Amid Continuous Losses in Large Models, Companies Turn to AI Learning Devices
-
At the 2023 Baidu World Conference held on October 17, Baidu introduced the Xiaodu Qinghe All-in-One Learning Machine, which combines large model technology, teaching methodologies, and personalized education, aiming to serve as a one-on-one AI home tutor for children.
The following day, Alibaba's Tmall Genie also released the Z20 True Smart Eye-Protecting Learning Tablet, equipped with large model and multimodal AI perception capabilities, supporting both 'precision reinforcement' and 'independent exploration' learning modes.
The consecutive launches of AI learning devices by Baidu and Alibaba are no coincidence. With technological maturity and strong policy support, educational informatization is becoming a new trend in the education sector. The large model technology actively pursued by tech companies aligns perfectly with the demands of this era of educational digitization.
As the Double 11 shopping festival approaches, learning devices have gained unprecedented popularity through influencer marketing. On October 28, 2023, 'Peking University prodigy' Liu Yuanyuan promoted the Xueersi Learning Tablet on Douyin at a price of 5,199 yuan, which sold out before she could finish her presentation.
However, it's worth noting that, overall, due to their high cost and the lack of widespread parental recognition of their educational value, the penetration rate of general AI devices remains low. This may also suggest that AI learning devices might not be the ideal terminal for tech companies to achieve a commercial closed-loop with large models.
The reason tech companies are 'crossing industries' to enter the education sector is largely due to the rapid development of educational informatization, driven by technological advancements and policy support.
Currently, the biggest problem in China's education industry is the unequal distribution of educational resources against the backdrop of varying regional economic development levels.
Education informatization, driven by mature technologies like the internet and AI, offers advantages such as breaking through time and space constraints, rapid replication and dissemination, and diverse presentation methods. These can promote educational equity and improve education quality.
For example, due to limited educational resources, traditional education models may struggle to accommodate students with poverty, physical disabilities, or reading difficulties. In contrast, scalable AI technology can provide personalized assistance to these students at low cost, thereby bridging the educational gap.
Recognizing these advantages of education informatization, relevant authorities are actively promoting the development of related industries. In 2018, the Ministry of Education issued the "Education Informatization 2.0 Action Plan," which states: "In response to the development of information technology, especially intelligent technology, we will actively promote 'Internet + Education,' adhere to the core concept of deep integration of information technology and education, and uphold the basic policy of application-driven and mechanism innovation to establish a sustainable development mechanism for education informatization."
Similarly, the State Council's "New Generation Artificial Intelligence Development Plan" further clarifies: "Using intelligent technology to accelerate the reform of talent training models and teaching methods, and to build a new education system that includes intelligent learning and interactive learning."
On one hand, the education informatization industry can achieve educational equity; on the other hand, relevant authorities are vigorously promoting the development of the education informatization industry, giving the sector immense potential for growth.
Data from the China Industrial Research Institute shows that from 2016 to 2021, China's education informatization market grew from 294.7 billion yuan to 472.4 billion yuan, with a compound annual growth rate of 9.9%. It is projected to reach 557.3 billion yuan in 2023, an 8.28% year-on-year increase.
Given their presence in the informatization sector, many internet and tech companies have already begun actively expanding into education informatization. As early as 2018, Baidu launched Baidu Education Brain 3.0, leveraging AI, big data, and cloud computing to enhance educational products and scenarios. During the pandemic, Alibaba's DingTalk Education service met the demand for "suspended classes, ongoing learning."
Since 2023, with the rise of ChatGPT, numerous tech companies have aggressively pursued AI large model products. On March 16, 2023, Baidu released its large language model and generative AI product, "Wenxin Yiyan." A month later, Alibaba introduced its own large-scale language model, "Tongyi Qianwen."
In May 2023, data from the China Science and Technology Information Research Institute revealed that 79 large models had been released domestically, marking a "hundred-model battle."
While tech companies are racing to adopt new technologies, the high training costs of large models have left most related businesses struggling with losses.
OpenAI's disclosed data shows that GPT-3's knowledge comes from a training corpus of 300 billion words. Guosheng Securities estimates that training GPT-3 once costs approximately $1.4 million. According to Fortune, OpenAI reported a loss of $545 million in 2022.
