grasp Hold heavier objects and perform more delicate operations.
Tesla has previously said that the second-generation Optimus robot will first be used in its manufacturing plants, and the company will start selling the robot once its practicality is proven.
Tesla CEO. Musk also responded to Tesla loyalists with a trillion-dollar robot prediction. Musk said Tesla could one day make $1 trillion a year through robots "if the price-to-earnings ratio is 20 or 25, or something like that." , that means the market value of Optimus alone will reach $25 trillion." Musk emphasized that Tesla's valuation may reach 10 times that of the most valuable company currently.
In the h1 booth on the outside, in front of the Humanoid Robot Pioneer Array booth, there are crowds of people watching the "Eighteen King Kong" - composed of eighteen humanoid robots, each of them has a code name, namely Qinglong, x02-lite , Qingbao Robot, Zhiyuan Expedition A2, Fourier gr-1, Dianke Robot No. 1, Xingdong No. 1, Kepler Forerunner k1, Little Crab, Kuafu 4th Generation, xr-4, h1, song, Guanghua One No., Titanium Tiger Yaoguang.
Among them, the gr-1, the first full-size humanoid robot released by Fourier last year, has taken the lead in mass production and delivery. The gr-1 on display today has achieved advanced upgrades in environmental perception, simulation models, motion control optimization and other aspects. The company said that in the future, Fourier will adhere to its original intention, continue to carry out research and development innovation in robotics technology, and empower life with robotics technology in a wider range of application fields through multi-party cooperation.
At the same time, Yunshen Technology disclosed for the first time at the exhibition the latest results of the cooperation between the robot dog and Zhejiang University's "AI Brain". The large-model-based robot intelligent decision-making system "Robot Cloud Brain" empowers the "brain" of the x30 robot dog, equipped with The system's x30 robot dog can not only speak and understand natural human language, but also learn to recognize and understand abstract images, appreciate Van Gogh's paintings, and sense and soothe human emotions.
In addition to humanoid robots and “embodied intelligence,” large models and generative AI applications are also key focuses.
Among them, Alibaba’s booth displayed Tongyi Qianwen and Digital Human products; Tencent mainly displayed Yuanbao, Yuanqi, Digital Human, AI implementation cases, etc.; Baidu did not bring “Kunlun Core Technology” this year, but directly displayed Baidu’s marketing efforts. AI large model assistant products in , finance and other fields.
"Star Chasing AI" - a public welfare project initiated by Alibaba Group and many internal and external developers made many participants feel the warmth of technology. Based on Alibaba's self-developed modelscope-agent framework, multiple services of Tongyi Big Model are called to realize the generation of a one-sentence story summary to a complete audio picture book, taking autistic children into the sea of stars.
Zhipu AI disclosed during the conference that the current Zhipu AI MAAS open platform has more than 400,000 corporate customers and developers, with an average daily call volume of 60 billion tokens. In the past four months, daily API consumption has increased 90 times. above. Its booth mainly displays the "Smart Town" that empowers thousands of industries with the open platform bigmodel.cn and the product matrix of the smart model as the core.
Baichuan Intelligent, which is less than 100 meters away, displayed the AI assistant application Baixiaoying and large-scale medical model implementation cases at the booth.
is the leading company of AI Infra. At the waic 2024 AI Infrastructure Forum held on July 4, Wuwen Core Dome co-founder and CEO Xia Lixue launched the world's first kilocalorie-scale heterogeneous computing power mixed training platform. The hybrid training cluster computing power utilization reached a maximum of 97.6%.
At the same time, Xia Lixue announced that the Wuwen Core Dome infini-ai cloud platform has integrated large-model heterogeneous kilo-calorie mixed training capabilities. It is the world's first platform that can perform single-task kilo-calorie mixed heterogeneous computing power mixed training, with 10,000-kilo-calorie expansion. It supports six types of heterogeneous computing power including amd, Tianshu Zhixin, Muxi, Moore Thread, and nvidia, more than 30 models in total such as qwen2, glm4, llama3, gemma, yi, baichuan2, chatglm3 series, as well as amd, Biren , Cambrian, Suiyuan, Haiguang, Tianshu Zhixin, Muxi, Moore Thread, nvidia and other large model mixed training with more than 10 kinds of computing cards.
In addition to the above-mentioned companies, this year's waic includes multinational companies such as Google, Microsoft, Schneider Electric, and Amazon Cloud Technology.
