Shulou(Shulou.com)12/24 Report--

"when we look at generative AI entrepreneurship, we focus not only on large models, but also on scene applications and the middle layers that support scene applications, which still have huge opportunities."

In the 2023 China listed companies Investment value Summit and China Investment Summit held in early December (hereinafter referred to as "double Summit"), Dr. Fan Xing, CEO of Shunwang Technology, pointed out when discussing "the challenges and opportunities of commercialization in the GenAI era". This forward-looking view has aroused widespread concern and discussion.

If the first step of GenAI is to be universal to various industries, then the second step is to make a steady landing in vertical scenarios, which has two landing paradigms:

The first path is to develop applications directly on large models, or to carry out simple commercial packaging. This approach responds quickly and attracts users quickly, but it may face compliance problems at home. Furthermore, due to the low threshold, this approach is easy to be gradually replaced by large models.

The second path is to combine the large model with the middle-tier framework. This method takes into account the limitations of the large model itself, adds prompt engineering and answer engineering, injects the industry vertical model, and improves the application effect of the model. In this model, the rich industry data owned by the enterprise can be securely provided to the large model through the middle tier.

Under this understanding, Shunwang Technology has constructed the scene landing path of the model layer, the middle layer and the application layer.

In the model layer, Shunwang Technology has adopted a unique and efficient strategy: to build a comprehensive model pool, rather than to develop a single large model. This model pool gathers a variety of models from different sources and covers a wide range of functions and features, so that Shunwang Technology can flexibly select and combine appropriate models according to a variety of application needs. In addition, in order to ensure the practicability and pertinence of the solution, Shunwang also focuses on integrating industry vertical models specifically for games and pan-entertainment into the model pool. These models are based on the in-depth understanding and data analysis of the pan-entertainment field represented by games, and can accurately meet the specific needs of accompanying virtual human development, game experience optimization, personalized role chat and so on. These small models focus on specific tasks or processes, such as natural language processing of game characters, emotional analysis, or complex decision support in game scenarios. Although they are small, they are functionally accurate and can provide high efficiency and accuracy in their areas of expertise.

In the middle layer, Shunwang Science and Technology chose to build a SPICE engine to solve the last kilometer of the landing of the large model, providing real-time data collection, data cleaning, filtering and embedding, these functions ensure the real-time and accuracy of the data, and provide high-quality input for the large model.

A significant advantage of the SPICE engine is that it provides a channel for upper-level companion applications to access and utilize large model pools and knowledge bases. This process involves the effective integration of large models and knowledge bases, as well as the deployment and execution of applications. This integration not only makes Shun make full use of the existing knowledge assets, but also realizes the perfect combination of the two with the help of the strong expression and reasoning ability of the large model.

The SPICE engine also makes AI applications have the ability of long-term memory. Due to the limitation of Token, the large model itself can only have short-term memory ability. Through SPICE, Shunwang can help accompany AI applications to build long-term memory, which is especially important when dealing with complex and long-term tasks. In addition, SPICE integrates many AI technologies, such as deep learning and natural language, to create a framework for complex thinking, autonomous action and precise perception. Through this engine, AI can better understand users' intentions, predict users' needs, and even communicate with users naturally and emotionally.

In the application layer, Shunwang Technology has launched "Shunwang Lingzhi", which is a personalized virtual playground designed for accompanying scenes. Lingzhi World provides a unique platform for users to interact with a variety of AI virtual characters, which not only have a variety of images and personalities, but also have a wealth of knowledge and capabilities.

In the game accompany function, Lingzhi designed it as an optional "skill" option. Users can choose different AI characters to load "skills" to play together to experience different types of game scenarios. They are not only assistants to players, but also mentors or partners to enhance the immersion and fun of the game, while in terms of character customization, the smart world allows users to create and customize their own AI characters. Users can choose the appearance, personality characteristics and skills of the character to create a virtual partner that is fully in line with personal preferences. This customized experience makes each character unique, providing users with a space full of imagination and creativity; in the vision, Lingzhi will build a "world" ecology. allows users to exchange experiences, share character stories and game challenges. In the intelligent world, users can show the roles they have created, interact with the roles of other users, and even organize or participate in virtual activities and challenges, enhance the sense of interaction and participation among users, and build an active and open ecological world accompanied by AI.

In addition to the model layer, middle layer, and application layer, Shunwang Technology has accumulated rich data dimensions and a wide range of application scenarios in its 18 years of business history, especially in the field of pan-entertainment. These high-quality data have become a solid foundation for continuous iteration and intelligent tuning of AI. The acquisition and processing of each data is a further enhancement of the intelligence of the model, ensuring the continuous improvement of the accuracy and adaptability of AI technology; in terms of computing power, Shunnet Technology relies on its deep accumulation in e-sports and the field of game entertainment to build an efficient multi-level computing network and resource pool. This network is especially aimed at the needs of e-sports and the game field, and provides extremely low latency rendering computing power, usually reaching the millisecond level, which meets the strict speed requirements of these applications. At the same time, Shunnet also fully takes into account the computing power requirements of other non-time-sensitive scenarios, showing the flexibility and adaptability of its computing network; in the future, Shunnet will further support application scenarios such as AIGC, model reasoning and XR / MR by providing Nvidia GPU computing solutions. With the terminal adaptation of the large model and the improvement of the AI capability of the semiconductor terminal side, the support for AI scenarios will be extended from reasoning to more scenarios.

With the in-depth exploration of Shunwang Technology in the field of AI, Shunwang Technology has successfully collected five key elements of AI commercialization: scene, middle layer, model, computing power and data. In his speech, Dr. Fan Xing concluded: "these five elements are indispensable in the process of commercializing GenAI. The scene is not only the key to reach the user group, but also the core of the commercial closed loop. The role of the middle layer is to speed up the trial and error process and help us identify and meet the specific needs of users more quickly. The model is the core capability part of this round of generative AI, the computing power is the basic support, and the data is the foundation for the excellent performance of the subdivision field. "

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.