College Entrance Exam Volunteer Application: The First Gold Mine of AI Search Commercialization?

07/02 2024 506

A few days ago, the college entrance exam results were released, and many friends were busy applying for universities for their children or relatives' children. College entrance exam volunteer consultation also became extremely popular, and Zhang Xuefeng's volunteer consultation products sold at astronomical prices.

On the eve of this year's college entrance exam, the consulting products on the "Fengxue Weilai" APP under Zhang Xuefeng's popular online celebrity brand were snapped up, and some people estimated that the sales exceeded 200 million yuan.

According to Tianyancha APP, Fengxue Weilai belongs to Suzhou Fengxue Weilai Education Technology Co., Ltd., which was established in 2001.

With sales revenue of over 200 million yuan in just three years, it may be a good performance for Zhang Xuefeng, but for internet giants with revenue often reaching tens of billions, this amount is indeed just a drop in the bucket.

Interestingly, in recent years, giants like Baidu, Alibaba, and ByteDance have also emerged in the field of AI volunteer application. AI filling out college entrance exam volunteer applications seems to have become a "new blue ocean" in the eyes of the giants.

The areas where giants gather are not necessarily very profitable, but they must have strategic value.

Why do giants all target AI volunteer application? Why would AI search + large models be the next battleground for big factories? The reasons behind this may be worth deeper consideration.

Is AI volunteer application reliable?

There are two layers of information in the internet world. The first layer is information that can be queried using search engines, such as tomorrow's weather or whether a flight is delayed. The second layer of information is processed information obtained through analysis of the first layer of information.

In the past, the internet mostly produced the first layer of information, with a small amount of second-layer information. In the AI era, more information is "processed" by large models.

However, at present, the value of AI-processed information is not as high as expected.

Taking volunteer application as an example, the current AI volunteer application's greatest value lies in helping you organize the first layer of information, such as the score lines of various universities and majors, and predicting admission references based on statistical data. Essentially, it replaces users in doing a lot of data collation work.

This is actually what large models are best at: doing a lot of data analysis and logical analysis.

The reason why Zhang Xuefeng's courses sell well is that he provides the second layer of information for volunteer applications. For example, whether a certain school major is strong or good for employment, what the working environment is like after employment, and whether the personality is suitable. These deep-level pieces of information are valuable.

Currently, AI volunteer applications on the market cannot go as deep as this, which means that from a product perspective, AI volunteer application is essentially still a data screening tool.

It can be said that from a pure tool perspective, AI volunteer application products are reliable, but from a user decision-making perspective, they are not very reliable, because large models cannot really help parents make decisions like Zhang Xuefeng.

Economics says that value determines price. Since AI volunteer applications cannot make decisions, their paid prices are naturally much lower than those of volunteer consultations. Therefore, from a commercial perspective, while Zhang Xuefeng makes a lot of money, AI volunteer application products may not necessarily be profitable.

However, AI volunteer application products also have an advantage, which is that the scale of covering users can theoretically be infinite. As long as the platform database is comprehensive enough and the data is updated in a timely manner, some people will pay for it. However, it is difficult for this payment to lead to repeat purchases.

Good businesses are either high-priced and low-frequency or high-frequency and low-priced. AI volunteer application fees are inferior to consultations and hardly generate repeat purchases, so it is not a good business.

In fact, from the perspective of the first principle of demand, volunteer application is actually a pseudo-demand.

In terms of decision-making logic, the college entrance exam is a major event in life, and decision-making ultimately relies on humans. As parents, they cannot entrust their children's future to an AI volunteer application. Therefore, essentially, they still need to integrate information from various channels and make decisions on their own.

So, paying for volunteer application is essentially spending money for peace of mind.

Although volunteer application is a pseudo-demand, what lies behind it is the "real demand" of examinee parents: future career planning.

Do examinee parents want a list of universities and majors? No. The essence of AI volunteer application is a mismatch between demand and product. Although the rules and forms of filling out volunteer applications are changing, it is not difficult to fill out a volunteer application form. The difficulty lies in combining subsequent career planning with volunteer application.

Therefore, filling out volunteer applications is only the first step of solving the problem. The key lies in providing career planning based on each family's environmental conditions, the examinee's scores, interests, and hobbies.

Zhang Xuefeng can achieve this, so his course products can sell for 200 million yuan, but obviously, the current large models cannot, even those from giants like Baidu and Alibaba.

So why are big factories still investing resources in this matter? The answer may lie in the large model technology itself.

Baidu and Alibaba Compete for New Territories in Internet Goods and Services

Baidu and Alibaba's entry into AI volunteer applications is not primarily to compete with Zhang Xuefeng and others.

Since the popularity of AI large model search, the industry has witnessed two waves of enthusiasm.

One is that everyone has started to promote their own large model APPs and AI searches, such as Baidu's Wenxin Yiyan, Alibaba's Quark AI Search, and Tencent's Tencent Yuanbao. The purpose of each company introducing APPs is straightforward: to build their own large model traffic pool and seize the initiative.

The other is that, in different scenarios, everyone hopes to connect their large models with specific applications. For example, Alibaba integrates DingTalk into large models, and ByteDance also integrates Feishu into large model capabilities.

