The first day of the 2023 World Artificial Intelligence Conference was crowded.
At the end of last year and the beginning of this year, a new round of artificial intelligence boom was set off by ChatGPT. As a hot spot in the field of artificial intelligence this year, large models and intelligent robots are the two "net celebrities" at the 2023 World Artificial Intelligence Conference, and this conference is being held in Shanghai. It is understood that more than 30 large models under Baidu, Huawei, Alibaba, iFLYTEK, SenseTime and other enterprises appeared on the scene, fully demonstrating the revolutionary impact of large models on the industry. More than 20 robots such as intelligent engineering robots, drones, and bipedal robots competed together.
The first echelon of the "universal" large model was unveiled
The "emergence" of ChatGPT and similar products have made "big model" a hot word of the year. From the appearance of more than 30 large models at this World Artificial Intelligence Conference, it can be seen that major Internet companies, communications and technology manufacturers are competing for the direction of "force" and the focus of public attention. At the scene, in addition to feeling the wave of change sweeping through all walks of life, you can also see that every detail of life is slowly being rewritten - sports, painting, writing... The novel scene with a sense of experience based on the large model attracted the young people who were watching the scene to "eagerly want to try".
The reporter saw at the conference site that almost all the products that occupy the "first echelon" of domestic general large models are in place, such as Baidu Wenxin big model, Huawei Cloud Pangu big model, Alibaba Cloud Tongyi Qianwen, SenseTime Ririxin, iFLYTEK Xinghuo cognitive big model, Jingdong Yunyanxi big model, Tsinghua University ChatGLM, Fudan University MOSS and other more than ten products were unveiled one by one. It is worth noting that in addition to the above-mentioned general-purpose large models that many audiences are familiar with, large model products such as SenseTime, Yunzhisheng Shanhai, and Tencent X music, which focus on vertical fields, also attracted the attention of the audience, focusing on the cutting-edge application and technology of the industry.
The reporter found that the largest number of onlookers on the scene was the "experience area" with various tricks, which led the audience to push open the magical door of AI through a relaxed and interesting experience. People can interact with AI in real time in the experience window of generative AI technology to understand a series of generative AI technologies such as natural language processing technology and machine deep learning. Play basketball on the smart basketball court and let the multi-camera visual motion capture system based on SenseTime's big model system capture your highlight moments; Dance with the game IP in front of the "Health Magic Mirror", let the AI human body motion sensor and motion algorithm engine recognition system capture and recognize the joint posture in motion, and experience the digital fitness method...
In an immersive, grounded experience, complex and grand concepts land in front of you. "We're starting to feel that the lives we're used to are changing dramatically." An audience member who has just experienced AI painting and a smart basketball court said with emotion.
Big coffee talks
HUAWEI CLOUD today released Pangu Model 3.0
Ken Hu, Huawei's rotating chairman, spent an afternoon walking around the conference complex, and at the opening ceremony, he said that the biggest difference between this year's conference and previous editions is the emergence of ChatGPT, which has pushed artificial intelligence to a new outlet.
Ken Hu said that the key to the development of artificial intelligence is to be down-to-earth, promote artificial intelligence to go deeper and practical, and empower industrial upgrading. Huawei has two focus points in the development of artificial intelligence: First, build a strong computing power base to support the development of the Chinese industry of industrial intelligence. Second, truly let artificial intelligence serve thousands of industries and serve scientific research and innovation.
Ken Hu said that over the years, Huawei has cultivated computing power, mainly focusing on the root technology of Kunpeng and Ascend. Through architecture innovation, ecological development and co-construction of computing power, etc., to support the construction of the future computing power foundation. In China, customers have diverse computing power needs, and Huawei adopts a variety of modes to build computing power based on actual conditions. For governments, large enterprises, and small and medium-sized enterprises, Huawei has flexible models to choose from and make decisions, so that computing power is no longer a bottleneck.
Ken Hu revealed that the key to the Pangu model released by HUAWEI CLOUD is to go deeper and more practical, face thousands of industries, serve production activities, serve scientific research and innovation, and create value for artificial intelligence. HUAWEI CLOUD Pangu models have been deployed in more than 10 industries, including finance, manufacturing, government affairs, electric power, coal mining, healthcare, and railways, supporting the implementation of AI applications in more than 400 service scenarios.
