"Can you feel that I'm with you?" In the story of the science fiction movie "Her" directed by Spike Jonze and released in 2014, the artificial intelligence operating system (AI OS) "Samantha", which has only a voice but no body, says this to the male protagonist. As a movie about the love between humans and artificial intelligence, the content seems to predict the current big companies’ bets on AI social networking. Since the advent of ChatGPT, model capabilities represented by emotion perception have rapidly climbed in the era of large models, and the emotional and emotional value that AI can provide after humanization has begun to be explored. Baidu, Douyin, Tencent, Meituan and other leading companies are using AI to return to the social track, and the entry point is surprisingly the same, that is, the interaction between real people and virtual artificial intelligence. Based on the fact that some large models will officially open services to the public on August 31, social networking seems to have become the second curve for leading manufacturers to develop for the C-end after tool applications. Judging from Baidu's "Wanhua", Douyin's "Xinqing", Meituan's "WOW" and other apps that have been tested online, AI social networking is currently the mainstream method of sensing user emotions and providing emotional value through companion chat. Among them, Kuaishou’s “AI Xiaokuai” also provides tool functions that are already familiar to us, such as drawings and copywriting modifications. In other words, the products launched by each company are highly consistent in design and function. This not only easily moves competition down to the level of model capabilities that are difficult to perceive, but may also lead to the dilemma of "born to attract new users and die to retain" in previous social products. At present, it seems that each player's understanding of the product is relatively "unified" and has not yet entered the mature stage of differentiation. Moreover, compared with the diverse gameplay of overseas AI social networking, the centralized launch of testing seems to be a bit rushed. AIGC, which claims to reshape thousands of industries, goes to social media. What kind of story can it tell? Let’s mature AI social networking together According to public information and Photon Planet experience, the current AI social products are still in the collective testing stage, and there are subtle differences in play styles. Players with social genes tend to build AI social networking into existing applications, which is equivalent to AI reconstruction of applications, such as Tencent Music’s “Weiban”, Douyin’s “Xinqing” and Weibo’s “Celebrity AI Emotional Companion Chat” "wait. Baidu and Meituan, which have not established a social presence in the mobile Internet era, have launched independent apps. Among them, Baidu has launched four apps to form a matrix. The former relies on existing products to facilitate the accumulation of users' existing relationship chains. The latter started anew, and although it lost the opportunity, it preferred AI-native. It's hard to say which of the two is better at the moment. Specific to the product level, the differences between players have begun to narrow, and they are basically making a fuss around the segmentation functionality of AI virtual characters. Douyin Xinqing is positioned as an emotional care robot, taking the form of the private messaging function of Douyin accounts. Before its official use, users are required to confirm the disclaimer agreement that "it cannot replace psychological counseling". It has the function of comforting emotions but it is also limited. The function it provides users is similar to the tree hole in the early days of the mobile Internet. At present, there is no possibility of linking real-person social networking and commercialization. Zhudengdao, a subsidiary of China Literature, is building an AI-native UGC community based on AI to mainly meet users’ entertainment needs. Users can enter settings to create virtual characters. On this basis, it also launched a "small theater" gameplay where users can create character stories and settings together. As an earlier application, the content of Dream Island is relatively complete, and its commercialization has gone further. It will give users 400 starlight points every day, which can support the AI ??character to reply 400 times. After that, you must pay to obtain starlight points to continue the conversation. In addition, paid content also includes the character’s memory focus ability improvement and chat bubbles, etc. There is only a limit of 400 replies, which is difficult for non-heavy users to reach, and chat bubble payment is also somewhat routine, which shows that its commercialization attempts are prudent. In comparison, Meituan’s WOW is relatively immature. After our experience, we found that WOW supports voice chat and users can create characters, but the voice is relatively lacking in human emotions, and there are few preset columns, which makes UGC characters somewhat "unstable." More importantly, the basic "recommendation-search-experience" link of its UGC platform is not perfect. Users who want to talk to new characters can only try their luck by swiping left and right in the discovery bar. Commercialization is currently out of the question. From the above three applications, we can easily see that although the curtain of AI social networking has been opened, no matter which company's product it is, it seems that it is still lagging behind. As a product based on large models, AI social applications are bound to have limited capabilities. For example, Dream Island relies on a corpus of mainly copyrighted novels from China Literature, and is several positions ahead in terms of character realism and details, and may even turn the data flywheel. Other manufacturers may be slow at one step and slower at the next, so they need to get their products out as soon as possible and obtain user data. This is not only the reason why major manufacturers have been swarming into the AI ??social track in recent times, but also evidence that their products need to be improved and are suspected of being rushed. Under such circumstances, it is understandable that the pace of commercialization is slightly slower. Considering the payment context of domestic C-side, it is unlikely that AI applications before scale can find a smooth commercialization path. Even after the product is gradually improved, AI social networking may not be enough to generate sufficient commercialization space. Take Character.AI, the most downloaded AI social product in the world in the first half of the year, as an example. As one of the most successful overseas C-side applications, its functions include agent-oriented assistants, psychologists, celebrities, virtual characters, and UGC customization. AI characters, AI group chat mode, and original UGC community with evaluation and scoring functions are also launched in terms of gameplay. Judging from its functions, we can easily identify traces of domestic AI social imitation. However, Character.AI is so popular that its subscription-based commercialization path is not smooth either. This unicorn has not disclosed revenue so far. Instead, it has raised multiple rounds of financing in order to expand its scale. There are also reports that it intends to consider developing B-side business after expanding the size of its team. Zhang Xiaolong, the father of WeChat, once posted a news on Fanfou Diary: "Every company publishing a Weibo is an act that makes users schizophrenic." It is foreseeable that AI social networking is likely to reproduce the WeChat trend of the mobile Internet 10 years ago. Bo battle. Even considering that the track is becoming increasingly crowded and the time window is tight, it is not enough to just use "usability" as a product standard.