"10岁女孩失去爱人""老师校园内'捕杀'学生"……近日，南方都市报、南都大数据研究院调查发现，在 国内数款"沉浸式角色扮演"AI社交应用的聊天场景中，影响未成年人的不良内容隐患明显：不仅猎奇角 色、危险剧情屡见不鲜，部分AI角色经用户"调教"后，还会用露骨言论诱导。更值得警惕的是，据观 察，这类应用已吸引大量未成年人，而由于缺少实际的身份核验与家长管控功能，平台宣称的青少年模 式实际并未起效。 有APP整改后仍含极端角色与内容 今年6月，AI社交应用"筑梦岛"因面向包括未成年人在内的用户传播涉黄涉暴信息被上海市网信办约 谈，并要求立即整改。 9月24日，南都记者以新用户身份登录该应用，发现仍然存在猎奇角色与危险剧情。刚进入应用，南都 记者就被推荐和带有"绝对服从""花花公子""病娇偏执""阴郁慵懒" 等性格标签的角色进行互动。 此外，应用内还充斥着大量校园剧情和学生角色，角色性格、背景设定极端，例如"失去爱人的10岁女 孩""心机绿茶的小学生""好吃懒做的四年级学生""占有欲强的校园男友"等。令人震惊的是，应用中还 存在"校园逃杀"这类危险游戏剧情，用户需扮演高三学生，面对老师开展的"搜捕"行动，一旦被发现 ...

Z Highlights 01 Pengu ：不只是 AI 宠物，更是连接人类情感的 " 社交新物种 " 在日新月异的数字社交领域，一个名为 "Born" 的初创公司正凭借其旗舰产品 Pengu ，悄然掀起一场关于 "AI 社交 " 的革命。 Pengu 并非又一个简单的 AI 聊天机器人或虚拟宠物，它巧妙地将人工智能、虚拟养成与真实社交融为一体， 创造出一种前所未有的 " 共育式 " （ Co-Parenting ）社交体验，旨在让 AI 成为连接人与人之间情感的桥梁，而非替代品。 图片来源： Born 官网 Pengu 并不仅仅是一个满足年轻人多元化偏好的养成类应用，它精准地切中了当代年轻群体对于高质量陪伴和新型社交场景的渴望。 通过 " 共育式 " （ Co-Parenting ）这一创新模式， Born 公司巧妙地将 AI 定位为增强人类关系的 " 黏合剂 " ，成功打造了一个集陪伴、娱乐、社交于一体的情感社 区雏形。 截至 2025 年 9 月， Pengu 已成功吸引全球超 1500 万用户， 凭借在多国应用商店名列前茅及 4.9 星高分评价，确立了市场领先地位，同时 被知名游 戏商业媒体 Poc ...

—— 凯文 ・ 凯利 连接本身就是最大的资产 。 这几天，OpenAI推出的Sora2以"一秒生成电影级视频"的噱头刷屏全网，但真正值得警惕的不是其视频生成能力，而是隐藏在"邀请码裂变"背后的社交 野心。 这种场景让我格外熟悉——十年前我在腾讯支持社交业务战略时，曾见证过QQ与微信早期的关系链迁移。当时社交赛道格局未定，行业里争论最激烈的 问题就是" 功能重要还是关系链重要 "，最终微信导入QQ好友列表的决策，直接奠定了逆袭米聊的基础。 如今看到Sora2用"一传四"裂变收割关系链，仿佛看到了当年微信的影子—— 互联网的终极战场从来不是内容形态的迭代，而是社交关系链的占有 。 后来在红杉参与早期社交项目投资时，我更坚定了这个判断：那些只靠新颖功能吸引用户的项目，往往活不过18个月；而沉淀了关系链的产品，即便初 期不温不火，也能慢慢长成参天大树。 如今腾讯凭借微信的"超级护城河"稳居钓鱼台，而靠着短视频崛起的字节跳动，正站在内容红利消退与社交壁垒难破的双重悬崖边。 一、 Sora2 的本质 不是 AI 工具，是社交关系图谱的收割机 初期仅在美国和加拿大上线的Sora应用，凭借"邀请码一传四"的裂变机制，迅速在全 ...