"On July 7, Huawei Developer Conference will release Pangu Model 3.0, providing unlimited possibilities for innovation in thousands of industries. General Artificial Intelligence (AGI) is ushering in the next golden decade, let us join hands to serve thousands of industries, serve scientific research, and win the new era of artificial intelligence. Ken Hu said.
We can subordinate AI to humans
When the capabilities of AI and robots become more and more powerful, is it becoming uncontrollable for humans? Yang Likun, winner of the 2018 Turing Award and chief artificial intelligence scientist of the Meta AI Basic Artificial Intelligence Research Institute, said that we can set goals that make artificial intelligence honest and subservient to humans, and at the same time, be careful of those goals that make them want to be curious or expose them to resources they shouldn't have. "Of course, it's a very difficult challenge." Yang Likun pointed out that you don't need to do a good job at the beginning, you can start by cultivating a "good little mouse", and then slowly go to "good dog" and "good monkey", and then put it into the simulation environment for testing.
On the other hand, Yang believes that in the long run, the only way to make an AI platform safe and practical is to make it open source. "The AI systems of the future should become a treasure trove of human knowledge, and the way they are trained needs to be based on many sources, so we want to see more open source LLM and more open source AI systems."
Shen Xiangyang, former executive vice president of Microsoft and a foreign academician of the US National Academy of Engineering, said that for artificial intelligence, there must still be a certain supervision and a certain guarantee. As artificial intelligence becomes more and more powerful, in order to achieve a point, we need to recognize the power of society, and we must work together to mobilize companies and institutions such as Microsoft and Google to participate.
The implementation of large models requires iterative tuning
"The general large model can solve 70%~80% of the problems in 100 scenarios, but it may not be able to meet 100% of the needs of a certain scenario of the enterprise." Tang Daosheng said that the general large model is generally trained based on extensive public literature and network information, and many professional knowledge and industry data accumulation is insufficient, resulting in insufficient industry pertinence and accuracy of answers. However, users have high requirements for the professional services provided by enterprises, low fault tolerance, and once enterprises provide wrong information to the public, it may cause serious consequences.
In addition, the landing process of large models also needs to go through a series of links of algorithm construction and model deployment, and each link cannot be "dropped from the chain". The model also needs to be continuously iteratively tuned, which requires the use of systematic and engineering tools.
"With the development of the big language model, industry and society will also move from digitalization and networking to intelligence. In this process, Tencent has always believed that the fundamental goal of AI development is to land in the industry and serve people. Tang Daosheng said.
The robot's uniform dance movements attracted onlookers
The ratio of robots to people will exceed 1:1 in the future
Tesla CEO Elon Musk pointed out in a speech through video that artificial intelligence is a profound change, including profound effects and impacts on civilization.
The "humanoid robot" is still in the development stage
At present, the number of computers and computing power is exploding, Musk pointed out, which means that over time, human intelligence will account for less and less of all intelligence than machine intelligence, which will have a fundamental impact. At present, Tesla's "humanoid robot" is still in the development stage, but Tesla will have a lot of robots in the future.
Musk predicts that there will be a large number of robots on the planet, and the ratio of robots to humans, "Now it seems that there will be a stage where the number of robots will exceed 1:1, and the number of robots on the planet will exceed the number of humans." "This is a development trend, and the positive side is that productivity will be much higher than it was when humans were dominant. Of course, the first thing to do is to make sure that robots are beneficial to humans.
This year, it is expected to achieve L4 or L5 fully autonomous driving capabilities
In terms of automatic driving of cars, Musk mentioned that self-driving technology will liberate people from boring driving, give people more freedom and time, and greatly improve the utilization rate of cars. "Under normal circumstances, a family car has about 10~20 hours of use a week, and most of the time is spent lying on its stomach in the parking lot. However, for a car with fully autonomous driving capabilities, the use time may reach 50~60 hours a week, and the utilization rate of autonomous vehicles is 5 times that of non-fully autonomous vehicles. In addition, Musk revealed that Tesla may have full self-driving capabilities at level 4 or L5 later this year.
Speaking of the development of the Chinese industry intelligence, Musk expressed optimism, "I have always admired the wisdom and drive of the Chinese people. "Once China is determined to do something, it will definitely be able to do this industry well." This is true of all sectors of the economy, including the artificial intelligence industry, which believes that China will have strong capabilities.
Text, photo/Guangzhou Daily all-media reporter Wen Jing Chen Xin
[ Editor: He Wenbiao ]