GuShiio.com Wen| Zinc Scale, Author| Li Xuanqi, Editor| Li Jinlin
“DeepSeek’s impact on the education industry has exceeded any major model in the past two years, and now is the true Normandy moment for educational AI.……”
The emotion expressed by Zhou Haifeng, director of strategic investment at Yuanhao Future, in an interview with Douxiao.com, is a faint reflection of the changes and challenges that the education industry is about to face. In the past few years, from traditional education hardware companies to education service players, to technology Internet companies involved in education hardware, all have launched AI models and related products. The emergence of DeepSeek has become a variable, such as the whale on its logo, which instantly stirred up waves throughout the industry.
Recently, many education companies such as Netease Youdao, Learning Ersi, Xiwo, Yi Xiaoxiang, Zhonggong, Listening Bear, Yunxuetang, and Shulang announced that they will embrace DeepSeek. Some have equipped this model in their intelligent education hardware products, and some will access its reasoning capabilities.
It is often wise to follow the general trend, but some have hidden ambitions and some are worried.
Expanding the circle of friends is mostly to take advantage of the situation“”
Shortly after the Spring Festival holiday ended, educational technology company Netease Youdao took the lead in announcing that it would fully embrace DeepSeek-R1.
According to reports, Netease Youdao’s AI general learning assistant Youdao Xiao P will combine the thinking and analysis capabilities provided by DeepSeek-R1 ‘s long thinking chain to further upgrade personalized answering questions, while Hi Echo, Youdao Zhiyun, QAnything and other products will also be fully connected to DeepSeek’s reasoning capabilities and will be upgraded one after another in the near future.“” In addition, new smart hardware products that integrate new reasoning model capabilities and Youdao Small P 2.0 version will be officially launched soon.“”……
Youdao Small P App is currently connected to DeepSeek Source: Netease Youdao
At that time, according to Netease Youdao’s official public account, Netease Youdao CEO Zhou Feng said that the maturity of the science model will greatly expand the application of AI in education and other industries. Last year, we predicted that the science model would mature rapidly, and actual technological innovation would come faster. DeepSeek -R1 was born in just a few months. This is very critical for educational applications. It can be said that the biggest scenarios and the most difficult problems in the teaching process rely on the capabilities of the science model, so we are very optimistic about the application of R1 and other reasoning models. rdquo;
It is worth mentioning that Netease Youdao’s internal test data shows that the accuracy rate of DeepSeek-R1 on Youdao’s K12 test set is as high as 88%. This outstanding performance further strengthens our determination to fully enable this model. rdquo;
DeepSeek-R1 has an accuracy rate of 88% on the Youdao k12 test set. Source: Netease Youdao
The next day, Yunxuetang also announced that all products in the series had been fully connected to the DeepSeek-R1 and DeepSeek-V3 models. Recently, Yunxuetang’s AI course making experts, AI learning maps, AI learning experts, AI pairs and other products can use a variety of large model capabilities including DeepSeek to provide intelligent services to customers.
Not only that, Learning and Thinking also recently announced that its intelligent education hardware products such as learning machines and learning machines will be connected to the DeepSeek artificial intelligence model. By embedding a deep thinking model, users ‘AI learning experience will be comprehensively upgraded and intelligent guidance will be redefined. Coaching scenarios. The new functions are expected to be launched on relevant models within February.
For example, when a child encounters a problem and calls Xiaosi, the intelligent assistant for learning and thinking, Xiaosi will combine large models such as DeepSeek and Nine Chapters to not only give answers suitable for the child, but also show the process of in-depth thinking like DeepSeek. The person in charge of Xueersi technology once revealed to the media that the team is exploring how to better incorporate outstanding industry models such as DeepSeek into the ecosystem of learning machines and expand Xiaosi’s circle of friends.
On February 9, the Yunxin vertical model under Zhonggong Education also completed the privatization deployment of the DeepSeek series of models. It is understood that through this deployment, the application logic of AI in scenarios such as content development, intelligent correction, and personalized answering has been optimized.
Also,Education companies such as Seewo, Yi Xiaoxiang, Listening Bear, and Shulang have also announced their embrace of DeepSeek.
This seems to be a major trend that cannot be ignored.
Recently, nearly a hundred brokerage research reports related to DeepSeek have been released. Many of these research reports also mentioned the possibility of DeepSeek and education. For example, Guotai Junan mentioned that DeepSeek’s exit will benefit e-commerce, education, and smart glasses, AI+ e-commerce, AI+ smart glasses, AI+ education, AI+ other directions, etc.
