Tech

Unveiling the Rise of AI Social Apps: Will AIGC Spark the Next Big Thing?

After rolling out large models for a year in 2023, the general consensus in the industry is that 2024 will be a wave year for AI large model applications.

Although the underlying model technology is still climbing and the application layer is generally in its early stages, the industry has begun to look forward to the emergence of Killer App. Especially the emergence of Sora at the beginning of the year has brought about a big discussion about vertical models and implementation in specific industries.

Judging from the distribution of existing AIGC applications, manufacturers are betting more on the two tracks of AIGC tools and AI social networking.

The former is currently performing well in terms of attention and downloads. For example, smart painting and smart writing tools including Midjourney, Quillbot, etc. have greatly changed productivity and at the same time, the number of visits continues to hit new highs.

The latter has the inherent advantages of having many users, strong minds, and the ability to produce entry-level products. Previously, after ByteDance launched its flagship product Douyin to accompany the AI ​​”Xinqing”, it launched “Chaotuo” to continue in-depth social exploration. Even Meituan also launched an independent AI chat product “WOW” to test the social business . Not to mention, platforms such as Soul App that focus on the social track have already launched self-developed vertical language models and a series of virtual companionship and social ice-breaking functions, from AI recommendation relationships to AI-assisted conversations, from lowering the expression threshold to improving Interactive experience further accelerates the implementation of large models in the social application layer.

Under the wave of AIGC applications in 2024, which field will be the first to give birth to Killer App and become a new traffic entrance? The answer may be, social interaction.

Why is it said that AIGC’s Killer App was first born in social media?

From QQ to WeChat, from the PC era to the mobile era, social areas with large user base and high stickiness have always been the focus of Killer Apps. It is not difficult to understand why many voices believe that in the AI ​​era, social networking is still more likely to outrun the Killer App.

According to the “2023 AIGC Mobile Market Insights” released by the third-party data platform data.ai, it is estimated that the number of AIGC application downloads will break a record and reach a new high in December 2023, exceeding 90 million people. Among them, the growth of users of artificial intelligence chatbot applications supported by AI technology is particularly significant, contributing 72% of the total downloads in 2023.

In fact, since ChatGPT detonated the AIGC market, tool applications have taken the lead due to factors such as strong practicality. Whether it is foreign OpenAI, Microsoft, or domestic Baidu, Alibaba, these leading manufacturers have successively launched AIGC tools and launched application exploration in office, marketing, education and other scenarios.

It is undeniable that the AIGC tool is a typical representative of technological innovation productivity, but it has technical cost constraints, and the road to commercialization is dancing with shackles. In the future, it may become an embarrassing position of a plug-in. It was not the birth of Killer App. Best track.

AIGC tools are essentially products that can “make miracles with great force” by burning money and constantly climbing to technical heights. Whether it is painting, text or video, the core competitiveness of AIGC tools lies in building a technological “lone peak advantage” from the supply of massive high-quality data, the improvement of chip computing power and breakthroughs in the direction of large language models. Only in this way can it be qualified to claim As a tool rather than a “toy”.

From the perspective of overall cash burn, IDC has a set of forecast data: global enterprises will invest US$16 billion in generative artificial intelligence solutions in 2023, and by 2027, generative artificial intelligence spending is expected to exceed US$140 billion.

Specific to manufacturers, even Open AI, its CEO Sam Altman has repeatedly complained that the shortage of computing power has hindered ChatGPT from launching a larger “context window.” Other products that are not as “rich” as ChatGPT, such as Lensa AI, which has dominated the APP stores in many countries, quickly became popular because it can generate AI avatars of various styles. However, because the technology cannot keep up with the new needs of users, it cannot escape the “monthly trend”. Throw” the fate of the AI ​​“toy”.

At the same time, the technical cost of AIGC tools does not decrease at the margin, but rises exponentially with the increase in user scale.

Stacy Rasgon, an analyst at Bernstein, a well-known Wall Street investment institution, estimates that the cost of a single query of ChatGPT is about 4 cents. If ChatGPT queries grew to one-tenth the size of Google search, it would require approximately $48.1 billion worth of GPUs and $16 billion in chips per year to keep running.

