/PRZWT/SHENZHEN, March 6, 2025 - In Bao'an District, Shenzhen, the government affairs model is connected to the dual-core drive of "Tencent Huanyuan + DeepSeek", empowering 31 business scenarios to be intelligently upgraded; Shanghai Pharmaceutical jointly develops intelligent agents with Tencent Cloud, and more than 30,000 employees enter a new era of smart office. The pharmacy knowledge base is combined with the DeepSeek model to help provide precision pharmacy advice...
DeepSeek's localized deployment is the epitome of the accelerated implementation of AI technology. In early 2025, the DeepSeek-R1 open-source model was "hot" with its superior reasoning ability and cost performance, and all over the country seized new opportunities in AI.
"Continuous breakthroughs in model algorithms and computing power have promoted the universalization of model applications, and the industry will also usher in a real'smart moment '." Tang Daisheng, senior executive vice president of Tencent Group and CEO of the Cloud and Smart Industry Business Group, said that Tencent hopes to provide users with easy-to-use AI, transform cutting-edge AI technology into smart products available to users, help the public's better life, promote innovation breakthroughs and high-quality development in the real industry, and accelerate the universalization of artificial intelligence.
Transforming cutting-edge AI technologies into useful AI products
The high-quality development conference of Guangdong Province held not long ago proposed that Guangdong has two advantages of electromechanical technology and digital intelligence technology, and should make great determination and concentrate on the two fields of artificial intelligence and robotics to build a new pillar of high-tech, high-growth and large-scale industry.
"Big model technology is the core scientific and technological foundation leading to the era of general artificial intelligence (AGI). It is crucial to adhere to self-research and innovation, and continue to explore AI cutting-edge technologies such as big models." Tang Daosheng introduced that in 2023, Tencent launched a full-link self-developed hybrid model, using a mixed expert model (MoE) structure, which has reached a trillion-level parameter scale.
Recently, Tencent has launched its self-developed hybrid T1 deep thinking model, which is designed for complex tasks, capable of understanding the multiple dimensions and underlying logical relationships of problems, and performing logical reasoning. After Tencent ingot is connected, user feedback Q & A is of high quality, stable and reliable, and is now fully open.
At the same time as model innovation, AI needs to be engineered to move from the laboratory to practical application. Tang Daosheng believes that only by turning advanced AI technology into useful AI products can it be truly transformed into productivity.
Tencent has been exploring the combination of cutting-edge model technology and product capabilities that have been accumulated for many years, starting from user requests, and constantly polishing high-availability and high-stability AI products in the scene. At present, it has launched intelligent products such as AI all-round assistant Tencent Yuanbao and intelligent workbench ima.copilot; at the same time, it has also combined AI with users' favorite applications such as WeChat search, Tencent conferences, Tencent documents, WeChat reading, and wealth management to greatly improve users' life and work efficiency and experience.
After the launch of the domestic open-source large model DeepSeek R1, Tencent took the lead in embracing and supporting access, creating a dual-engine architecture of "Huanyuan Large Model + DeepSeek R1", which supports product access for hundreds of millions of users such as Tencent ingot, WeChat search, ima, QQ browser, Tencent documents, and Tencent maps. With more than 20 years of accumulated massive user operation and maintenance and computing power resources, it provides users with a deep reasoning experience.
Tencent Yuanbao launched a "dual-engine" service.
Promote the integration of AI and industry to help enterprises make breakthroughs in innovation
Compared with C-end internet services, the combination of industrial scenarios and AI is also full of imagination and is also the key to promoting high-quality development of the industry.
"We believe that the building of the model is only the starting point, and the landing of the technology in the industrial scene and the creation of value is the end point." Tang Daosheng introduced that when Tencent took the first step of the big model, it insisted on doing "the AI closest to the industry", integrating AI into the industrial scene, helping corporate customers reduce costs and increase efficiency, and innovate breakthroughs. It has been implemented in more than 20 industries such as retail, finance, industry, medical care, education, and cultural tourism.
Open the software and you can quickly write reports and code with AI, and enjoy the improvement of office efficiency brought by AI... Such a scenario has become a reality. In the enterprise application layer (SaaS), AI capabilities have been incorporated into applications such as Tencent Conferences, Tencent Docs, Tencent Lexiang, and AI Code Assistant. Recently, these enterprise applications have also been connected to DeepSeek to improve product capabilities and user experience. For example, Tencent enterprise customer service has fully adapted DeepSeek, linking with DeepSeek's large model through the enterprise knowledge base to improve the quality of Q & A and user satisfaction.
Enterprises can also use Tencent Enjoy to build a knowledge base of industry documents and knowhow accumulated by enterprises over the years. After that, employees can ask questions to intelligent assistants, make efficient use of the enterprise knowledge base, and realize rapid retrieval and application of knowledge. Not only that, individual users can also build their own exclusive knowledge base based on ima to meet personalized knowledge management needs.
In order to lower the development threshold and shorten the distance from model to application, Tencent provides full-stack artificial intelligence development services for enterprise users and developers based on the Tencent Cloud TI platform, covering the whole process of data acquisition, processing, model training, evaluation, and deployment to applications. It supports one-click fine-tuning and inference of mainstream models such as DeepSeek, Hybrid Large Model, and Llama3.
How to reduce the AI "illusion" and make the answer content more secure and reliable? Tencent Cloud Knowledge Engine adopts the RAG (Retrieval Enhanced Generation) technology architecture, and enterprises can use the exclusive knowledge base and professional documents that are highly relevant to their own business, so that the big model can take the "textbook" to the "exam". When solving user problems, the big model will first query the relevant textbook content as the basis, and then make accurate answers.
These power builders, with the high degree of synergy between Tencent's hybrid and DeepSeek dual-mode capabilities, have continuously upgraded their capabilities to better help the industry reduce costs and increase efficiency.
"The big model has landed in thousands of industries and realized new breakthroughs in the industry with intelligence, and the prospect is broad." Tang Daoseng said that as a technology platform enterprise, Tencent will actively play the role of innovation traction, continue to promote technological innovation and ecological open-source construction, and help the industry to upgrade intelligently and achieve high-quality development.