On Baidu's side, in addition to launching AI search products, it recently released Wenxin Large Model 4.0 Turbo, further enhancing capabilities in coding, understanding, logic, generation, memory, and other comprehensive AI capabilities, and opening these capabilities to B-end.

Behind these two waves of enthusiasm, the leading internet giants are actually striving to achieve one goal: using large models to amplify their traffic pool and hoping to consolidate their entrance status in the old era.

So what does the giants' grab for entrance have to do with AI volunteer application?

In fact, there is a relationship. The internet believes that the giants selling AI volunteers seems to prove that large-scale user payments are feasible for the product form of AI search + large models.

The motivation for users to pay for AI volunteer applications is strong. Whether it is AI volunteer applications or AI-assisted applications, the underlying logic is that there is non-subscription large-scale user payment behavior on the product form of AI search + large models.

What does this mean?

It means that in some rigid demand scenarios, people may pay for products and services based on large model capabilities, rather than for the functions of the large model itself.

As we all know, the earliest internet business model was selling software, which essentially paid for the internet product itself. Later, people found that they could not only sell software but also sell advertisements, goods, and even services, leading to the current internet business.

In a nutshell, highly developed internet businesses are actually based on users paying for products and services on the internet.

Will large models be the same?

There is no definitive answer yet, but it is almost certain that the increment of short videos + live streaming has gradually reached the end of its cycle, and the last wave of dividends in the mobile internet business era is approaching. Giants urgently need a new incremental field to drive revenue growth.

Therefore, even though large model products are not yet perfect and have many limitations in landing scenarios, we can see that Baidu, Alibaba, and ByteDance are constantly making new attempts.

Currently, this attempt can actually generate revenue. Goldman Sachs once judged that Baidu's AI solutions would bring Baidu an increment of 3 to 6 billion yuan.

Although this increment seems insufficient compared to the revenue scale of the giants, at the recent Wenxin Large Model 4.0 Turbo version launch conference, Baidu mentioned a statistic. Baidu launched Wenxin Large Model 1.0 in March 2019. Currently, the cumulative user scale of Wenxin Yiyan has reached 300 million, with 500 million daily invocations.

In just over three years, with a user base of 300 million, this data seems to indicate that AI products centered on large models are rapidly forming their own traffic pool.

How to quickly increase the volume and build a traffic pool centered on their own large model products is actually a question that large factories have been thinking about all along.

If large factories can build a traffic pool centered on large models within the next three years, it will be easy to replicate their past success and put more goods and services in this new territory. By then, new traffic will drive new growth, and the commercialization of AI large models may take on a different look.

The deciding factor may not be the product itself, but the ability to productize

For internet giants, under the impact of large models, everyone is also facing a transformation window period:

That is, the revenue and profit structure centered on goods and services and mobile internet needs to gradually transform into a revenue and profit structure centered on AGI.

The former is represented by business models such as short videos and live streaming, while the latter is where the future growth lies.

Accomplishing this is not easy.

Since the explosion of AI large models, I have been thinking that perhaps one day, just like the era of earning money through search engines will end, the era of earning money through short video traffic will also end. By then, what will be the next product to occupy users' rigid needs? AI search? Or a large model APP?

The internet believes that large models may eventually become the next "rigid demand" capability, but not necessarily a specific large model product.

Whether AI was implemented on the B-end in the past or large models nowadays, AI's true value essentially lies in its "embedded" capabilities. Remember the Apple product launch? Although there was not much talk about large models, their shadows can actually be seen in the product details.

Therefore, for users, what is truly useful is not necessarily a dedicated "large model APP," but rather the large model capabilities embedded in a product.

For large factories, this is actually easier to achieve.

On the one hand, large factories already have mature application products and scenarios, and the user base is large enough. How to better integrate AI large model capabilities into their own products may be more meaningful for the company's transformation into a revenue and profit structure centered on AGI.

In this sense, Baidu's continuous refinement of AI + search, Alibaba's integration of large models into DingTalk, and ByteDance's creation of large models + Feishu all have similar considerations.

On the other hand, for large factories, they actually have sufficient resources to continuously iterate on such capabilities.

Essentially, "large model product power" needs continuous iteration and upgrading, supported by a large amount of data training, which is a unique advantage of the giants themselves.

When it comes to products, WeChat, the social media giant, must be mentioned.

With large models so popular, I'm really curious about when and how Tencent WeChat will enter this field.

Currently, Baidu, Alibaba, and ByteDance are quickly grabbing users, though with different focuses, but they are moving quickly. In contrast, WeChat does not seem to have revealed more plans regarding large model capabilities.

WeChat is a very special and potential product. In the era of short videos, Douyin and Kuaishou rose rapidly, while WeChat Video took a step back. The field of AI large models cannot miss this opportunity again.

In fact, there are many scenarios where large models can be applied in WeChat.

For example, search is also important for WeChat. WeChat's content and service ecosystem is vast, not just a social APP, but hiding a huge amount of content and services.

The value of large models + search lies in efficiently screening out valuable parts from this vast amount of content and services.

In the past, subscription accounts were the first to do this job, followed by mini-programs, and today it is video numbers. Therefore, video numbers have been emphasizing live streaming and their GMV has been increasing.

In the future, will large models become another video number or mini-program? What new opportunities will there be for brand merchants and content creators?

It's worth looking forward to.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.