Not just OpenAI, The Wall Street Journal reported that Microsoft's GitHub Copilot, one of its first generative AI products, is also deeply mired in losses. The service charges users $10 per month but incurs an average loss of $20 per user monthly, with some users even costing up to $80.
Although domestic large model companies have not disclosed detailed financial data for their large model businesses, under current circumstances, most companies are likely struggling to achieve profitability. At the Qualcomm Summit on October 25, 2023, Honor CEO Zhao Ming commented, "No one has claimed that large-scale network models are profitable yet, as the computational power consumption is still too high."
In this context, many tech companies naturally need to continuously explore cutting-edge business models to establish a commercial closed-loop for large models. For example, Huawei's Pangu model focuses on industries like government affairs, finance, and manufacturing, while Baidu's ERNIE Bot actively collaborates with automakers, integrating into vehicles from Changan, Geely, and VOYAH.
Given that some internet companies have accumulated experience in the education informatization sector, educational hardware has naturally become a key component for implementing AI large models. Consequently, companies like Baidu and Alibaba have recently been rolling out AI-powered learning devices.
Despite strong policy support for the education informatization industry, the market performance of AI-related terminals has not been particularly impressive.
According to iResearch data, China's online education market reached 582.5 billion yuan in 2022, while the market size of general AI products was only 21.11 billion yuan, with a penetration rate of merely 3.6%.
Against this backdrop, AI-related terminal businesses of some education companies have struggled to achieve impressive performance. For example, Readboy reported revenue of 126 million yuan in the first half of 2023, a 52% year-on-year decline, with revenue from student personal tablets dropping 55% to 104 million yuan.
Similarly, in the first half of 2023, iFlytek's revenue fell 2.26% to 7.842 billion yuan, with education products and services revenue growing only 3.63% to 2.285 billion yuan.
The low penetration of AI products in China's education sector is attributed to two main factors: the high prices of related products, and the fact that the value of AI in teaching has not been widely recognized by parents, resulting in weak willingness to pay directly.
Taking the Xiaodu Qinghe All-in-One Learning Machine as an example, its retail price is as high as 9,999 yuan. Such a high cost is not attractive compared to traditional tutoring, especially since the effectiveness of AI learning machines has not been strongly validated in the market, making it difficult for parents to accept these products.
In fact, due to the unfavorable feedback from the consumer market, some education companies have begun exploring alternative approaches, hoping to improve the utilization of their AI learning machines through AI smart study rooms.
On one hand, AI smart study rooms represent a "resource integration" by education companies after multiple rounds of technological development and market validation in areas such as smart hardware devices and education informatization. Concepts like "precision learning and improvement" only began to take shape a few years ago, and the development of AI and big data technologies has gradually given AI-assisted learning some market competitiveness. The final implementation in AI smart study rooms allows students to learn with AI assistance without the need for human teachers.
On the other hand, as off-campus academic tutoring faces strict limitations under the 'Double Reduction' policy, the demand for supplementary education remains strong among students (C-end users), creating a surge in alternative solutions. AI-powered learning labs have emerged as a substitute for traditional tutoring centers, serving as a transitional space for institutions adapting to the new regulatory environment.
On August 9, 2023, Readboy held a franchise conference for its AI Learning Lab project, with 35 operational centers nationwide signing agreements. These labs provide physical learning spaces where students can use Readboy tablets to access personalized learning systems.
While the commercial prospects of AI learning labs remain uncertain, one thing is clear: purely AI-driven learning devices currently struggle to form a complete commercial ecosystem.
Tech companies' simultaneous push into AI learning devices is no coincidence. On one hand, they recognize the growing trend of educational digitization, where their large language model technologies align with current needs. On the other hand, the financial losses from large model development push these companies to seek viable business models, making AI learning devices an emerging hotspot in education technology.
However, market acceptance of AI learning devices faces limitations due to pricing and public awareness, forcing industry players to continuously explore alternative business models for product adoption. This suggests that tech companies' hopes of using AI learning devices to create a commercial ecosystem for large models may not proceed as smoothly as envisioned.