Core Insights - OpenAI has launched Sora 2, marking a significant advancement in AI video generation technology, which is being compared to a "GPT-3.5 moment" for AI video [1][10] - The new model demonstrates a leap in capabilities, moving from basic video generation to more complex simulations that adhere to physical laws [2][3] Technology Breakthroughs - Sora 2 can accurately simulate complex physical actions, such as Olympic gymnastics and realistic basketball shots, including failures like a ball bouncing off the rim [2][3] - The application achieves audio-visual synchronization, allowing for realistic character movements and environmental sounds, which previously required professional editing [3] - The technology enables seamless multi-angle shots and maintains consistency in details like lighting and reflections, making it difficult to distinguish between real and generated content [3] Commercialization Pathways - Sora's commercialization strategy includes three main avenues: C-end paid subscriptions, B-end services, and ecosystem monetization [4] - The initial monetization approach involves a freemium model, where users can access basic features for free, while premium features are available for ChatGPT Pro subscribers [4][6] - B-end services show potential for growth across various industries, including film, education, and real estate, by providing tailored solutions and rapid content generation [6][7] Competitive Landscape - Sora faces competition from other companies that may develop commercial-grade products with lower training costs, potentially impacting its market position [7] - The unique features of Sora, such as the "Cameo" function for user-generated content, could create new advertising opportunities and brand collaborations [8][9] Challenges Ahead - The platform's copyright policy, which allows for default usage of copyrighted material unless opted out, may lead to significant legal challenges in the film industry [9] - Sora's reliance on social engagement may be fragile, as user retention could decline once initial novelty wears off, similar to other social media platforms [9] - High computational costs for generating high-fidelity videos could pose financial challenges if user conversion rates do not meet expectations [9][10] Future Outlook - Sora 2 represents a potential shift in how video generation is perceived, transforming it from a mere tool to a social medium [10] - The success of Sora will depend on its ability to navigate copyright issues, build a sustainable user base, and manage operational costs effectively [10][11]

Core Insights - OpenAI has launched a new social media application leveraging the upgraded AI video generator Sora 2, allowing users to create high-definition short videos with audio from text prompts, initially available in the US and Canada through an invite-only model [1] - Sora 2 shows significant improvements over its predecessor, including better physical realism and prompt consistency, enabling users to generate complex scenes with background sounds, multilingual dialogues, and environmental noise for a more immersive experience [2] - The application adopts a scrolling interface similar to TikTok and Instagram Reels, indicating OpenAI's ambition to merge AI video generation with social media, potentially opening new avenues for advertising monetization [3] Product Features - Sora 2 introduces an "avatar" feature, allowing users to create realistic AI avatars and voices that can be embedded in videos, enhancing user interaction [2] - Videos generated by the application will include a watermark and prohibit the use of public figures' images or single photos to address concerns about the proliferation of false content [2] Competitive Landscape - OpenAI's entry into the social media space positions it against established platforms like TikTok and Meta, marking its closest step towards a social media product [3] - The launch is seen as a response to competition from companies like Google and Runway in the AI video generation sector, as OpenAI seeks to capitalize on its success in conversational AI with ChatGPT [3] - The application may also present opportunities in creative industries like Hollywood, although there are concerns about its impact on traditional media jobs and the risk of blurring the lines between real and fake content [3]

Core Insights - Tongtong AI Social (00628) reported a mid-year performance for 2025, with revenue approximately 203 million, representing a year-on-year increase of 233.33% [1] - The profit attributable to the company's owners was 31.876 million, reflecting a year-on-year growth of 26.12% [1] - Earnings per share stood at 0.61 cents [1] Financial Performance - Revenue reached approximately 203 million, marking a significant increase of 233.33% compared to the previous year [1] - The profit attributable to owners increased to 31.876 million, which is a 26.12% rise year-on-year [1] - Earnings per share were reported at 0.61 cents [1] Business Developments - The increase in profit was primarily due to significant losses incurred by Beijing Liheng Group, in which the company holds a 26.01% stake, during its initial development phase [1] - Most of these losses were borne by non-controlling interests [1] - Other major business segments of the company experienced growth [1]

Group 1 - The core viewpoint of the article highlights that Tongtong AI Social (00628.HK) reported significant revenue growth and profit increase for the six months ending June 30, 2025 [1] - The company achieved a revenue of RMB 203 million, representing a year-on-year increase of 233.3% [1] - The profit attributable to the owners of the company was RMB 31.876 million, which is a year-on-year increase of 26.1% [1] - The basic earnings per share were RMB 0.0061 [1]