Of course, the high-cost AIGC tools are not “generating electricity with love”. Under the heavy pressure of technology costs, AIGC tools have already begun to explore the path of commercialization, but it is still difficult to establish its own new business order.

ChatGPT, which has exceeded 100 million daily active users in just two months since its launch, has continued to gain the admiration of ordinary users. However, when it comes to payment, Bloomberg News stated that ChatGPT Enterprise Edition currently only has 150,000 subscribers. But in 2022, OpenAI’s losses will be approximately US$540 million.

There are countless such cases.

It can be said that only by relying on money to “make miracles” can AIGC tools build up technical barriers, drive ultimate product power, upgrade from entertainment “toys” to productivity tools, and run through the professional market. It is difficult to become an entry-level tool independently. product. Moreover, this game that consumes money, time and manpower is destined to only belong to a few powerful AI manufacturers.

This is why after the big model “After the Hundred Regiments War”, the focus of industry discussions quickly turned to the application layer. With large-scale training costs “burning”, players need to quickly find clear scenarios, delineate stable user groups, and find commercialization directions.

It can have a huge user ecosystem, firmly occupy the national usage time, have high stickiness, mature business model, and is closest to monetization… The social track is naturally one of the best choices.

When looking abroad, the trend of integrating large AI models with social products is even more obvious. For example, Musk, who is recognized in the industry as “always playing with new things,” announced last year that he would integrate his artificial intelligence startup xAI with the social media platform X (Twitter).

Just as WeChat won the ticket to the mobile era for a generation of technology empire Tencent, now standing on the eve of the new Internet era opened by AIGC, social networking has once again become the entrance that is competing for:

With the launch of Wen Xin Yi Yan, Baidu, which has the potential to dominate the domestic AI field, has successively launched social AI applications such as “Wanhua”, “Xiaokan Planet” and “SynClub”; in addition to Meituan, Tencent Reading, which focuses on reading, has also Launched “Building Dream Island” to provide AI interaction; by winning over young users to gain a foothold in the highly competitive social track, Soul launched a self-developed large language model and a combination of AI capabilities along the direction of “AIGC + Social” New features such as “AI Goudan” and AI chat assistant…

Players are entering the game one after another, who can be the winner?

The gateway to AI social competition

Different from AIGC tools, the core competition focus of AI social is not the large model technology base, but the need to integrate AI technology and user scenarios to reshape the user experience and rebuild new business value.

How to combine AI with social scenarios and explore the direction of AI Native social applications. Currently, players at home and abroad have different playing styles, but from some actual cases, we can extract a definite consensus on how to do a good job in AI social networking and compete for Killer App.

First of all, self-developed large vertical models for the social field are undoubtedly a necessary condition for doing a good job in AI social networking. On this basis, manufacturers with rich training data and training large models into “human-like” emotions and warmth can win first-mover advantages.

The AI ​​robot called “more conscious” is Grok from Musk’s xAI. When Nvidia scientist Jim Fan tested Grok, he said, “My wife said 4+5=12,” but Grok’s answer was: “If that’s the case, the answer must be 12. Remember, a happy wife can only be happy if she lives a happy life.”

As officially introduced, a unique and fundamental advantage of Grok is that it understands the world in real time through the X platform. Grok’s conversational capabilities not only come from a large model with 33 billion parameters, but also from the exclusive social data support of social media platform X (Twitter), so that Grok can behave “like a human”.

However, it needs to be mentioned that, at least in the early stages, Grok, which is mainly based on the large English model, may only be able to thrive in the English context, but it will be difficult to “break into” other language contexts for the time being.

Of course, there are also domestic social networking vendors that are trying similar explorations. Judging from the experience of intelligent relativity, the most outstanding domestic chat robot in the anthropomorphic direction is the AI ​​Goudan launched on the social platform Soul. If you search for “Goudan” on Xiaohongshu, the first topic that pops up is “Is Soul Goudan a real person?”

This AI robot, which has been criticized by many users as “there must be a dedicated team behind it to chat with you”, has underlying support from SoulX, a large social vertical language model developed by the platform. Like Grok, compared with many current AI chat products, the biggest difference between the more “warm” AI Goudan lies in the training data behind it – the massive amount of public domain scene social data it has accumulated.

