news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author｜Chang Minxiao

Editor: Qiu Xiaofen

From the Hundred Model Wars in the past two years to the implementation of large model applications this year, compared to the excitement when people first mentioned large models, people are now more concerned about what large models can do for daily life.

36Kr learned that the artificial intelligence big model company "Zhuo Shi Technology" has completed a B+ round of financing of 100 million yuan. This financing was led by the well-known investment institution Shenzhen Tongchuang Weiye, and followed by Qingdao state-owned asset platform Qingdao Haifa and TusStar and other institutions.

"Compared with the research and development of basic large models, we pay more attention to the application and commercialization of large models to help enterprise users truly achieve cost reduction and efficiency improvement." said Tu Jing, founder and CEO of "Zhuo Shi Technology".

It is difficult to implement big models at present, not only because the capabilities of the middle layer and application layer of the big models are not perfect enough, but also because it is difficult to find suitable industries to achieve application scenario focus.

At the same time, on the basis of obtaining market feasibility verification, the large model also needs to have the ability to be standardized and replicated on a large scale in order to truly realize its application.

In response to the difficulties in the current large model market application, Zhuo Shi Technology has provided solutions from three perspectives: computing power, algorithms, and big data.

Image source: ZhuoShi Technology

In terms of algorithms, ZhuoShi Technology independently built the underlying large model, and on this basis built the "Mixture of Experts (MoE)", which is an efficient model architecture used in deep learning models. The MoE architecture dynamically selects different expert sub-models to process input data to reduce the computing power cost of training and improve the efficiency of supporting the application end.

At the same time, the company independently completes the pre-training of the model, and then conducts targeted secondary pre-training and fine-tuning based on the needs of corporate customers. The company also has the ability to build a large model intermediate layer, completing the entire process from basic large models, exclusive large models to model enhancement (knowledge enhancement/decision enhancement/retrieval enhancement/cross-modal enhancement), intelligent agents and business result output.

The company's industry large-scale model scenarios are widely used, such as: high-precision silicon smelting, liquor brewing and other work that is completed "by experience and feeling", and its work process is standardized to generate accurate judgments.

Based on the model, ZhuoShi Technology built the Maas-Agent platform, providing a series of low-code development services for industry models, including basic computing services, data services, model training services, application deployment and construction, and low-code drag-and-drop capabilities, which reduced the cost of large model training by 60%. The company also significantly saved computing power through intent recognition.

Zhuo Shi Technology has cooperated with the government, including the Provincial Finance Bureau, Beijing Civil Affairs Bureau, and Beijing Kangyang Group, to obtain accurate, market-oriented, high-quality user data to feed back into model pre-training.

Image source: ZhuoShi Technology

The company's pre-trained large model products currently have elderly care and education as the main application scenarios, and have launched end-to-end products that combine AI digital humans and industry models.

"We regard the elderly and children as the main scenarios for the large-scale popularization of large-scale model applications," CEO Tu Jing explained the progress of Zhuo Shi Technology's commercialization in an interview. "The elderly care industry is not only a rigid demand, but also has certain national policies and financial resources to support it. Nowadays, the elderly are not very good at using smartphones, and there is a lack of emotional companionship around them. Digital humans driven by large models can interact with the elderly in real time and capture their preferences in health and diet. At the same time, they can give them targeted and accurate advice on chronic disease management and medication guidance."

Due to the lack of domestic health care talents, Zhuo Shi Technology was the first to apply AI digital human lecturer IP to the elderly care education and training scene. According to the needs of different regions and groups, interactive courses are customized to answer and solve the elderly's daily life, basic diseases and other related issues, so as to alleviate the problems of uneven distribution and insufficient number of full-time lecturers.

Image source: ZhuoShi Technology

"Zhuo Shi Technology" was founded in Beijing in 2019. Tu Jing, a former executive of Baidu Maps and Baidu APP, is the founder of the company. The founding team and core members mainly come from major companies such as Baidu and Lenovo.

Currently, ZhuoShi Technology has established cooperation with governments in many parts of the country and established three R&D centers in Beijing, Chengdu and Hainan, which are responsible for R&D operations, AI R&D and data services respectively.

Investors in this round unanimously stated: "ZhuoShi Technology's potential and determination in commercializing industry large models are key factors in attracting investors. In today's rapidly developing technology field, the ability to transform advanced large model technology into practical business solutions is an important criterion for measuring a company's innovation ability and market competitiveness. ZhuoShi Technology has not only achieved relevant advantages in artificial intelligence large model product technology, but more importantly, they have successfully applied these research results to practical problems, providing practical solutions for multiple industries and creating huge business value."

Welcome to communicate, please add WeChat and indicate your purpose

end