To be frank, we don’t have any competitors in the college entrance examination. It’s not that we don’t look at others. For so many years, the entire team has only focused on what users need and turning technology into products
Zheng Sishou, Product Manager of Alibaba’s Qianwen Business Unit, recently stated that the team has been working on college entrance examination tools for eight years, and AI is making things different.
Before this year, the mindset of users filling out college entrance examination preferences was Quark, but after a year of integration and adjustment, the product that came to the forefront has become Qianwen. On June 10th, Qianwen released the full cycle college entrance examination volunteer filling agent, which is officially referred to as an agent. At present, among the three similar products of Tencent, Baidu, and Alibaba, Tencent’s Yuanbao Gaokao Tong and Alibaba’s Qianwen Gaokao have both focused on Agent features. The real difference between Qianwen and Gaokao is that they have extended the product application cycle.
What supports Qianwen is the AI Agent’s set of thinking, planning, execution, and reflection, while also possessing proactive planning, long-term memory, and personalized services. Internally, it is believed that tool based products cannot solve the long cycle and personalized needs of college entrance examination volunteer filling. Agents can cover the entire cycle of volunteer filling and even career planning.
Qianwen Gaokao replaces Quark Gaokao, and the iteration of Gaokao products from other manufacturers reflects the shift from search+tools to large models+agents.
A complex mathematical problem
The college entrance examination application product focuses on a matching related mathematical problem: finding the optimal solution from a massive amount of information from nearly 3000 universities, over 2000 majors, and potentially billions of combinations across the country.
The standard answer may be found, but it’s too slow and difficult for people to solve this set of questions. Even if there are corresponding experts, resources are extremely scarce and ultimately need to be filled with tools.
Before making a decision, candidates must first grasp historical data and estimate probabilities based on past admission rankings and line differences. When filling out the application, it is also influenced by multiple factors such as school level, professional strength, personal interests, location, and employment prospects.
Traditional tools have completed information preprocessing and standardization, and users can obtain corresponding answers by simply searching. But in reality, filling out college entrance examination preferences is a complex task chain that lasts for more than 20 days. Before scoring, it is necessary to estimate the score and preliminarily determine the direction based on personal interests.
The most laborious part is to match schools, majors, and regions based on the scores after scoring. After making the decision, it is necessary to carefully check and verify whether there are any errors. What is parallel volunteering, which majors of physics and chemistry students can apply for, and what expectations parents have? The game in the real world goes far beyond the first principle of using up tools and leaving.
In the process of filling out volunteer applications, despite the lack of manpower and limited tools, AI excels the most. The combination space for voluntary application can be estimated by AI to cut out obviously inferior combinations; Admission probability prediction is a natural landing scenario for pattern recognition and regression analysis; In the end, there are AI recommendation systems and multi-objective optimization that can cover individual recommendation schemes.
Tencent and Alibaba have introduced agents and provided two completely different solutions to make up for the huge space left by the tools.
On June 5th, Tencent announced that Yuanbao and QQ Browser have released the college entrance examination information agent “Yuanbao Gaokao Tong”, with the main function being the college entrance examination consultant agent. Yuanbao is a mobile portal, while QQ Browser provides a PC portal. In terms of product usage process, candidates input scores, volunteer recommendations, college and major screening, and past data queries. Apart from multiple rounds of conversations and long memories, there is not much difference from past products.
Tencent’s restrained product strategy has certain considerations. Volunteer application is not a life simulation game, and the impact of one choice is very long-term. Enterprises can only provide auxiliary tools and cannot participate in decision-making.
Five days later, Alibaba Qianwen released its own college entrance examination application agent, which is much more aggressive than Tencent’s. The product consists of three core competencies: volunteer calendar, volunteer report, and volunteer Q&A. Qianwen Agent has divided the college entrance examination volunteer filling scenario into three core scenarios: on time, by report reference, and by dialogue fine-tuning, thereby transforming the volunteer filling process from user initiated retrieval to AI assisted decision-making.
After filling in the province and other information, the candidate enters an automated process. AI generates a schedule from the first day after the exam to the submission of preferences, and pushes cognitive establishment, score positioning, direction exploration, preliminary selection of plans, and formal filling, using structured methods to establish rational cognition for candidates.
In 2025, generative AI will enter the field of volunteer reporting for the first time, and all companies will provide the ability to generate reports. However, due to the personalized nature of voluntary application, it is not possible to simply evaluate what is good and what is bad. Therefore, product comparison is not only about the accuracy of the report, but also about the breadth of information provided.
Both Qianwen Gaokao and Yuanbao Gaokao offer stable recommendations and dynamic adjustments, but Qianwen has set more constraints on recommendations, such as professional preferences and specialized qualifications.
In the past few years, volunteer Q&A has been the most distinctive feature of various college entrance examination products. The five supports mentioned by Qianwen Agent at the conference can be summarized as dialogue memory, data accumulation, score conversion using line difference and ranking method, visual display of structured data, and turning past recommendations into active questioning.
The decline of tools is not wrong, but the voluntary application can only be for assisted driving
Before the big model+Agent, the college entrance examination application products had always been in a state of homogeneous competition.