Among them, Amazon Cloud Technology will demonstrate a wealth of generative AI application scenarios during this conference, the powerful functions of enterprise-level generative AI assistant Amazon Q, and will announce the Shanghai Lingang Science and Technology Innovation Investment Management Co., Ltd./Shanghai AI Industry Investment Fund Accelerate cooperation with Amazon Cloud Technology to build and upgrade businesses.
's new Amazon Q developer agents are a unique feature that can autonomously perform a series of tasks from function implementation, documentation writing, code refactoring to software upgrades. In performing these tasks, Amazon Q achieved the highest scores of any software development assistant available today, scoring 13.82% on the benchmark coding proficiency database swe-bench and 13.82% on the swe-bench (light version) The score is 20.33%.
Amazon Cloud Technology showed in a document presented to TMTpost AGI that the value proposition of Amazon Q: 1) Improve the work efficiency of all employees of the enterprise; 2) The number of data sources that can be integrated exceeds other similar products on the market; 3) Enterprise Level data security privacy control; 4) helps every employee turn conversations into generative AI-driven applications in seconds.
In addition, on July 4, Infinite Lightyear (inf), a credible large-scale model company founded by Qi Yuan, Dean of the Shanghai Institute of Scientific Intelligence (hereinafter referred to as: Shangzhi Institute) and Haoqing Distinguished Professor of Fudan University, was released at waic 2024 Trusted light language large model and its technical report. It is reported that the trusted light language large model combines large language models and symbolic reasoning to effectively solve the hallucination problem, greatly enhance the credibility of the model, and empower vertical fields such as financial services and medical diagnosis, making generative AI truly a new productivity tool. In the evaluation of financial and medical vertical fields, the trusted light language large model surpassed OpenAI's trillion-parameter large model gpt4-turbo. At the same time, as a "specialized" tens-billion-scale large model, it effectively improves inference accuracy and reduces service costs.By solving the major challenge of model illusion problems for enterprise-level applications, Infinite Lightyear has penetrated deeply into financial and medical scenarios, and has currently served many leading companies and institutions. Dr. Xu Yinghui, co-founder of
Infinite Lightyear, believes that after experiencing the "Battle of Hundreds of Models", the industrial application of models has become the focus of the development of large models. However, when going deep into specialized scenarios, the performance of current general-purpose large models is not ideal. Instead, large models in vertical fields need to be more accurate and more credible to become "financial advisors", "professional doctors" and experts in more fields. Let everyone use AI technology and benefit from it.
Domestic AI company Hehe Information was at the booth, focusing on its Scanner Almighty team and the team from South China University of Technology, applying aigc technology to the image digital restoration of the fragments of Dunhuang manuscripts, and jointly created an AI ancient book restoration model. During the
conference, the Hehe Information booth opened a text restoration experience project on synthetic samples of Dunhuang posthumous writings. The public can scan the sample scrolls at different locations and witness that AIGC technology can restore damaged ancient books through glyph repair, fading repair, background completion, etc. Hope for recovery.
“The increasing maturity of intelligent scanning technology can not only facilitate the improvement of users’ current work and study efficiency, but also provide users with faster and more accurate access to and retrieval of information and the creation of personal digital information asset libraries in the long run. An effective tool is also valuable for the inheritance of culture, art, and history. "According to Cao Chaoyang, head of Hehe Information Scanning Almighty King Division, at this stage, intelligent high-definition filters can use AI technology to provide excellent image scene processing. Solution to solve dozens of scanned image problems with one click. Intelligent scanning technology has rich application potential in different fields, and it is expected to explore more innovative directions in the process of deepening the needs of scenarios.
To summarize, waic seems to be still popular this year, and hundreds of large model and AI-derived companies are ushering in a new era. But today, more than a year after the surge, hundreds of domestic large models have shown indistinguishable features and innovations, which may require relevant companies to rethink the technology and commercialization of large models. At the same time, Musk, a frequent visitor, was absent from this year's waic, but Tesla robots appeared, confirming the importance of AI technology products. Yan Junjie, founder and CEO of
minimax, also admitted to TMTpost that in the first half of this year, the company began to realize that in some productivity-oriented scenarios, “we began to have local advantages.” "Frankly speaking, I think most (domestic) companies have not differentiated themselves yet. Everyone is similar, maybe the model level is similar, the products are similar, and then they start to 'compete for price.' I don't think this is a bad thing. In fact, it forces everyone to do better technological innovation.”
(This article was first published on Titanium Media app, author｜Lin Zhijia, editor｜Hu Runfeng)
.