In recent developments within China's education landscape, New Oriental has unveiled its Harmony OS native version of the New Oriental Online App, alongside an intelligent study system powered by large model technology, dubbed the "Smart Learning System." This launch marks a significant milestone as the last among major education firms in China, with a market capitalization exceeding 80 billion yuan (approximately $12 billion), to introduce a large model product.

Interestingly, New Oriental’s initial large model offering focuses on postgraduate entrance exam preparation rather than the more prevalent K12 education sectorThis raises the question: Why did New Oriental choose to pioneer the application of large model technology in its postgraduate programs?

New Oriental's entrance into the realm of large models signifies a notable shift, emphasizing the growing impact of technology in the education sector

However, while the era of educational large models is upon us, the industry remains in its infancySeveral challenges loom ahead as companies look to refine and transition their technological advancements.

The newly launched Smart Learning System incorporates features such as automatic grading and question generationHistorically, New Oriental has not been at the forefront of tech-driven educationWith the recent adoption of large model technology, it appears to be playing catch-up in an industry where competitors have already harnessed this capability.

In July 2023, NetEase Youdao set the stage by releasing the first educational large model in China, "Zi Yue." Following suit, several other companies, including Zuoyebang, TAL Education, Seewo, and Yuanfudao, have developed their own models, with several firms producing multiple iterationsAdditionally, many well-established education companies like Zhonggong Education and Daosheng Education have rolled out large model products

Even smaller education firms have ventured into applying large models, demonstrating widespread industry adoption.

New Oriental has already made moves to integrate large model technology into its operationsDuring Alibaba Cloud’s AI Leadership Summit in April 2023, Zhang Chen, product manager at New Oriental’s university division, shared that the company is leveraging large models to analyze staff-student communications for enhancement suggestions, consequently boosting student satisfactionMoreover, they are utilizing this technology to evaluate courses and study materials, breaking down knowledge structures to improve overall classroom efficiency.

This early phase of experimentation primarily focuses on employing large model technology to streamline internal processes and enhance productivityThe introduction of the Smart Learning System now extends these advancements directly to students

For instance, its automatic grading feature caters to all subjects in the postgraduate entrance exam, with a particular emphasis on the grading of English essays.

Writing plays a crucial role in a student's academic performance, making essay evaluations a key requirement during exam preparationThe speed and quality of grading significantly influence student satisfactionAccording to Hong Zhe, General Manager of the Online Domestic Division at New Oriental, educators previously required approximately ten minutes to grade a single mock exam paper, while the new human-computer collaboration can reduce that time to about two minutes.

The system also includes a question-generation feature, addressing the limited availability of real exam questionsGiven the need for ample practice, large model technology can autonomously generate mock exam questions based on core subject knowledge and difficulty levels

alefox

For example, for the compulsory subject of Clinical Medicine, New Oriental's system generated 4,000 questions aligned with 6,000 key knowledge points.

However, the challenge lies in achieving a balance between the accuracy and applicability of the automatically generated questions, avoiding the creation of irrelevant or overly simplistic itemsTo facilitate this, educators must pre-load labeled and stratified knowledge points into the system, enabling dynamic matching of various aspects and difficulty levels for automatic question generation.

One might wonder why New Oriental's first large model product would target the postgraduate entrance exam sectorSeveral factors contribute to this strategic choiceFirstly, this decision aligns with New Oriental's internal business structureThe number of students in non-academic training programs is substantial — approximately 484,000 individuals as of August 31 this year

However, these programs are still in their infancy, characterized by diverse offerings and varying degrees of maturity, posing significant challenges for the large-scale application of AI technologiesIn contrast, preparation for overseas exams like TOEFL and IELTS already has a wealth of mature resources available.

New Oriental's postgraduate exam preparation segment also boasts a large volume of users, with over 400,000 participants this year alone, nearing one-tenth of the expected national enrollment figures by 2025. This considerable user base presents ample opportunities for the application of large model technology.