The products launched by any giant are essentially the output of information connection capability of the Internet, completion of data accumulation, optimization of algorithm testing, and finally landing in the form of search. For example, Tencent first piloted the KNN+LM algorithm in 2017, then collected admission, college, and policy data, and launched the new Gaokao platform in 2021.
Since 2021, the college entrance examination volunteer filling, like the New Year’s red envelope, has gradually become a node for Internet companies to show strength, transform traffic and rotate the data flywheel. During this period, the college entrance examination tool went through a process from searching to filling out college applications, and then generating college application reports last year, with a uniform product form.
For example, Quark’s college entrance examination went through search queries, launched the “Chongzhenbao” tool, and introduced a volunteer filling function. After adding the generation function last year, the continuous stacking of functions did not change the background of the search+tool. Moreover, for large companies, these tools do not have a moat. Their launch this year is exclusive, and next year they will become infrastructure.
The product strategy of Qianwen Gaokao Agent implies two business intentions.
There is no moat for the tools under homogeneous competition, and the examinees simply do not care who is the pioneer of “Chong Wen Bao”. The score line query is not the evaluation standard for the quality of the manufacturer’s products, but the passing requirement. The Internet has long erased the information gap between colleges and universities. In other words, search+tools have reached the end of diminishing marginal utility.
As long as we stay at the level of tools, there is no moat for filling out college entrance examination preferences. The AI generated volunteer analysis report, which was launched last year, has become a standard feature this year. Homogenization ultimately becomes an arms race at the functional level, and products cannot form differentiated pricing, making free pricing the norm.
The biggest variable brought by Agent is the reconstruction of the interaction itself. In the tool stage, the user’s behavior is active search, while the agent relies on the inference, memory, tool invocation of the large model, as well as the continuous acquisition of user information, to rotate the data flywheel.
Zheng Sishou, the product manager of Qianwen Business Unit, said, “In the process of doing Qianwen, we want to more directly let candidates and parents discover their needs, hobbies, and interests during the Q&A process. This is a different logic from the past, it has moved from one-way use to anthropomorphic interaction
Due to the unclear purpose of candidates and parents, it is difficult for tools to intervene in subsequent actions. The solution proposed by Qianwen is to align data through multiple rounds of dialogue between humans and AI, and this approach does not have any novelty: it imitates human behavior.
College entrance examination consultants not only have access to data on various regions’ college entrance examinations over the years, the enrollment situation of colleges and majors, and rich experience, but their more important ability is to quickly obtain information about candidates and parents. Whether it is the examinee’s the final say or the parents’ the final say, whether they want to go to the coast or the mainland, and what is their desire to seek jobs.
The big model+agent controls structured and standardized data, and can quickly use tools to make judgments. However, if candidates and parents cannot actively speak up and provide information, the entire path cannot be completed.
However, strictly speaking, interactive anthropomorphism still lacks a moat, and Qianwen Agent still harbors another more ambitious plan.
The application of college entrance examination preferences is a cyclical scenario and lacks long-term sustainability. After agentization, it will be recommended through dialogue and constraints, continuously expanding to more areas along the scope of the preferences, and completing the binding of user habits.
Photon Planet has learned that this year’s college entrance examination is a test, and Qianwen will plan and develop follow-up products based on the test results. For example, updating the calendar and Agent mechanism to provide a wider range of services, from this perspective, Qianwen Gaokao Agent is just one of the many incisions in AI penetration into the learning and growth process.
Flower Token, Buy Trust?
Qianwen Gaokao Agent did not break away from Alibaba’s AI strategy of “producing tokens, transporting tokens, and consuming tokens”.
The college entrance examination application provides a high consumption, high concurrency, and super high trust scenario for Qianwen Agent, which provides momentum for driving Qianwen’s C-end stickiness and adjustment volume, and driving the Alibaba Cloud MaaS platform.
In 2025, Quark will launch its first AI volunteer report, and behind the 13 million reports generated is a huge investment in computing power. Zheng Sishou did not give any direct response when talking about the issue of computing power consumption recently, only saying that “from the perspective of the group, there are no restrictions on the entire matter, and comprehensive computing power support has been provided”.
The cost of tokens is considerable, and it is a high-quality consumption of tokens. Every interaction between candidates and parents involves deep participation, genuine decision-making, and trust investment. Qianwen Gaokao Agent consumes tokens, but in exchange, it gains trust assets.
In addition, the college entrance examination is also one of the high-frequency and essential needs scenarios at the national level. Alibaba is trying to enter each scenario and create a “Double 11” similar to annual token consumption, covering all aspects of life, study, and work through the accumulation of peak values. The day after the release of the Qianwen Gaokao Agent, it once again plunged into the riverbed of the World Cup.
The scene behind the college entrance examination is profound, and the accumulation of trust assets can easily transform one-time consumption into long-term dependence. And where can the prediction of the World Cup be “acted upon”? Apart from buying lottery tickets, there is currently no sign of a backer.
Perhaps Qianwen just wants to ‘gamble’.
相关文章
