Soul App: The Gateway to AI-Enhanced Social Networking

In the current digital landscape, the value of content has become increasingly prominent, particularly in its role as a vital resource for training AI models.

One clear example is Reddit. Despite its relatively problematic commercial performance and a history of operating at a loss, Reddit managed to secure a high valuation when it went public. Since its IPO on March 21, the social media platform’s market value soared from the initial $6.5 billion to as much as $10.68 billion. A significant contributor to this valuation is Reddit’s strategic decision to sell its data to Google’s AI chatbot Gemini for use as a training resource, which brings in $60 million per year. 

This successful monetization of data stands in contrast to China’s Zhihu, a platform often compared to Reddit. Despite a similar approach of supplying user-generated content to Kimi – Chinese AI startup Moonshot’s chatbot – Zhihu‘s market performance has not seen a similar boost as Reddit. This difference in outcomes suggests that the integration of content supply and AI training, and the subsequent monetization of this process, is not necessarily a guaranteed success. 

The question that arises, then, is whether there could be a platform that integrates the supply of content and AI training into one single efficient system, while also capitalizing on the value of content in the AI era. One potential contender for this innovative model is Soul App, a social networking app backed by Tencent and MIHOYO, with a robust user base among Chinese Gen-Z. 

In this context, Soul’s approach to integrating its own user data with self-developed large-scale AI model presents an interesting case.

Evolving app-specific models to better serve public domain scenes

While it is easy to get caught up in the race for the latest technology, we’ve seen time and again that a singular focus on technology alone doesn’t guarantee success, and in the case of AI, the true value lies not in the sheer size or complexity of the large model, but in how it can be applied to improve user experiences in specific, meaningful ways. 

Spotify’s music recommendation system, for example, utilizes AI technology to understand both user behavior and the music itself, offering recommendations tailored to individual tastes. The success of features like the “Discover Weekly” playlist, which suggests new songs based on user preferences, underscores the importance of practical AI applications over mere model size.

Following a similar path, as a social networking app centered around shared interests, Soul devotes itself into improving its capacity to deliver precise and significant enhancements to user experiences, to which Soul’s unique focus on “public domain scenes” is a testament. 

These scenes, such as “Soul Square” (similar to Instagram’s “Feed”) and “Audio Partyrooms” (similar to Clubhouse) where all users can interact with each other, allow them to forge diverse relationships, be it one-to-one, one-to-many, or many-to-many connections, transcending gender limitations. 

Hence, as the company integrates AI into its business model, its strategy pivots around the construction of application-specific models, rather than pursuing the development of broadly defined, large-scale ones. 

“When AIGC emerged, our initial decision was not to limit ourselves to just focusing on AGI technology or creating generic large-scale models. I’m convinced that technology will eventually become more egalitarian, and we shouldn’t lose sight of our unique strengths by chasing after the latest trends. We’ve crafted our own approach, focusing on the Mixture of Experts (MoE) model. Our strengths lie in our application scenarios and data”, says Soul App chief executive Zhang Lu in her interview with Chinese media Geekpark.

This approach signifies a conscious shift towards harnessing the power of AI to create a more engaging and meaningful social networking experience:

When looking at Soul’s targeted audience, it is clear that Gen-Z is a nomadic lot that frequently hops between different content types, from gaming to chats and from videos to memes and music. To catch the fancy of Gen-Z, Soul introduced AIGC to its online game “Werewolf Awakening”. The NPC(Non-player Character)s integrated with AI capabilities in this game now can make precise anthropomorphic decisions to disguise, trust, lead, and confront, etc. AIGC empowers multi-role playing and more sound selections, which enhance users’ interactions with virtual characters.

Also, Soul’s “Lazy KTV” offers a multi-person UGC “AI Duet” experience. Users start by recording a 30-second audio clip, which serves as a personalized voice model. This clip is then put through one of the three AI singing modes – balanced, similarity, and expressiveness in preparation for further synthesis. After that, users can further invite friends to complete voice cloning together and perform a musical “duet” across time and space.

“Some people readily adopt the current AIGC technology products, while others hesitate, often because they’re unsure how to engage with them. Therefore, when designing and optimizing these products, it’s crucial to consider a broad spectrum of user needs. This includes making the technology intuitive and easy to grasp, as well as ensuring it remains engaging over time”, says Zhang Lu, adding that it is essential to understand users and adapt strategies to ensure retention.

To better anchor its AI capabilities and safeguard the security, Soul launched a self-developed specialized large model, SoulX, which plays an important role in the platform’s social networking ecosystem. SoulX is equipped with advanced functionalities such as prompt-driven, conditionally controllable generation, context understanding, and multimodal comprehension. This model ensures seamless, natural, emotionally resonant, and dynamic conversations while encompassing a wide array of finely categorized risks. 