Additionally, trends in user demographics for postgraduate preparation are evolving, with an increasing variety of studentsAlong with fresh graduates, there has been a significant rise in individuals with prior degrees and those already in the workforce“About 20% of those engaging with New Oriental's postgraduate services are working professionals,” revealed Li Lin, Project Director for the university division

This diverse user base introduces a complex array of demands, given the variations in students’ academic backgrounds, learning habits, and schedulesAI technology is essential in providing customized and personalized support to meet these needs.

The competitive landscape of the postgraduate training sector further underscores the importance of this focusRecent shifts have created a more dynamic market, moving from a predominantly top-heavy structure to one resembling a pyramid, where smaller, agile institutions are rising in prominenceThese "mid-tier institutions" are excelling in specific niches, such as art or MBA preparation, gaining substantial revenues despite a smaller user baseSome even provide training targeted to students aspiring to join specific universities.

While these smaller entities may lack the abundant resources or resilience of larger institutions, they leverage their unique, specialized offerings to establish a competitive edge

Big companies must also pivot towards providing tailored services in vertical markets, utilizing AI technology to remain relevant.

By starting with postgraduate education, New Oriental's foray into large model technology is a reflection of overarching industry trendsComparatively, in the civil service exam training market, major players like Fenbi, Huatu Education, and Zhonggong Education have introduced their models, each deploying them in distinct ways — Fenbi emphasizes question-answered tutoring, Huatu focuses on interview preparation, while Zhonggong integrates digital personas with their teaching assistance.

As one education sector insider noted, "Greater verticality in the application of large models often correlates with increased professionalism and efficiency." This encapsulates a clear trajectory for educational technology as the initial wave of major companies rolls out their large model products.

In the broader context, at the 2024 Artificial Intelligence Large Model Industry Conference, the China Enterprise Evaluation Association and the Zhongguancun Digital Intelligence AI Industry Alliance jointly released the "2024 China Large Model Enterprise TOP50 Development Report." The report indicated that two educational large models, Seewo and Xueersi Jiuzhang, made it into the top fifty list.

Importantly, the report highlighted a significant movement in parameter scales, indicating a shift from models with hundreds of billions of parameters to those reaching trillions

Currently, there are 25 companies boasting parameters exceeding 100 billion, with only seven crossing the trillion mark.

The parameter levels of educational large models cannot yet compete with those of general-purpose models that have already reached the trillion level, emphasizing that “models equate to applications” is the path forward for educational large models.

As the insider elaborated, "Most educational firms don’t need to develop foundational models, akin to how they don’t need to create cloud servicesSuch standardized foundational services can be procured from leading tech giants, allowing education companies to adjust and apply their data atop these bases." However, the transition towards this model will face several hurdles moving forward.

First among these is selecting the right ecosystem“NVIDIA holds an impressive 80% market share in AI chips, which can be attributed to its development of the GPU-based general parallel computing platform and programming model, CUDA, thereby creating a robust software ecosystem.”

New Oriental's choice to deploy its postgraduate large model product within the Harmony OS ecosystem stems from multiple factors, including early investments in Harmony's research and development, product lines developed on the Harmony OS, and a growing number of postgraduate students transitioning to club the Harmony OS on their mobile devices.

Nonetheless, the Harmony ecosystem is still in its infancy

To reach a broader audience, New Oriental's products will need to extend beyond Harmony to larger ecosystems like AndroidThe outcomes of future large model ecosystems will depend on which model emerges as the most widely used can still be determined.

Another crucial challenge lies in cost reductionWhile the "model equals application" business model has been validated within education, finding a balance between investment and return continues to be a work in progressTake the postgraduate entrance exam preparation market—worth around 30 billion yuan annually—only a mere 20% of users are willing to spend over 5,000 yuanCombined with decreasing consumer spending power and a dwindling number of students registering for exams, education companies need to weigh costs and benefits carefully during the development and rollout of large model products.

The education insider predicts a growing tension in this phase of development