The Shadow Under the Blooming Flowers: The "iPhone Moment" for AI Applications Has Not Yet Arrived
As smartphones are the mainstream computing platform, most AI applications currently take the form of apps. Due to the billions of users in mobile internet, many AIGC apps have successfully entered the market since 2024.
Data disclosed by QuestMobile shows that in June 2024, there were 2 AIGC apps in China with over 10 million monthly active users, namely Douyin and Wenxin Yiyan. There were 8 AIGC apps with over 1 million monthly active users, including Tiangong, Kimi, Xingye, Xunfei Xinghuo, Tongyi Qianwen, Zhipu Qingyan, Hailuo AI, and Maoxiang.
In addition to standalone AIGC apps, mini-programs and application plugins are also becoming "AI-enabled". For example, Youdao Dictionary's Little P module has introduced NetEase's Ziwei large language model, which can be applied to translation, article polishing, grammar analysis, and other scenarios.
It can be said that against the backdrop of rapidly diminishing traffic dividends, the mobile internet industry has not seen such a vibrant wave of product innovation for a long time. The "AI-ification" of internet products has become a new competitive point in the industry. If tech companies want to demonstrate imagination, they cannot avoid launching AI-related application products.
The launch of these AI application products largely relies on the promotional efforts of related companies. For example, according to AppGrowing statistics, from April to May 2024, Douyin's advertising spend was about 15-17.5 million yuan. By early June, ByteDance launched a new round of large-scale advertising for Douyin, with an investment of 124 million yuan.
Additionally, according to a research report by Open Source Securities, in January 2024, Kimi Chat began advertising on Bilibili, collaborating with university professors, academic KOLs, and AIGC geeks with mid-tier followings (500,000 to 4 million fans) to release videos tagged with "Free ChatGPT", "Thesis Helper", "English Learning Assistant", etc., precisely targeting Bilibili's core user group for product promotion.
From this, we can also see that the strong rise of AI large model applications is still based on the "high wall" and "heavy investment" strategies of the mobile internet era. This greatly helps the initial traffic growth of emerging products, but judging from some existing problems, tech companies may need to further identify their value anchors to extend the lifecycle of AI applications integrating into users' lives.
The true "iPhone moment" for AI applications may not have arrived yet. The "iPhone moment" is often used to describe how a technological innovation or product launch rapidly changes people's lifestyles and becomes an indispensable part of daily life.
Although there are voices in the market claiming that OpenAI has brought about the "iPhone moment" of artificial intelligence by launching shocking AI application products such as the natural language chat product ChatGPT and the text-to-video tool Sora.
However, looking at the current application status, these products have not yet demonstrated the "infrastructure" attributes similar to WeChat, Alipay, Pinduoduo, and other products that have caused widespread changes in a vertical industry.
Furthermore, looking at the bigger picture, although the user base of generative AI continues to expand, user stickiness still needs to be strengthened. Data from QuestMobile shows that over the past year, more than 40% of AIGC apps experienced traffic decline, and some applications have high uninstallation rates, such as Xinghui, Nieta, and WOW, with monthly uninstallation rates as high as 92.0%, 78.1%, and 67.5% respectively.
This largely indicates that existing AI applications are very much like "seller's show" items, which may look extremely sci-fi, but are not necessarily "fitting" for users.
In this situation, where should AI manufacturers go?
Super Application or "Super Capable" Application? Professionalism and Ecosystem are Key Competitive Points
In fact, at the 2024 World Artificial Intelligence Conference, AI manufacturers have already seen the ultimate answer to the above problems - only by avoiding traffic-oriented thinking and identifying application scenarios can they truly tap into the potential value of AI technology.
In this regard, Robin Li, founder, chairman, and CEO of Baidu, once said, "Many people are focusing on when GPT-5 will be released, but I'm more interested in which applications can fully utilize all the capabilities of large language models."
At the 2024 World Artificial Intelligence Conference, Li also reflected, "We should avoid falling into the 'super application trap', thinking that only an app with 1 billion DAU can be considered successful. This is the thinking logic of the mobile era."
In short, Li's current focus is not on isolated model parameters or product monthly active scale, but on how to create products that fully utilize existing AI capabilities and bring value to users. Only in this way can AI large model technology escape the fate of being too high-brow.
In fact, the super applications in the current mobile internet track have billions of users not simply by "building high walls" or "heavy investment", but by being built on the basis of bringing tangible value to users.
Take WeChat as an example. Tencent's financial report shows that in the first quarter of this year, WeChat's monthly active users reached 1.359 billion, a year-on-year increase of 3%. Achieving such a scale is largely because WeChat's various functions highly meet user needs. For example, WeChat Pay can reduce users' transaction costs, and official accounts can help users easily access information.
From this perspective, regardless of the form of AI carrier, if it wants to stand out and successfully close the business loop, it needs to find suitable scenarios and achieve high-quality user experience. At present, to achieve these goals, AI manufacturers need to continue to improve in terms of specialization and ecosystem.
First, AI large models usually have excellent data processing capabilities and deep learning abilities, which are actually very suitable for vertical tracks that pursue specialization and accuracy, such as medicine, law, and finance.
Based on this, Baidu, Baichuan Intelligence, SenseTime, Tencent, and other tech companies have brought AI applications represented by "intelligent agents" and have begun to focus on vertical fields such as healthcare, education, and finance.
For example, at the 2024 World Artificial Intelligence Conference, Baichuan Intelligence launched a beta version of the medical application AI Health Advisor, built on Baichuan Intelligence's general medical enhanced large model, possessing rich medical knowledge and doctor's thinking.
Regarding this, Wang Xiaochuan, founder and CEO of Baichuan Intelligence, said that the current severe shortage of doctors presents a great development opportunity for AI applications. "Healthcare is the jewel in the crown of large models." In addition, it is understood that Baichuan Intelligence is also committed to creating fully automated health management, which can surpass AGI, managing the entire health process of patients from prevention and diagnosis to treatment.
Of course, delving into vertical fields particularly tests the professional capabilities of AI large models, and current related technologies still have significant room for improvement. As Liu Yuhong, Vice President of Tencent Cloud and head of Tencent's Hunyuan large model, pointed out, "Large model technology has only developed for one to two years since the most popular ChatGPT-3.5, and products are still in a very early stage with insufficient maturity."
Going forward, "capability" will further become a new consensus for measuring AI value, and AI manufacturers still need to settle down to accumulate industry experience and business scenario know-how.
In addition to building vertical professional capabilities in depth, AI manufacturers are also committed to horizontal development, building open AI ecosystems. As AI technology applications enter deep waters, scenarios become more diverse and complex, which will to some extent increase the difficulty and cost for relevant manufacturers to gain insights into needs and conduct technological research and development. In this situation, embracing more independent software developers can, on one hand, accelerate AI application innovation, and on the other hand, facilitate the realization of scale effects.
For example, in June 2024, DingTalk announced that it would open up to all large model manufacturers, building an open AI ecosystem to provide customers with services such as model training and optimization, AI solution creation, and AI custom application development. It is reported that DingTalk's AI ecosystem partners currently exceed 100.
Furthermore, in early July 2024, Baidu Intelligent Cloud announced that the Wenxin large model 4.0 Turbo is fully open to enterprise customers, and the flagship models ERNIE 4.0 and ERNIE 3.5 have significantly reduced prices.
To further lower the threshold for downstream customers to implement large models, Baidu Intelligent Cloud also launched the Qianfan industry scenario solution, creating reference "samples" such as scenario models, Prompt templates, model fine-tuning showrooms, and application showrooms, making it convenient for customers to use directly or quickly complete large model application development through copying.