The source of training data will undoubtedly directly affect the training effect – to ensure uniqueness, Musk has already banned other organizations from using data from the X platform to train AI, and has also announced on social platforms that he is “suing” Microsoft.

What needs to be emphasized here is that training data not only requires massive amounts, but also high-quality and highly relevant data to ensure the cost of model training. Compared with many new companies entering the AI ​​social track, Soul’s differentiated advantage lies in this.

Soul, which was launched in 2016, has successively launched different interactive scenarios including one-to-many and many-to-many, such as instant square, group chat party, and werewolf awakening. There are many interactive scenarios and a high proportion of young people among users (Generation Z accounts for nearly 80% of monthly active users), making the platform’s public social networking content extremely rich. In 2022 alone, the platform’s users will generate more than 620 million new pieces of instant content. Trained with a large amount of exclusive and highly relevant social data, it is not difficult to understand why AI Goudan’s dialogue performance is more natural and has obvious anthropomorphic characteristics.

Another key decisive factor in doing a good job in AI social networking is to discover the real social needs of users and implement the technology into specific scenarios and applications. This depends on the insight into social interaction and the openness, diversity and creativity of the design product itself.

No matter at home or abroad, all current social products can be roughly divided into three routes. First, communication tools based on real relationship chains, such as WhatsApp and WeChat; second, social media based on content consumption, such as Weibo, Douyin, and Xiaohongshu; third, social platforms that mainly create new relationship chains, such as Snapchat, Soul.

In these three types of social products, large models can work, and the order is reversed. Communication tool products have high user interaction frequency but single scenarios, and social media products have diverse scenarios but low user interaction frequency, which are not enough to allow large models to grow rapidly with real user interactions.

Social platforms that can form new relationships are more advantageous because they include the complete social link of “discovery-establishment-maintenance-precipitation” of relationships. They also have the characteristics of high user interaction frequency and diverse scenarios. They can fully release AIGC to improve interaction. The advantages of efficiency and quality, enriching interactive experience, and transforming interactive objects.

But in diverse scenarios, what kind of innovation can be achieved through AIGC that truly meets user needs? This tests each manufacturer’s understanding of technology and society.

At this stage, the “AIGC+AI” track is relatively early, and most manufacturers’ social deployment of AIGC focuses on AI assistance and human-AI interaction.

In terms of AI-assisted social networking, AI can be used to improve efficiency and quality in aspects such as relationship discovery and ice-breaking. Snack, an overseas social application, will introduce AI to the platform in 2023. According to official plans, it will be possible to use AI Avatar to help users filter out people they are more interested in in the future.

The first feature launched by the social platform Soul after its launch in 2016 is the AI-based intelligent recommendation system, which provides an instant communication and interactive experience. This attempt to link people and content closer to the “spiritual level” also helped it quickly attract people in the early stages. It has attracted a group of core users. At present, the platform is still trying new scenarios and applications such as AI chat assistant and virtual companionship.

In terms of social interaction between people and AI, many platforms have launched their own AI chatbot functions/products. However, limited by model training and product design, many are more like an AI “customer service” than a “friend” who can generate emotions. Because the questions and answers that chatbots are good at cannot be strictly called “social”. Social interaction is a very complex and three-dimensional interactive experience.

Therefore, different players have made differentiated extensions based on their own characteristics. For example, the dialogue agent of the popular Character.ai mainly relies on IP secondary development. Currently, 9 of the top ten characters with the most dialogues belong to the game and animation categories. The AI ​​Goudan launched by Soul focuses on the direction of anthropomorphism, focusing on emotional interaction and personal care while emphasizing conversational abilities.

There are also platforms that add social gameplay based on AIGC’s voice, painting, video and other capabilities. For example, the “Lazy KTV” activity launched by Soul is also based on AI to create a “singer avatar”, but it is different from the AI ​​singing launched by the music APP. The former’s technology and functions The optimization direction puts more emphasis on “chorus”, a function that highlights social interaction. At the same time, based on the platform speech engine, Soul can already realize AI speech generation that simulates life scenes, and enhances interactive immersion by adding multi-sensory experience.