The effectiveness of Soul’s strategy is demonstrated by the success of Character.ai, argubly the most popular AI chatbot in the world akin to ChatGPT. In June 2023, it received an average of 28 minutes per visit from its 280 million visits. Such stunning figure can be largely attributed to its innovative user interaction experience, advanced AI technology, and precise market positioning. Moreover, its focus on a young user base that has a natural affinity for anime culture and virtual socialization has allowed it to establish a strong brand in the AI social field. 

Similarly, by offering enhanced scenarios and membership perks, Soul has effectively capitalized on this trend, appealing to the preferences of Chinese youth. By the end of 2022, value-added services such as avatar and virtual gifts, recommendation privileges, and game props accounted for 91.1% of the company’s revenue. These figures suggest that young users in China are not only receptive to but also willing to pay for the virtual experiences offered by Soul.

An AI Playground for Gen-Z, A Self-sustaining Ecosystem

So, what’s the potential value of Soul’s data? 

By the end of 2022, Soul boasted an impressive MAU (Monthly Active User) count of 29.4 million, with a DAU (Daily Active User) count of 9.5 million. This substantial user base translates into a significant data flow for the platform. For instance, if a single user generates around 60 messages daily, the cumulative effect across a DAU of approximately 10 million users results in an astounding 600 million messages collected by Soul on a daily basis.

In comparison, the DAU of the ChatGPT app is also only at the tens of millions level. Thus, having a traffic entry point is key, and Soul is well-positioned to become a significant traffic entry point in the future AI era, where conversation is the primary form of interaction.

Most importantly, with 78.4% of its average monthly active users belonging to Chinese Gen-Z, Soul’s data collection is not only extensive but also highly engaging. This sheer volume of user-generated content provides Soul with a rich and diverse dataset crucial for training its large AI models, enabling the platform to tailor its AI development specifically to the preferences and interactions of its predominantly Chinese Gen-Z user base. 

As Zhang Lu explains, “User feedback is crucial for model improvement. The current model training requires not just massive amounts of data, but a large volume of highly relevant, high-quality data. This directly determines the performance, iteration speed, and training cost of the vertical model.”

The integration of AI features has transformed Soul into a self-sustaining ecosystem, fostering vibrant community engagement. For example, the platform utilises an AI chatbot called “AI Goudan” to generate contextualised comments, which might turn out to be genuine, funny, or quirky, under the posts sent by users in “Soul Square” and spark real interactions among users, breathing life into the platform’s community. 

“A model should not only boost efficiency but also offer emotional and personalized experiences. It should sometimes agree and sometimes disagree with the user. This is fundamental for AI in social scenarios, a direction also pursued by Soul’s conversational AI”, says Zhang Lu.

The back-and-forth “community” interactions create possibilities for further 1-1 “social” interactions, which forms the bedrock of Soul’s commercial success, underlining the pivotal role of AI in nurturing and sustaining its ecosystem for future growth and innovation. 

Such success is also evident in the high user engagement levels on its platform, with 53.9% of daily active users engaging in private messaging, spending an average of 46.4 minutes per day on the platform, and exchanging 66.9 messages daily. Notably, Soul witnesses 10.5 billion one-to-one messages per month, highlighting its robust user interaction and the platform’s impressive user retention rate of 78.6%. 

“In my opinion, humans and agents will coexist in the future. Given the current technological capabilities, agents may not yet provide you with particular surprise, but it won’t be too bad, ensuring a baseline experience. When technological capabilities achieve a breakthrough, in this uphill process, the younger generation growing up with new technology may naturally become accustomed to interacting with agents”, says Zhang Lu.  

Interestingly, according to a social trends report released by Soul in early 2024, alongside friends, hobbies, and family, 9.1% of Chinese youth consider “virtual humans and idols” as a preferred form of companionship in moments of loneliness.

Conclusion

In a landscape where global AI model investments are witnessing a decline, as highlighted by a recent report from Stanford’s Institute for Human-Centered Artificial Intelligence (HAI), the emergence of platforms harnessing the power of AI set to drive innovation stands out to have transformative potentials. 

In the rapidly evolving AI era, for social networking platforms, understanding user behavior and adaptability are key strategic factors. Companies like Soul are leveraging these insights, integrating AI with unique user datasets to enhance interaction and engagement. 

Therefore, the focus is now shifting towards improving user response rates, dialogue rounds, and content publication, all of which can foster a vibrant virtual community. These metrics provide valuable feedback on the efficiency and quality of AI-assisted human interaction, shaping strategic direction in the industry.