news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Yoky
Email: [email protected]

What will be the development trend of large-scale model products?

What will the next national-level AI application look like?

In February 2023, ChatGPT was born, igniting the concept of AI Native and making a number of entrepreneurs and giants excited. They tried to break the tradition and explore new forms of independent apps, subverting the application ecosystem from top to bottom. However, 18 months later, we found that everything was not as fast as we imagined.

On August 13, QuestMobile released the "2024 Generative AI Large Model Application Ecosystem Research Report". The positive thing is that according to QuestMobile's monitoring data, independent apps have run two AI applications with tens of millions of user traffic, Douyin Group's Doubao and Baidu's Wenxin Yiyan with 27.52 million and 11.34 million respectively. But surprisingly, 80% of independent applications have less than 500,000 traffic.

The fundamental reason is that AI applications are very different from the mobile Internet era. At that time, there were a large number of unmet user demands on mobile devices, requiring innovative applications from 0 to 1. However, today, untapped demands are not the majority, and just painting or writing poems cannot make AI applications "big" quickly.

This may be the reason why another form of AI application has emerged. QuestMobile data shows that 90% of the top Internet apps are deploying embedded AI applications, using AI to transform the existing service experience and improve the quality and efficiency of solving problems. Among them, the number of users of Zhixiaobao, an "AI financial assistant" that provides services on Alipay, has reached 59.08 million, and the number of users has surpassed that of independent AI apps.

This makes us realize that whether it is an AI-native independent APP or embedded AI, the external "face" form is actually not important. AI that can truly be used by the public is the key to the implementation of technology.

1

AI independent apps cannot take a “crash course”

Looking back at the development from the Internet to the mobile Internet, we find that the short video platform Douyin, the LBS-based food delivery platform Meituan, and the smartphone-based WeChat and Alipay are all mobile-native products. They have created new product forms and new business models, and quickly become the top products in the mobile Internet era.

That is why, at the beginning of the AI ​​era, the entire industry is exploring the form of AI Native and wants to create an AI version of the super APP.

From a technical perspective, AI Native applications are indeed "sexy". They can develop unlimited application functions around AI technology, provide users with highly personalized and novel experiences, and enhance brand recognition. Such product forms also create more space for intelligent interaction. AI Native applications can achieve a more natural and immersive interactive experience.

But at the same time, it also leads to a misunderstanding in application innovation: AI products must be Native in order for AI technology to develop further.

Facts have proved that the product form of AI itself, whether it is an independent APP, Agent, mini-program or application plug-in, is not that important. What is important is to find the right scenarios, needs and user groups.

We often refer to this stage as “looking for nails with a hammer”, meaning that AI capabilities are available, but the key lies in what problems it can solve.

In the continuous breakthrough of scenarios, GPTs released at the OpenAI Developer Conference last November brought Agent to the public's attention, as if embedding an Agent could create the so-called AI Native application in one fell swoop. However, due to the lack of support from real demand scenarios, many GPTs projects eventually had to withdraw from the market, and even Microsoft directly cancelled the GPTs project.

Obviously, it is inevitable that giving AI independent apps a "crash course" is a bit premature, and the industry's rapid follow-up and catch-up have resulted in applications becoming "more and more similar." According to the QuestMoblie report, the types of applications are highly concentrated, whether it is LLM language processing or text and image generation, video generation, they are concentrated in the category of productivity tools in various industries.

The homogeneity of scenarios has led to the homogeneity of products. Many LLM products are almost identical in product form, capabilities and even UI design.

An independent developer told us: "AI applications are now facing a very low threshold for user migration. After a certain APP becomes charged, many users will choose another free one, because the product capabilities are homogeneous and there is almost no irreplaceability."

More importantly, AI products are very different from products in the traditional Internet and mobile Internet eras - they need to learn and evolve through continuous use. This requires feedback from real user data and continuous polishing of specific scenarios to make the products increasingly accurate and perfect.

For this reason, the reality is that many AI products lack sufficient user base and application scenarios in the early stages, resulting in the inability to collect enough data to optimize the model, creating difficulties at the start.

However, the emergence of independent apps with tens of millions of user traffic also gives the industry hope for continued growth. The key is that everyone should downplay the illusion of "quick success" and give them more time and space to find new breakthroughs in the ever-changing market and find a sustainable business model.

1

Embedding in super apps is another main channel for AI implementation

When we try to create AI super apps, we ignore the fact that we already have super apps.

The rise of embedded AI has allowed us to see more possibilities for AI applications. Take Alipay as an example. As the second largest commercial open ecosystem in China, it has more than 4 million mini-programs, covering 90% of merchants in major industries such as retail and catering. Nearly 700 million people use its life service mini-programs every day. This huge ecosystem provides a ready-made and active landing scenario for AI applications.

The national APP is also using its verified user needs and scenario soil to create more practical AI products.

Based on the Alipay platform, Ant's AI technology attempts to implement "three butlers", covering three popular scenarios: financial management, life services, and medical health. Thanks to the existing business scenarios and traffic in the Alipay ecosystem, it quickly welcomed the first wave of users.

Among them, Alipay's smart assistant, which serves as a life butler, can help users use AI to call up various services and complete daily tasks such as buying coffee and shopping, allowing AI to integrate into the lives of more ordinary people. The AI ​​financial assistant "Zhi Xiaobao", which was launched earlier, focuses on financial management and insurance professional knowledge Q&A, and can provide professional services such as market interpretation, position analysis, insurance configuration, and investment education. The latest monthly active users have reached 59.08 million, which only accounts for 6.6% of Alipay APP users.

By being embedded in the ecosystem of super apps, they provide a ready-made, active landing platform for AI applications, allowing the AI ​​flywheel to turn quickly without having to find scenarios and build a user base from scratch. On the other hand, AI capabilities also feed back to the ecosystem, bringing a more intelligent and personalized service experience to users of super apps such as Alipay.

Similarly, Kuaishou's Keling also provides a vivid example. After Keling became popular, many users directly enriched Kuaishou's short video ecosystem by generating videos through Keling.

In contrast, the scenario challenges faced by some independent apps during their implementation can be easily solved by embedding AI in super apps. Embedded AI applications can also be smoothly migrated to different service scenarios with lower switching costs.

QuestMobile's report believes that AI application plug-ins rely on ecological traffic pools such as super apps to reach business users at a low cost, and can quickly integrate with existing business scenarios and capabilities within the ecosystem. Under the competition of the existing Internet ecosystem, plug-in AI applications have a higher chance of breaking through.

Back to the topic at the beginning, one of the trends that big models have identified is to solve user needs and pain points, find suitable landing scenarios, and create predictable business models. In essence, these are the "inside" of big model product landing, rather than short-lived concepts.

It is foreseeable that, whether it is the attempt of independent APP or the strategy of embedding AI in super APP, they will find the best path for each adaptive market on their own exploration path. The diversity of technology and the richness of innovation will coexist, jointly promoting the development of AI applications to a deeper level and wider fields, allowing more ordinary people to feel the difference brought by AI.