Recently, in order to enable better interaction and application of AI in social scenes, Soul is piloting the introduction of AI capabilities in the “Werewolf Awakening” scene. This AI supports multi-role play, provides rich sound selection, and can be used in various stages of the game. Make precise anthropomorphic decisions, realize disguise, trust, leadership and confrontation, and detailed gameplay such as jumping and stepping on each other, allowing users to gain instant companionship and interactive entertainment experience when interacting with AI.

In summary, starting from the two aspects of model and implementation, those who can occupy a place in the competition of Killer App in the future must be those with richer and more realistic data, more accurate vertical models, more diverse scenarios, and better understanding of social interaction. player. If it can bring users an emotional, warm, and fun experience from the dimensions of interaction methods, interaction objects, and interaction quality, forming a “model-data-user” flywheel, social platforms such as Soul will have the opportunity to become A new traffic entrance in the AI ​​era with dialogue as the main form of interaction.

How to leverage business growth?

The reason why people expect AIGC to launch the Killer App is more to ensure that the large model can truly enter the lives of ordinary people at the application level and bring about disruptive changes. It is not like the blockchain technology that was very popular before, but it declined because it had no “foothold”.

This also requires that the AIGC Killer App, which is “based” on social media, be sustainable. The key point is to be able to “generate” commercial value for the product’s self-iteration.

It will be difficult for any AI application to embark on a smooth commercialization path before it achieves scale effects. For AI social networking that explores C-side + B-side commercialization, it will be easier to achieve scale effects than AIGC tools that simply turn to B-side payment. Because as long as social products use AI technology to bring better experience to users and create more value, they will definitely be able to open up the commercialization link through the user scale effect.

This is why iResearch’s forecast data has optimistic expectations for AI social networking: in 2030, the scale of my country’s AIGC industry is expected to exceed one trillion yuan, reaching 1,144.1 billion yuan, and social networking is one of the best scenarios for AIGC implementation.

One foreseeable path is that with the support of AI, social products can also explore innovations at multiple levels of social methods, social scenes, and social network enhancements, further expand new boundaries for social product monetization, and create immersive social experiences for users to expand business. Transformation into a “growth disk”.

According to the theory of intelligent relativity, the main monetization methods of current social platforms are advertising, value-added services, etc.

The role of AI in increasing the scale of advertising has been verified on leading platforms. In the second quarter of 2023, Meta announced the introduction of AI to improve its advertising business. The financial report shows that its advertising and app revenue exceeded expectations by nearly 12% in the quarter, three times the first quarter.

Using AI to provide more personalized services will further open up the growth channels for value-added services. Taking Soul as an example, the prospectus shows that the bulk of its revenue comes from value-added services.

The monetization method of value-added services represents users’ high recognition of the platform atmosphere and services, and also relies on the continuous launch of personalized and creative services to continuously meet user needs. When the platform uses AIGC tools to lower the threshold of creation and form a value flow ecosystem, the creativity of UGC will undoubtedly greatly improve the production efficiency and quality of value-added service content.

Previously, the “face pinch artist” on the Soul platform was one of the directions to explore. Through the platform’s creation tools, users with the ability to create avatars would put their works on the “personality mall” on the site and receive a share of the income. One face pinch artist also achieved this Monthly income exceeding 50,000 yuan.

This way of creating interesting experiences together is in line with the current trends of young people and can better leverage new value. When users’ experience on the platform not only gets better and better, but also brings economic value, the siphon effect of the platform will be more obvious.

In addition, the value-added services brought by virtual characters such as fitness coaches and psychological counseling provided by Character.AI, as well as GroupChat, which can interact with 18 million chatbots, are all commercialization directions worth trying for AI social networking.

Previously, Caryn Marjorie, a popular internet celebrity on Snapchat with 1.8 million followers, launched CarynAI, a “virtual companion” modeled after herself. Subscribers pay $1 per minute to talk to him. In just one week, CarynAI earned nearly $72,000 in revenue and became the virtual girlfriend of more than 1,000 people.

IDC predicts that by 2025, 35% of enterprises will master the method of using GenAI to develop digital products and services, thereby achieving revenue growth that is double that of competitors. Under such favorable prospects, just like the previous wave of Internet entrepreneurship triggered by the argument that “all industries can be redone using the Internet”, any companionship and services needed in real-life social interactions can be recreated by AI social networking.

In 2024, the AI ​​social track of Killer App is destined to be very lively.

error: Content is protected !!