Another company entering the big model, what is the advantage of SenseTime's launch of "Daily New SenseNova"?

Wang Xiaogang, co-founder and chief scientist of SenseTime, believes that SenseTime's big model has two advantages: the first is infrastructure, and the second is that SenseTime has penetrated into industries and scenarios in the past few years, accumulating a variety of multimodal tasks, which are like raw materials.

Tencent News "Periscope" Liu Yong

On April 10, 2023, SenseTime held a technical exchange day to announce the strategic layout of promoting the development of AGI (General Artificial Intelligence) with "big model + big computing power", and announced SenseNova's "Daily New SenseNova" big model system under this strategy, launching a variety of large models and capabilities such as natural language processing, content generation, automatic data annotation, and custom model training.

At the event, SenseTime displayed a series of generative AI models and applications such as AI literate diagram creation, 2D/3D digital human generation, and large scene/small object generation, as well as SenseTime's R&D system for integrating and innovating "big model + big computing power" based on the AI device SenseCore.

It is reported that SenseTime has a total of 27,000 GPU chip cards on the device, which can output 5.0 exaFLOPS total computing power, which is one of the largest intelligent computing platforms in Asia.

Why is it called "Daily Update"? SenseTime said that "Gou is new, every day is new, and every day is new" is selected from the Chinese Confucian classic "Liji University", which means that if it can be updated every day, it should be kept new every day, and it should be updated every day. This is like a new productivity revolution brought to the industry by the "daily new" big model, and SenseTime has incorporated AGI services into the production and life of enterprises, quietly setting off a storm of productivity innovation.

Xu Li, chairman and CEO of SenseTime, said at the press conference on the 10th: "In the era of AI big models, the three elements of data, algorithms and computing power are also undergoing new evolutions, and the amount of large model parameters will increase at an exponential rate, and the amount of data will also grow on a large scale with the introduction of multi-modality, so it will inevitably lead to a sharp increase in demand for computing power." We build infrastructure in the AGI era with SenseCore, an AI device. ”

"AGI has spawned a new research paradigm based on a powerful multimodal base model that continuously unlocks new capabilities of the base model through reinforcement learning and human feedback to solve massive open tasks more efficiently. AGI will realize the evolution from 'data flywheel' to 'intelligent flywheel', and eventually move towards human-machine common intelligence. Wang Xiaogang, co-founder and chief scientist of SenseTime, said: "SenseTime has established a full-stack large model R&D system and has been implemented in multiple industry scenarios, and the diversity of scenarios, the complexity of tasks, and the richness of data fully demonstrate the capabilities and future potential of our large models. We will continue to promote infrastructure construction and look forward to joining the era of AGI with our partners. ”

A few days ago, Wang Xiaogang said when talking about SenseTime's advantages in entering the field of large models, he believes that SenseTime's advantages have the following two aspects:

First, in artificial intelligence startups, it is rare to see a company like SenseTime that puts a lot of resources into infrastructure construction, SenseTime has made a large device, in addition to powerful computing power, there is also 5000 P computing power, thus building a powerful training system and supercomputing system.

Second, as a platform company, SenseTime covers many industry lines. In various industry lines, different types of data have been accumulated, as well as a description of the problem.

"When we want to make a general artificial intelligence big model, we compare it to nuclear fusion, to produce a nuclear weapon, you have to have a nuclear device, which is some of our basic hardware systems." You also have to have nuclear raw materials, nuclear raw materials, it is to have some very rich data in various industries, whether it is a task. He said.

He went on to say, why did ChatGPT succeed? That's because when it gets this data from the Internet, it will find that natural language can cover a variety of complex tasks. But if we extend to multimodality, to vision, these complex tasks you can't get data directly from the Internet.

Therefore, in his view, SenseTime has the innate advantage of doing a large model of general artificial intelligence, a good infrastructure, a systematic infrastructure. The other is that SenseTime has gone deep into the industry and scenarios in the past few years, accumulating a variety of multimodal tasks, which are used as raw materials to finally make this model successful.

In addition, "SenseNova" provides a variety of flexible API interfaces and services for government and enterprise customers, including image generation, natural language generation, visual perception general tasks and annotation services. Customers can call the AI technical capabilities of the "Daily New SenseNova" large model according to actual application requirements, and realize various AI applications with low threshold, low cost and high efficiency.

Under the strategic system of "one platform and four pillars", SenseTime's "SenseNova" large model system has supported business sectors such as smart cars, smart life, smart business, and smart cities, opening up the application closed loop of multiple fields and industries.

