On January 19, Giant Network recently launched a new "AI Large Model Challenge" in its hit mobile game *Supernatural Action Team*, marking the first time AI large model capabilities have been integrated into the core gameplay layer of a high-DAU game and made available to its entire user base. Supported by the AI capabilities and ecosystem collaboration of leading technology platforms such as Alibaba Cloud, Volcano Engine, and Tencent Cloud, this feature has achieved stable operation and large-scale implementation of AI large models in games with highly active user populations, and is regarded by the industry as a landmark case of native AI embedding and validation within a major high-DAU product in China. In this new gameplay mode, in-game NPC "dummies" are driven in real-time by AI large models, transforming them from characters following fixed scripts into direct participants in matches as "player opponents." These NPCs can engage in voice communication, mimic real player behavior to blend into teams, and proactively initiate attacks at critical moments to confront human players head-on, making it difficult for players to distinguish between "who is real and who is AI" during a match. To date, AI has participated in over 25 million matches, becoming one of the most discussed new features among players. To promote the large-scale implementation of native AI gameplay in high-DAU games, Giant Network, in this initiative, collaborated with several leading technology platforms to explore key challenges such as large model capability adaptation, high-concurrency real-time inference, and stable operation in complex interactive scenarios, collectively advancing the application of AI in real-game environments. This has also made *Supernatural Action Team* one of the few cases in the industry to achieve large-scale validation of native AI gameplay within a major high-DAU product. From a gameplay perspective, unlike traditional scripted NPCs or conversational AI assistants, the "AI dummies" in *Supernatural Action Team* possess stronger autonomous decision-making capabilities. Through the synergy of capabilities like Automatic Speech Recognition (ASR), Large Language Models (LLM), and Text-to-Speech (TTS), the "AI dummies" can understand player semantics, imitate the timbre and linguistic style of real players, and generate counter-strategies based on the real-time match environment. They create confusion by lingering in disguise, following and approaching, or misleading the way, and launch attacks on human players at opportune moments, infusing each match with unpredictability. This design significantly enhances the game's randomness and strategic experience, quickly igniting passionate discussion among players. On social media platforms, short videos and topics centered on content like "AI's deceptive realism," "dancing with AI dummies," and "getting backstabbed by AI" have continued to gain traction. Many players have expressed that this is the first time they have clearly felt AI having a substantive impact on the gameplay experience in a high-DAU game. It is understood that the core objective of introducing AI into the game is not to replace players or increase difficulty, but rather to use AI participation to make the gaming process more immersive and entertaining, allowing familiar gameplay rules to continuously generate new variations and experiences. From an industry perspective, although attention on AI in the gaming sector has been steadily increasing in recent years, most applications remain concentrated in auxiliary systems or localized feature testing. Deeply embedding AI large models into core gameplay and rolling them out to the entire user base of a high-DAU product is still rare. This update for *Supernatural Action Team* is the first to fully demonstrate the real-world operational state of AI large models integrated into the core gameplay layer of a high-DAU game, validating their stability and feasibility in high-concurrency, strongly interactive environments, and providing a tangible, observable case study for "native AI gameplay." This type of high-frequency interactive, sustainably evolving native AI gameplay is opening up new content formats and long-term commercial possibilities for the gaming industry. As an annual dark horse hit, *Supernatural Action Team* has maintained its popularity since launch through its differentiated gameplay combining "Chinese-style mild horror with multiplayer cooperation," consistently ranking at the top of the App Store free and top-grossing charts and driving the expansion of the mild horror genre. While maintaining its leadership in the genre, the game has taken the lead in introducing AI into the core gameplay layer, further expanding the boundaries of the gaming experience. Prior to this, Giant Network had been continuously exploring native AI gameplay in its social deduction game *Space Kill*, successively launching modes such as "AI Deduction Theater," "AI Endgame Challenge," and "AI Endgame Duel." These related gameplay features have cumulatively attracted millions of players and generated tens of millions of matches, accumulating practical experience for the large-scale application of AI in games. At the corporate level, Giant Network has consistently increased its investment in AI in recent years, promoting the deep integration of AI with its core gaming business through self-research, investment, and industrial collaboration. It is widely believed within the industry that the implementation of the AI large model gameplay in *Supernatural Action Team* signifies a shift in the role of AI in games from "tool-layer application" to "gameplay and content driver." Moving forward, Giant Network will continue to collaborate with partners to explore more scalable and sustainable innovative paths centered around native AI gameplay.
Comments