Mission Hills at Sea ｜ Where are the four major innovations ＂significant＂？

On July 6, 2023 World Artificial Intelligence Conference was held in Shanghai with the theme of "Smart World Generates the Future". The convening of the conference coincided with the booming global model and generative artificial intelligence, and the appearance of the latest AI achievements in the world became a highlight of the conference.

At the opening ceremony of the 2023 World Artificial Intelligence Congress, four major innovations made their debut: the scholar’s universal big model system, the digital twin brain of Fudan University, Zhangjiang SUPERCITY, and the universal humanoid robot GR-1 of Fourier. Where are these four major innovations "significant"?

Scholar model can reach the world leading or advanced level in more than 130 evaluations.

"This time, the scholar’s general big model system has been fully upgraded, and the model family has been further enriched, with stronger capabilities and wider coverage modes and tasks. At the same time, the laboratory also conducts all-round open source at the levels of models, data, tools and evaluation to promote technological innovation and industrial progress. " Qiao Yu, assistant director of Shanghai Artificial Intelligence Laboratory, said.

Shanghai Artificial Intelligence Laboratory and Shangtang Technology, together with the Chinese University of Hong Kong, Fudan University, Shanghai Jiaotong University and Tsinghua University, released a brand-new and upgraded "Scholar General Large Model System", which includes three basic models, namely Scholar Multimodal, Scholar Puyu and Scholar Horizon, and the first full-chain open source system for the development and application of large models.

Seven-character quatrains created by "Scholar" according to Zhang Daqian’s "Lake Mountain and Summer Scene" (Photo courtesy of Shanghai Artificial Intelligence Laboratory posted by Yangguang. com)

At present, the scholar’s general large model system can reach the world leading or advanced level in more than 130 evaluations. Among them, Scholar Multi-modal seamlessly integrates various modes such as language, image and video, and realizes the definition of visual tasks through natural language for the first time, and has the ability of multi-modal interaction and cross-modal generation; The upgraded Scholar Puyu is the first officially released 100-billion-parameter-level language model supporting 8K context length in China, and many evaluations have surpassed ChatGPT；. Scholar Sky is the world’s first three-dimensional large-scale model of city-level NeRF real scene, with hundreds of billions of parameters, and it is the first time in the world to realize 4K high-precision modeling and editing of 100 square kilometers of urban real scene.

Qiao Yu introduced that Scholar Big Model has promoted the intelligent process in many industries and fields such as autonomous driving, smart medical care and earth science. In earth science, the global medium-term weather forecast model "Fengwu" has achieved the first time that the effective weather forecast time is 10 days, and it only takes 30 seconds for "Fengwu" to generate the global high-precision forecast results for the next 10 days, which is much better than the traditional model in efficiency.

Universal humanoid robot is expected to lead AI into the "era of embodied intelligence"

With the emergence of "super brain" represented by ChatGPT, the evolution route of artificial intelligence is becoming clearer. At present, the industry generally believes that the next wave of artificial intelligence is embodied intelligence. As the carrier of artificial intelligence entering the physical world, embodied intelligence combines artificial intelligence technology with robot entities, giving the "brain" a "body" that can be dominated, perceived, interacted and acted.

The intelligent universal humanoid robot GR-1 is 1.65 meters tall, weighs 55 kilograms, has 40 degrees of freedom, the maximum peak torque of joint module can reach 300N·m, the walking speed can reach 5KM/h, and it can carry 50 kilograms.

Fu Liye intelligent universal humanoid robot GR-1 (photo by Yang Guangwang reporter Han Xiaoyu)

Gu Jie, the founder of Fu Liye Intelligent, believes that as the next generation ontology of AI with body intelligence, universal humanoid robot is expected to lead AI into the "era of body intelligence", and will also promote the development of special robots to general robots. "Just like the development trend from desktop computers to personal computers to mobile phones, universal humanoid robots will gradually become the mainstream of the industry in the future. At present, the hardware foundation of the universal humanoid robot is basically laid, and the application wave driven by the big model is coming soon. "

"Once the humanoid robot is generalized, it will meet the huge market demand." Gu Jie introduced that GR-1 adopts a highly scalable design, which can realize more AI model and algorithm verification, and explore and try the landing and cutting-edge technology integration in application scenarios such as rescue and disaster relief, old-age care, and domestic service.

The world’s first digital twin brain with the scale of 86 billion neurons.

The human brain consists of 86 billion neurons, and each neuron is connected with as many as 1,000 to tens of thousands of other neurons, forming the most complex neural network known at present, which can generate emotion, self-learning and thinking, and has the ability to adapt to the complex external environment. The goal of the digital twin brain project is to try to reconstruct the simulated brain on the computer, so that there is a significant correlation between the digital brain and the biological brain in the mesoscale spatio-temporal dynamic mechanism.

Through nearly five years’ efforts, the research teams of Feng Jianfeng, Lu Wenlian and Zheng Qibao of Fudan University used the multimodal neuroimaging data of human brain collected by high-precision nuclear magnetic resonance equipment to develop the method of data assimilation and reconstruction of pulse neural network on the supercomputer cluster connected by high-speed network, and constructed the world’s first digital twin brain with the scale of 86 billion neurons.

According to reports, the construction of digital twin brain can be used as a cognitive experimental platform, and also as a medical experiment platform for brain diseases, brain-computer interface and drug testing. By simulating and analyzing the pathological areas related to the basal ganglia of the brain, the research team gave a variety of different deep brain stimuli in the digital twin brain, providing a new digital twin means for better understanding the pathogenesis of Parkinson’s disease and optimizing the clinical diagnosis and treatment plan.

Urban digital twin base to create digital application scene

The development of artificial intelligence industry is inseparable from the landing of scene applications. At the industrial development plenary meeting of the 2022 World Artificial Intelligence Conference, Zhangjiang Group released the concept of "Zhangjiang Future City SUPER CITY", and proposed for the first time to build a pilot area of Zhangjiang Digital Twin "Future City" with the area of 4.1 square kilometers in Zhangjiang Central District.

This year, supported by the industrial cluster of Zhangjiang Artificial Intelligence and Metauniverse Digital Chain, Zhangjiang Group and Overlapping Digital have jointly created a new AI 2.0 paradigm of digital twin "City of the Future", and formed five digital application scenarios for future life, future transportation, future medical care, future finance and future manufacturing by developing the digital twin base of Zhangjiang Science City, which is inclusive, open and win-win.

Immersive navigation solution (Photo courtesy of respondents sent by Yangguang. com)

Shi Zhiru, vice president of Overlapping Digital, introduced that "Zhangjiang SUPERCITY" has built a digital urban cloud native base, which realizes the fusion of time and space data and holographic transparency of traffic information, and has the ability of cross-modal AI intelligent analysis and dynamic simulation of traffic situation. "In the future, the first set of immersive navigation solutions will be launched in China, and citizens can no longer be limited by spatial positioning, experience realistic three-dimensional scenes and appreciate immersive space scenes."

"The digital twin system will become the basic application platform for many applications in the future. If the low-cost rendering technology can be broken through, the digital twin platform will probably become a three-dimensional map service, and both indoor and outdoor scenes can be vividly presented to people at low cost. It has truly become an indispensable part of people’s livelihood work." Zhang Zhaohua, CEO of Overlapping Digital, said.