In the field of intelligent driving, a large number of long-tail categories are required to require high-precision vehicle-end models. By producing high-precision vehicle-end models with large models, the accuracy of FEW/One/Zero Shots in the long-tail category has been greatly improved, and the average accuracy of the focus categories has been improved by 3%. In addition, the large model provides high-precision intelligent annotation capabilities, provides core functions for data closed-loop, greatly reduces the amount of data that needs to be manually annotated, and accelerates the improvement of model accuracy. Thanks to the large model capability, SenseTime has realized BEV surround view perception, achieved high-precision recognition of 3,000 types of objects, and built a multi-modal large model of autonomous driving that integrates perception and decision-making, bringing stronger environment, behavior, and motivation decoding capabilities.

In the field of biomedicine, SenseTime's AI device provides AI inference computing power for large protein structure models, and provides R&D platform and training computing power for protein interaction models. SenseTime collaborated with Baiying Technology to train an antibody affinity prediction model. Through high-performance computing optimization, the inference time of large model protein structure prediction is reduced from hours to minutes, so that the protein structure prediction performance meets the standard of industrial applications, and the antibody screening efficiency is increased by 60%.

From the perspective of SenseTime's Chinese natural language, digital human generation, large scene roaming, small object reconstruction, Wensheng diagram, etc., enterprises can call the AI capabilities of large models more flexibly according to their own needs, realize the deployment of AI technology in actual business links with low threshold, low cost and high efficiency, and effectively realize the transformation of technical power into productivity.

At the same time, "SenseNova" has also brought breakthroughs to SenseTime's own business. For example, in the field of intelligent driving, based on the visual large model, SenseTime has realized the mass production of BEV surround view general perception algorithm that can recognize 3,000 types of objects, and also built a multi-modal model of autonomous driving that integrates perception and decision-making, bringing stronger environment, behavior, and motivation decoding capabilities.

Natural language is a key means of human-machine communication, and SenseNova has also launched SenseTime's newly developed language model "Discussion SenseChat". As a natural language processing model with hundreds of billions of parameters, "Discussion SenseChat" uses a large amount of data training and fully considers Chinese context, which can better understand and process Chinese text. At the event, "Talk SenseChat" demonstrated excellent multi-round dialogue and comprehension of very long texts. SenseTime also showcased several innovative applications supported by the language model, including: programming assistants that help developers write and debug code more efficiently; Health consultation assistant to provide users with personalized medical advice; PDF reading assistant that makes it easy to extract and summarize information from complex documents.

The diffusion model sparked the popularity of AIGC applications, and SenseTime showcased a series of generative AI models and applications such as "SenseNova", such as various AI graphics creation, 2D/3D digital human generation, and large scene/small object generation:

The "SenseMirage" Wen Sheng Picture Creation Platform shows the powerful Wen Sheng picture ability with real light and shadow, rich details and varied style, and can support the generation of 6K high-definition images; Customers can also train and build models based on their needs.

The "SenseAvatar" AI digital human video generation platform only needs a 5-minute live video material to generate a digital human avatar with natural voice and movement, accurate lip shape, and multilingual proficiency.

The "Qiongyu SenseSpace" and "SenseThings" 3D content generation platforms can efficiently and cost-effectively generate large-scale three-dimensional scenes and refined objects, opening up new imagination space for metaverse and virtual and real integration applications.

These powerful and easy-to-use content generation capabilities brought by "SenseNova" will change the production paradigm of the content production industry, break through the ceiling of content creativity, and will reshape the ecology of the content production industry and open up new growth space.

Whether it is a large language model, a literary graph or a digital human generation, it is inseparable from the computing power support of large-scale AI infrastructure. SenseCore, a SenseTime AI device, with industry-leading computing power output capabilities, ultra-large model training and large-scale inference capabilities, will become a leader in infrastructure services in the era of AGI and large models.

Based on the AI device SenseCore and the "SenseNova" big model system, SenseTime provides industry partners with a variety of large model-as-a-service (Model-as-a-Service) covering automated data annotation, custom large model training, model incremental training, model inference deployment, and development efficiency improvement.

Automatic data labeling based on pre-trained large models can achieve nearly 100 times more efficient than manual data labeling.

Large model parallel training and model incremental training services can help customers quickly train models using their own data, including the development of vertical industry models on top of pre-trained large models, and the production of custom models with thousands of rows and faces.

The model inference deployment service can improve the efficiency of large model inference by more than 100% and reduce the cost of providing services with models.

SenseTime has also opened a large number of pre-trained models and AI development toolchains to industry developers, fully empowering customers to improve development efficiency.

AGI is not a theatrical carnival, but an opportunity for productivity upgrading. In the process of becoming a programmer, artist and creator in ChatGPT, SenseTime has injected the productivity iteration driven by AGI into more fields, industries, enterprises and scenarios. It can be said that on the road to driving AGI, SenseTime has found a key breaking point.