To create an immersive science fiction space for "The Three Body Problem", the "Shangtang Daily New" big model system has been comprehensively upgraded. Shangtang | Model | System
At the 2023 World Artificial Intelligence Conference "Da Ai Wu Jiang · Ri Ri Xin" Artificial Intelligence Forum held yesterday, the reporter learned that the "SenseNova" large model system of Shangtang Ri Ri Xin has completed multiple upgrades and is being applied in industries such as finance, healthcare, and automotive. Xu Li, Chairman and CEO of Shangtang Technology, stated that the company hopes to continuously promote the improvement of AI infrastructure capabilities through "big models+big devices", build professional big models that understand the industry more and have more expertise, and let the industrial value of big models bloom in thousands of industries.
Under the AGI strategic layout of "big model+big device", Shangtang's big model system is undergoing rapid iteration. As a natural language processing model with billions of parameters, SenseChat 2.0 has broken through the input length limitation of large language models and launched model versions with different parameter levels, which can adapt to the application needs of different terminals and scenarios such as mobile and cloud. Since its first release in April this year, the model parameters of the generative large model "SenseMirage 3.0" have been increased from 1 billion to 7 billion, enabling professional photography level image detail characterization.
Compared with version 1.0, the 2.0 version of the "SenseAvatar" digital human generation platform has improved the fluency of voice and mouth movements by more than 30%, achieved 4K high-definition video effects, and added functions for image generation and digital human singing. The spatial reconstruction efficiency of Qiongyu SenseSpace 2.0 version has been improved by 20%, rendering performance has been improved by 50%, and the mapping time for every 100 square kilometers of the scene can be completed in just 38 hours. The 2.0 version of "Grid SenseThings" achieves millimeter level precision in texture and material restoration of small objects, and breaks through the difficulty of collecting high reflective and mirror like objects.
Xu Li introduced that relying on the rapid iteration of the big model system in the underlying technology, Shangtang is leveraging the multimodal capabilities of big models to empower multiple industrial fields.
In the financial field, Shangtang collaborates with clients such as banks, insurance, and securities firms to use digital humans to carry out intelligent customer service, intelligent marketing, and other work. By integrating big language model capabilities, it provides new functions such as investment research analysis and report writing, achieving cost reduction and efficiency improvement. After mounting the financial knowledge base, digital humans can still 100% output content Q&A based on customer product descriptions and achieve timely updates of information.
In the medical scene, Shangtang has created a Chinese medical language model called "Da Yi" based on massive medical knowledge and clinical data. It has the ability to engage in multiple rounds of conversation in scenarios such as guidance, consultation, health counseling, and decision-making assistance. It is capable of supporting multimodal comprehensive analysis of medical images, texts, structured data, and continuously improving medical language understanding and reasoning abilities.
Combining the comprehensive capabilities of "Negotiate" version 2.0 and "Second Draw" version 3.0, the company also brings various intelligent interaction solutions to mobile terminal customers. In the immersive sci-fi experience space of "Beyond the Gravity" based on Liu Cixin's novel "The Three Body Problem", Shang Tang utilizes the ability of large models to break through the boundaries of imagination and create a futuristic sci-fi journey.
In the field of intelligent vehicles, industry applications such as Shangtang's "Jueying" intelligent cockpit, intelligent driving, and vehicle road collaboration are also breaking through innovation boundaries with the support of large models. In the intelligent cockpit, multi-modal fusion such as vision and hearing is used to comprehensively perceive user needs, and user habits and preferences are recorded through labeled data to provide exclusive personalized services.
Outside the cabin, relying on "large models+large devices", "Jueying" deploys end-to-end cloud collaboration, unifies traffic entry, supports private deployment and tens of millions of application requirements. At the recent 2023 CVPR, Shangtang and its collaborating units proposed the UniAD universal model for autonomous driving, which integrates perception and decision-making. They pioneered the architecture of an autonomous driving model with global tasks as the goal, and their related papers won the CVPR Best Paper Award, proposing new directions for the development of autonomous driving technology and industry.
Based on this, the company is building a transportation system for vehicle road cloud collaboration, utilizing a multimodal and multi task universal large model to develop a roadside visual perception large model, and combining "Qiongyu" 2.0 and "Gewu" 2.0 versions to build intelligent transportation twins and simulations, promoting the evolution of vehicle road cloud collaboration towards a dialogue based interactive mode of the large model.