Obviously, education companies embrace DeepSeek mostly to take advantage of the momentum. Just as learning and thinking said about expanding the circle of friends, Duan Yitao, chief scientist of Netease Youdao, also mentioned in an interview with Douzhi.com that embracing DeepSeek-R1 ‘s new technology will help it build Ziyue into a more powerful educational model.
Is education hardware saved?
In fact, before embracing DeepSeek, education companies had been working on the AI model for a long time.
As early as 2023, we will develop the nine-chapter model in the future, becoming the first education technology company in China to use the mathematical reasoning model, and then expand into a full-discipline education vertical model; in the same year, Netease said that it was the first to launch the country’s first education model Ziyue; Homework Group also independently developed Homework Group Model, and ranked first in the multi-task test lists of the two authoritative lists, C-Eval and CMMLU……
Moreover, looking at the current learning machine market, almost all manufacturers have launched AI learning machines supported by self-developed large models. Among them are iFlytek, Youdao, Homework Help, Xiaodu, etc., all of which apply large models to learning machines of leading products, such as AI composition assistants, AI spoken grading exercises, AI mathematics search and answering tools, etc.
Photo source: Sandalwood
On the other side is huge market prospects. According to a research report by Ai Media Consulting, the overall scale of my country’s educational intelligent hardware market is on the rise and is expected to exceed 100 billion yuan by 2025.
ButIn terms of product quality and technical stability, AI+ education hardware is uneven.
Previously, according to Red Star News, many consumers reported that there were many problems with the 12.35-inch Xueersi flagship learning machine they purchased. For example, only 5 of the 9-fold eye protection certifications have been certified. AI correction operations can not be identified as mistakes. Don’t judge the correct answer as wrong answers. Math questions give English answers over-the-top answers or answer the wrong answers.
Regarding the current development of domestic education hardware, Zinc Scale also asked DeepSeek, and the answer was: From the perspective of development status, domestic leading companies (such as iFlytek, Netease Youdao, Homework, etc.) are based on self-research or optimization. AI models (such as Spark Model, Ziyue Education Model) have launched hardware products such as intelligent learning machines, dictionary pens, and AI exercise books to support multi-disciplinary tutoring, composition correction, and spoken dialogue. The introduction of large models has significantly improved the interactive capabilities of educational hardware, such as realizing real-life teachers ‘answering questions and problem-solving guidance through natural language processing.
Moreover, driven by policies, educational hardware has become an important alternative to off-campus training, and the market size is expected to exceed 100 billion yuan in 2024. rdquo;
Photo source: DeepSeek
However, DeepSeek also mentioned that the current core shortcomings of domestic education hardware include technical limitations. For example, large models have limited logical analysis capabilities in complex disciplines (such as mathematical reasoning and physical experiments) and may generate wrong answers; in-depth evaluation of long texts (such as compositions) still relies on manual intervention. Moreover, most hardware functions are similar (such as question bank and voice assistant) and lack differentiated innovation. Some enterprises rely on hardware + paid content models, with insufficient technical barriers. rdquo;
So, when major education companies begin to develop large models on their own and embrace DeepSeek with the other, can the AI strength of educational hardware be improved to a higher level?
“Relying on its understanding of the education industry and the accumulation of product user data, Youdao has previously developed its own reasoning model in the education field. Open source technology will help us make better use of these resource advantages and reduce our trial and error costs, so as to customize and optimize our own models according to our own needs, and build a solid moat in the field of education. rdquo; There is a public account in Netease, and Zhou Feng seems very optimistic about this. He emphasized that high-quality open source and low-cost models will inevitably bring great prosperity to AI.
DeepSeek gives a more specific answer: the big education model (such as subject knowledge questions and answers, personalized learning recommendation system) and DeepSeek (Assuming it is efficient search or knowledge mapping technology) After combining, educational resource management can be optimized and knowledge retrieval can be accurately carried out in the following ways: DeepSeek’s rapid indexing capability can help the education model retrieve textbooks, papers, question bank and other resources in real time to improve the accuracy and richness of answering questions; dynamic knowledge mapping: By building a subject knowledge association network through DeepSeek, the education model can more efficiently understand the logical relationships between knowledge points, such as automatically associating relevant formulas and examples in mathematical problem solving.
However, when DeepSeek became standard, major education companies seemed to have stepped into the same river again. As DeepSeek reminds, avoiding AI for AI’s sake may become a more critical proposition.
