Caixin Weekly | Examining the State of “AI Teachers” (AI Translation)

财新周刊｜摸底“AI老师”

Paragraph 1-2: Since 2025, subject education has emerged as a highly practical and fast-implementing scenario for AI, with "AI teachers" offering free services such as essay corrections, math explanations, and English practice across K12, higher, and adult education. Solutions like AI agents even help students fill in college entrance exam (gaokao) applications. The combination of AI’s content generation and reasoning capabilities has created rapid commercial value, as seen in the MiniMax-Gaotu partnership’s AI English class (featuring a digital likeness of celebrity Daniel Wu), which sold 20,000 units at 398 yuan each on its launch night. Similarly, Stanford students’ Agent-based VideoTutor solution raised $1 million shortly after launch. [para. 1] Model improvements in reasoning have elevated AI's ability to assist with subject Q&A, with many education companies integrating advanced reasoning models like DeepSeek since February, aligning with long-advocated "Socratic" teaching methods. This has led to a general rebranding of educational apps to emphasize their AI foundations. [para. 2]

Paragraph 3-5: Both internet-native and traditional educational service providers have adopted the AI teacher form, pushing custom, precise summer products featuring AI-powered personalization in 2025. ByteDance’s cloud engine Volcengine reported a 12x increase in K12 education token consumption from December 2024 to May 2025, indicating the flood of educational AI app usage. QuestMobile estimated monthly active users (MAU) in the AI educational space at 22.18 million in May 2025, up 1.6% month-on-month. ByteDance’s domestic and overseas AI tools, such as Doubao AiXue (MAU >7 million) and Gauth, are leading user engagement. Despite policy constraints (post-“Double Reduction”), tech companies continue to push AI educational apps, with about 20% of queries from K12 subject questions on major AI chatbots. Major tech companies have promoted AI-powered application features for university entrance. [para. 3][para. 4][para. 5][para. 6][para. 7]

Paragraph 6-8: A primary motivation for edtech companies to deploy AI lies in automating repetitive labor—reducing teaching costs and boosting product price competitiveness. For example, after implementing AI for online feedback, the cap on students managed per teacher at Yuanfudao rose from 150 to 800, saving 4-5x staffing costs for equivalent growth. Generative AI’s evolution is gradually educating users and promoting AI as an everyday educational tool—enabling real-time interactive learning machines and the emergence of "AI one-on-one" tutoring scenarios. However, integrating AI into lessons and products is an ongoing process, and reactions from parents show that the replacement of real teachers by AI sometimes leads to dissatisfaction if value is perceived as diminished. [para. 8][para. 9]

Paragraph 9-12: AI’s growing power boosts the digitalization of education, but challenges remain. Issues include differences between personalized learning and test outcomes, and concerns that easy access to AI-generated answers may impair students’ initiative and creativity. The Ministry of Education has issued guidelines warning students to use AI critically, not copy AI outputs for assignments, and avoid cheating. During the 2025 gaokao, major domestic AI assistants proactively blocked gaokao-related topics. Meanwhile, companies like ByteDance (Doubao 1.6), DeepSeek, and Tencent’s Yuanbao lead the growing AI educational app market. ByteDance’s Doubao AiXue, evolving from earlier educational products, now supports multi-question photo-based Q&A and has quickly surpassed similar offerings by competitors. Similar global products—like ByteDance's Gauth and Zuoyebang’s Question AI—have topped the US App Store, reflecting worldwide need for instant Q&A via photos. However, regulatory hurdles in China and involvements like the US government’s action against TikTok have affected global reach. [para. 10][para. 11][para. 12][para. 13][para. 14][para. 15][para. 16]

Paragraph 13-19: AI-driven "teacher" products, including high-fidelity digital instructors, interactive classrooms, and personalized AI assistants, are quickly becoming differentiated selling points among educational companies. The new wave of AI courses, from large companies to startups, stresses affordability, responsiveness, and customization. However, real-life challenges in mimicking the impact of charismatic live teachers remain. Product differences are emerging mainly in the integration of teacher behavior, content interaction, and question handling. Most AI courses still focus on video-based Q&A and progress monitoring, rather than fundamentally transforming pedagogy. Building robust question banks and accurate reasoning remain core solutions for high-quality AI educational products. [para. 17][para. 18][para. 19][para. 20][para. 21][para. 22][para. 23][para. 24][para. 25][para. 26]

Paragraph 20-24: Advances in large AI models mean most can now score above 130 on China’s gaokao math section—a major leap from earlier performances. For example, ByteDance’s Doubao AiXue scored 144 on one national math test and full marks on another. Nevertheless, actual product reliability remains an issue: correct answer rates average 80-90%, with frequent mismatches in questions sought by students. Some educational companies are developing their own models or optimizing open-source platforms like DeepSeek or Alibaba’s Qwen, striving for specialization and data privacy, but high hardware costs limit broader adoption. Machine assessment is most advanced in English, with large-scale automated grading now highly accurate, while math and science essay grading remains problematic due to complexity. [para. 27][para. 28][para. 29][para. 30][para. 31][para. 32][para. 33]

Paragraph 25-29: Standardization in China’s K12 exams and the abundance of learning resources prompt ongoing debates about the value of personalized learning guided by AI. Effective education requires more than just topic coverage; interest-based, personalized delivery is vital. New AI agents that can synthesize content in diverse forms (e.g., thematic video lessons or using popular culture contexts) are emerging. However, “pain points” like personalized college counselling remain—most students and families still resort to self-research due to the high cost (over 10,000 yuan) of professional planning services, which only about 2% of students access. Companies like Alibaba and Tencent are leveraging their browsers and AI to offer free, inclusive college counselling tools, with over 10 million reports generated in 2025, primarily serving users in smaller cities. Recent Ministry of Education guidelines have emphasized equal access and the accommodation of diverse needs—a major goal for the continued integration of AI in China’s education ecosystem. [para. 34][para. 35][para. 36][para. 37][para. 38][para. 39]

[para. 1]