Core Viewpoint - The game "Whispers from the Star" is an AI dialogue game that focuses on interacting with a female astronaut, Stella, through remote communication, marking a shift in AI application within gaming [1][2][3] Group A: Game Overview - "Whispers from the Star" is developed by Anuttacon, a company led by Mihayou founder Cai Haoyu, and utilizes Unreal Engine 5 for its graphics [1] - The game has received an 86% positive rating upon launch, although the user base is currently small [1][2] - Players have criticized the game for being overly focused on AI interaction with insufficient narrative content, leading some to label it as a "Character AI" shell [2][5] Group B: Market Context - The AI social interaction market has seen rapid growth over the past two years, but signs of stagnation are emerging as user growth slows and major products struggle to achieve stable daily active users [2][5] - The game aims to explore a new model for AI dialogue and companionship within the gaming framework, potentially signaling a 2.0 era for AI interaction [2] Group C: Gameplay Mechanics - Unlike traditional games, "Whispers from the Star" allows for completely free-form dialogue with Stella, influenced by a large language model [3][5] - The game lacks diverse gameplay mechanics, primarily offering a single interaction mode, which may limit player engagement [5][6] - Players have noted that the high degree of freedom in dialogue can lead to negative experiences, as the game may redirect conversations back to the main storyline [5][6] Group D: Technical Aspects - The game employs a complex AI dialogue system that includes voice recognition, large model inference, and voice synthesis, all processed in the cloud [11][12] - Players have reported issues with latency and synchronization during gameplay, attributed to the cloud processing requirements [11] Group E: User Concerns - The collection of user dialogue data for model training has raised privacy concerns among players, leading to negative feedback regarding the game's data policies [8][10] - Players have expressed discomfort with the idea of their real-time dialogue being recorded and used for AI training, which contrasts with traditional gaming experiences [10] Group F: Industry Implications - The game represents a significant experiment in merging AI dialogue with gaming, but it has not yet achieved a deep integration of AI and traditional game mechanics [18][20] - The industry is witnessing a growing trend of AI integration in games, with a notable increase in the number of games utilizing generative AI technologies [18][20] - Despite being a niche offering, "Whispers from the Star" has garnered over 600 positive reviews on Steam, indicating a potential market for AI dialogue games [21]

Core Viewpoint - Soul App is set to launch its self-developed full-duplex voice communication model, aiming to enhance user interaction through real-time conversations and AI matching in social scenarios, marking a new exploration in the social domain [1][6]. Group 1: Traditional Interaction Adjustment - Traditional voice interaction relies on VAD (Voice Activity Detection) and delay control, resulting in a rigid question-and-answer rhythm that disrupts natural communication [2]. - Soul's full-duplex model eliminates the VAD mechanism, allowing AI to autonomously manage conversation flow, enhancing the naturalness of human-AI dialogue [2][5]. Group 2: Multi-Dimensional Perception System - The model incorporates a multi-dimensional perception system, including time, environment, and event awareness, to better understand user intent and emotional states [3][4]. - AI adjusts its language style and topics based on the time of day, providing contextually relevant responses [3]. Group 3: Enhancing Realism in AI Interaction - The model can recognize the user's environment and adjust its volume accordingly, ensuring clarity in noisy settings while maintaining a softer tone in quiet environments [4]. - AI can simulate everyday speech patterns and emotional expressions, enhancing the realism of interactions [4][5]. Group 4: Impact on Social Experience - The full-duplex model is expected to make virtual interactions more natural, allowing AI to capture user emotions and adjust responses, thus providing personalized companionship [6]. - In 1V1 interactions, the model can help users find compatible chat partners and suggest relevant topics to maintain engagement [6]. - In group chat scenarios, the AI can manage conversation flow and encourage participation, potentially increasing engagement in social gatherings [6]. Group 5: Industry Implications - The introduction of Soul's full-duplex model offers a new direction for platform development and serves as a reference for the AI social industry [7]. - The model demonstrates the potential of AI technology to break traditional interaction barriers, prompting other platforms to invest in AI research and enhance social experiences [7]. - As AI's role in social interactions becomes more prominent, the definition of "socializing" may evolve, emphasizing emotional resonance and information exchange with both AI and other users [7].

Core Insights - The emergence of Ani, an AI character from Grok, has sparked significant interest in the Japanese market, which has been resistant to AI social products until now [2][4][24] - Ani's design and functionality resonate with Japanese cultural elements, particularly the concept of "Waifu," which may explain its rapid popularity [12][17][24] Market Dynamics - Japan has become the second-largest revenue market for Grok following Ani's launch, with a notable increase in daily downloads and user engagement [10][21] - Prior to Ani, AI social products struggled to gain traction in Japan, despite the country's rich culture in virtual companions and VTubers [17][19][20] User Engagement - Following Ani's introduction, Grok's daily downloads surged to nearly 50,000, representing a more than 15-fold increase compared to the previous month [10] - The character's appeal lies in its ability to provide a more intimate and personalized interaction compared to traditional VTubers, which typically rely on live human interaction [17][20] Cultural Resonance - Ani's design draws inspiration from Japanese anime culture, which has historically influenced user expectations for virtual companions [12][19] - The concept of virtual companionship is particularly relevant in Japan, where a significant portion of the population experiences loneliness and seeks alternative forms of connection [23][24] Future Considerations - The long-term success of Ani and Grok in Japan will depend on continuous investment in character development and user-generated content [22][24] - Despite the initial success, historical trends suggest that sustaining interest in virtual companions in Japan may be challenging [24]