36Kr Exclusive | Focusing on the implementation of enterprise-level large models, "Shi Yin Zhi Neng" has completed the first round of financing of tens of millions of yuan.

Author | Deng Yongyi

Editor | Su Jianxun

36Kr has learned that "Knowledge Intelligence" has recently completed its first round of financing of several million yuan, led by Beijing Shijingshan District Modern Innovation Industry Development Fund Co., Ltd. After this round of financing, the company's valuation has reached approximately 500 million yuan.

Knowledge Intelligence is an artificial intelligence startup founded in 2019. Its CEO, Wang Chunhui, previously served as the Technical Vice President of Oracle Japan and the Asia-Pacific region, and a Senior Researcher at Google Brain. Its core team also comes from top teams in the NLP field such as Google Brain, Microsoft Research Asia, and DeepMind. 60% of the technical team members hold a doctoral or master's degree, have rich experience in project 0-1, and have published many academic papers in well-known domestic and foreign journals.

The main product of Knowledge Intelligence is an enterprise-level intelligent application platform, focusing on implementing AI in the production processes of enterprises.

CEO Wang Chunhui told 36Kr that since its establishment in 2019, Knowledge Intelligence has been focused on the research and development of key technologies such as NLP (Natural Language Processing) and speech recognition. Before the arrival of this wave of generative AI, Knowledge Intelligence has already conducted a long-term exploration and implementation attempt for large models, including scenario applications such as intelligent customer service and industrial equipment monitoring.

Until the middle of 2023, Knowledge Intelligence has initially completed the research and development of the first version of the large model product and completed the national filing in January 2024.

The "One Leaf Boat" large model independently developed by Knowledge Intelligence is a large language model that has undergone fine-tuning on a base of approximately 540 billion parameters and possesses general artificial intelligence capabilities such as multimodality and multitasking in speech, image, and video.

In terms of size, the current "One Leaf Boat" large model has multiple commercially available models ranging from 0.7B to 45B, which can be efficiently implemented in various scenarios of enterprise applications.

Wang Chunhui believes that after the rapid development of the large model field in the past two years, it has now reached a critical stage of finding landing scenarios. However, once it enters the core links of enterprises, problems such as the large model hallucination will still increase the difficulty and uncontrollability of implementation. And the current business direction of Knowledge Intelligence is mainly oriented towards the To B market. In terms of business implementation, Knowledge Intelligence is more inclined to adopt an application method that combines large models and professional models.

From the perspective of market positioning, Knowledge Intelligence has currently launched solutions including scenarios in smart government affairs, smart policing, smart industry, and smart enterprises.

Taking the landing application in the smart city field as an example, in the past, for law enforcement scenarios such as public security and transportation, or production scenarios in the industrial field, business personnel often had to manually view and review massive amounts of image, audio, and text data.

After the large model can be applied, the product of "One Leaf Boat" can understand and summarize the existing image, video, and other data through multimodal capabilities, greatly improving business efficiency.

Facing the enterprise and government markets, Knowledge Intelligence has also launched a hardware product of AI Xinchuang All-in-One in 2024, which includes computing power, algorithms, model layers, and multi-functional applications, enabling out-of-the-box use and safeguarding user data security.

After being equipped with the "Boat Intelligent Computing Service Operation and Control System" independently developed by Knowledge Intelligence, under the same model conditions, the system can save half of the computing resources without affecting the model performance, and increase the computing resource utilization rate from 40% under the default condition to approximately 90%.

Currently, Knowledge Intelligence has multiple customers in the government and enterprise sectors and has generated stable commercial revenue.

Regarding future plans, Wang Chunhui stated that 2025 will be a year for Knowledge Intelligence to accelerate its marketization. This round of financing will also be mainly used to accelerate the company's research and development in cutting-edge technologies and promote the company's market expansion and application implementation in more fields.