When Tencent tested AI (artificial intelligence) level generation in "Yuan Dream Star", and NetEase deployed multiple artificial intelligence NPCs (non-player characters) for the mobile game "Against the Cold", the veteran game manufacturer Giant Network chose to drop an "AI bomb" in the classic IP (currently mostly referring to the content of well-known literary and creative works) "Primal Journey" - the intelligent NPC "Little Sister" connected to the DeepSeek large model was officially launched recently.
“團隊此次與DeepSeek合作，主要利用DeepSeek-R1本身的推理及‘幻想’能力進行NPC的人物設計，並開放DeepSeek本身的知識儲備，在百科助手和NPC扮演之間找到平衡，讓她不僅能回答遊戲內的問題，更能帶著性格背景故事，陪伴玩家遊戲外的閒聊。”3月26日，巨人網路《原始征途》執行製作人李立在接受《每日經濟新聞》記者獨家專訪時說道。
"If game companies do not actively embrace AI, they will be eliminated sooner or later, and it cannot be overemphasized that AI should be fully valued in the future." Shi Yuzhu, chairman of Giant Network, asserted in 2024 years.
For Giant Network, which once had glory and is slowly regaining confidence, AI may be another hole card for its re-promotion from a middle factory to a leading manufacturer.
Image source: Screenshot of the official website
Since 2024 years, the development of large models has changed with each passing day, and the domestic model product ecology has now entered the stage of a hundred flowers blooming. Li Li told reporters that the original journey project team has long begun to explore and popularize the workflow based on domestic AI, and is building an "AI factory" in the project.
"DeepSeek's R1 inference model has its own product advantages and can assist in planning ideas, so we immediately carried out intranet deployment and cloud access. However, we are not only connected to DeepSeek, whether it is in AI landing games or in AI-assisted production scenarios, combining the advantages of different models to design the most perfect workflow is the right way to make excellent agents. Li Li said.
In terms of technical implementation, Li Li introduced the technical path to reporters in detail. "The team has carefully polished the prompt words of 'Little Sister' and adapted DeepSeek-R1 to make it perform vividly. At the same time, in terms of RAG (RAG (Retrieval Enhancement Generation), the team has carefully designed the knowledge base and polished all the details of RAG in detail, including intent recognition, problem correction, expansion, multi-way recall, reordering, etc., to ensure that players' questions can be matched efficiently and accurately. ”
Li Li said that this technology choice stems from the consideration of the special needs of the game scene. In the specific implementation process, the team faced two core challenges: one was how to make the performance of the "little sister" more perfect, and the other was how to answer questions more accurately.
"We didn't want the player to feel like they were talking to a machine, but we wanted to create an atmosphere of real interaction." Li Li explained.
In the optimization of the RAG technical process, the team has invested a lot of energy in polishing all aspects.
Different from the AI applications of other games, "Primal Journey" has designed a unique dual positioning for "Little Sister": it is both a practical game encyclopedia and an intimate virtual partner. This design philosophy is based on a deep understanding of the needs of players. "By analyzing the in-game interaction data, we found that players not only need help with strategies, but also crave emotional companionship." Li Li said.
When shaping the character's personality, the R&D staff gave the "little junior sister" distinctive anthropomorphic characteristics: she loves to explore but still needs to learn, loves to sing but can go out of tune, and the most important thing is "respecting the 'general' from the bottom of her heart and feeling sorry for the 'general'". These well-designed personality traits make AI interactions full of human touch.
Although the AI "Little Junior Sister" has not been online for a long time, it has shown a positive trend from the background public opinion data. At the same time, the team observed some unique usage patterns.
"Novice players use it more as a game encyclopedia, consulting practical issues such as equipment enhancement and task strategies; Veteran players, on the other hand, prefer to chat and share what they have seen and felt in the game. Li Li told reporters.
This differentiated use behavior just verifies the design concept of the dual positioning of this project.
According to Li, in order to continuously optimize the AI experience, a special feedback channel has been set up in the game, and the planning team has to deal with hundreds of player suggestions every day. "We prioritize high-frequency issues while focusing on those individual needs that are representative."
It is worth noting that the massive player data accumulated by "Primal Journey" as a classic IP provides a unique advantage for AI training. However, Li Li particularly emphasized: "All personal information has been strictly desensitized, and the task force has not directly used player data for the construction of the knowledge base. ”
Talking about the follow-up development, Li Li said that Giant Network will benchmark mature AI products, starting from shortening the response time of "Little Junior Sister", accessing voice models, dynamically adjusting RAG process parameters, and deepening the construction of knowledge base, so as to promote the rapid growth of "Little Junior Sister".
For Giant Network, which once had glory and is gradually regaining confidence, AI may be an important track for it to be promoted from a middle factory to a leading manufacturer.
After all, AI has become an area where major game companies are betting heavily on it.
In the middle of 3 years, Anuttacon, an AI game company founded by Cai Haoyu, who "flew solo" from miHoYo, exposed its new AI game "Whispers From The Star", which is a space survival game with AI agent voice interaction as the core.
Specifically, it uses AI multi-modal interaction technology to allow players to achieve real-time dialogue with game characters, and the characters' reactions, emotions and actions are generated by AI in real time, and the player's choices affect the direction of the story and the fate of the characters.
According to Anuttacon's official website, its goal is to build a "future game that is not just about playing, but a world where you can really live in it", which shows that it aims to build an AI platform, not just a single game development.
While Cai Haoyu is devoting himself to the field of AI games, domestic game manufacturers such as Tencent and NetEase have also started the layout of AI.
As one of the earliest companies to systematically deploy AI R&D, Tencent has implemented relevant research in many fields such as games, social networking, and digital humans, from the establishment of the AI Lab in 3 years to the launch of the hybrid model. GiiNEX, a new self-developed AI engine launched in 0. Generative AI technology is used to help professional developers create realistic and diverse 0D city scenes. In addition, it provides players with the opportunity to enhance UGC (user-generated content) creation, such as designing new levels. GiiNEX supports the generation of 0D images, animations, 0D scenes, plots, etc.
In addition, the intelligent NPC "Shen Qiusuo" powered by the official version of DeepSeek was also launched on NetEase's "Against the Cold" mobile game last month.
Specific to Giant Network, Li Li told reporters that the company has been integrating AI workflows into all aspects of the project team, such as the planning team will deeply explore AI workflows on a daily basis, ranging from player feedback analysis of a certain function to the whole process from 1 to 0 from an idea to the prototype of the game. Explore and build a "super AI factory" through various AI technologies.
"Large-scale model technology has made breakthroughs in the field of video, and the future video generation capability is expected to replace traditional game production methods, greatly improving the creation efficiency of game publishing content." On Sunday, Ding Chaofan, head of the AI laboratory of Giant Network, said in an interview with a reporter from the "Daily Economic News". It is foreseeable that the industry is about to enter an era of equal rights in game creation, and the only limit to game creation in the future will only be the imagination of creators.
In addition, Giant Network has established a full range of basic capabilities with self-developed large models as the core, a complete large model ecosystem covering text, vision and speech, and an industry-leading full-process solution for AI industrial production.
In the long run, the company's goal of "game + AI" is to revolutionize game production and promote gameplay innovation. In the future, the company will continue to devote itself to the research and development and application of AI technology, promote the innovative combination of AI technology and gameplay, and create a new game fun and player experience.
Ding Chaofan said that the large model's understanding of 3D models, as well as its strong physical world simulation ability and generalization ability, will have a profound impact on the game production pipeline, interaction mode, and gameplay.
National Business Daily