Chen Jialan, a reporter from this newspaper, reported from Guangzhou
The AIGC (Generative AI) craze brought about by ChatGPT is sweeping through all walks of life.
The smartphone market has declined for many consecutive quarters and has not seen obvious signs of recovery, but Baidu has entered the mobile phone to attract attention from the outside world.
In fact, Baidu's first mobile phone is a mobile phone equipped with AI learning large models. Previously, iFLYTEK has landed its Spark model in the iFLYTEK AI learning machine T20 Pro. Xueersi also said that it is conducting research and development of self-developed mathematical large models, which is expected to be launched within this year, and its learning opportunities will launch an "AI assistant", covering composition assistant, speaking assistant, reading assistant, mathematics assistant and other related functions. Recently, there is also the first learning machine that supports children's version of ChatGPT in China, the dolphin Xiaomeng bilingual learning machine T10, which allows children to communicate with them through voice, text, pictures and other ways.
The reporter of "China Business News" noted that whether Baidu made mobile phones did not choose the mass consumption track, or iFLYTEK and other companies intend to open up the education market through large models, especially in educational intelligent hardware products such as learning machines, the wind of AI large model + education is rising.
So, what sparks can the crazy AI big model and learning products make? Who players are already in the integration of AI models and learning products? How will AI big models empower learning products? The reporter contacted and interviewed many industry insiders.
Giants have laid out
On May 17, Baidu's Xiaodu announced that it will launch a new brand Xiaodu Qinghe, and on May 22, it released its first new species product Xiaodu Qinghe learning mobile phone.
According to reports, Xiaodu Qinghe learning mobile phone is equipped with Xiaodu Lingji large model, providing AI photo lectures, Chinese and English AI composition guidance, AI speaking practice and other learning guidance; It also has a customized mobile phone system for teenagers, free learning resources for elementary schools, and the ability of parental application control, real-time location, and vision protection. The mobile phone is currently only available in a "6+128GB" version, with a pre-sale price of 1499 yuan.
Jing Kun, vice president of Baidu Group, mentioned the combination of large models and mobile phones many times in an interview with reporters, such as Xiaodu mobile phone has a large model-based operation assistant, with the blessing of large models, its accuracy of modifying the operation is much higher than the original. He also imagined that the AI model may become a language tutor next to every primary school student, and the child uses the AI mobile phone to take pictures of his own essays, and the AI can teach him how to write a better article.
Although Baidu launched a smartphone aimed at the student population, its competitors are mainly educational smart hardware products.
On May 6, iFLYTEK officially released the iFLYTEK Spark cognitive large model and application results, and took the lead in applying its large model to the educational intelligent hardware AI learning machine T20 Pro, which not only realizes the ability to refine the correction and personalized comments of Chinese and English compositions like a real teacher, guide children to improve their writing level, but also supports students to directly communicate with the learning machine for English chat, mathematics personalized and accurate human-like interactive auxiliary learning and other functions.
Xueersi is carrying out the research and development of self-developed mathematical large model, named MathGPT, for global mathematics enthusiasts and scientific research institutions, with problem-solving and lecture algorithms in the field of mathematics as the core, has achieved phased results, and will launch product-level applications based on the self-developed large model within this year. Xueersi said that its learning machine will soon launch an "AI assistant", covering composition assistant, speaking assistant, reading assistant, math assistant and other related functions.
At the same time, NetEase Youdao recently officially released a spoiler video of AI speaking teachers developed based on the "Ziyue" large model, NetEase Youdao AI speaking teachers can provide a variety of practice scenarios, and will play a variety of roles according to user needs, and then guide users to have multiple rounds of dialogue. It is said that NetEase Youdao is also developing the application of subdivided learning scenarios Chinese composition grading.
It can be seen that in the current popularity of generative AI based on large models, there are many companies like Baidu and iFLYTEK that seek new increments, and they hope to introduce AIGC technology into their own products, in order to improve the interactive experience of education products and enable products to serve users in a wider range of scenarios.
The reporter noted that the educational intelligent hardware products represented by the learning machine have become the carrier of the giants' concentrated landing of AIGC technology in the field of education. Some analysts pointed out that AIGC is expected to bring about revolutionary changes in educational resources, education systems and teaching methods.
"AIGC technology will provide a broad market space for traditional educational software and hardware tracks." Xu Ying, an analyst at Canalys China, told reporters that the market-oriented development of learning machines in the direction of intelligence and human-computer interaction is indeed a more feasible commercial application scenario, and there should be broad prospects in education.
Can it stir up the educational hardware track?
At present, the mainstream suppliers of intelligent hardware products related to education include traditional education hardware enterprises represented by BBK, Yoshulang, Youxuepai, etc., as well as new technology enterprises represented by iFLYTEK and Baidu. Since the implementation of the "double reduction" policy, relevant enterprises have sought business transformation, and educational intelligent hardware products have become a new battlefield. Coupled with the impact of the epidemic, the rapid growth of demand has also promoted the growth of the learning machine market space. After the relevant giants have participated in the layout, the market competition is gradually intensifying.
However, at present, educational intelligent hardware products still tend to be homogenized. Huafu Securities Research pointed out that educational intelligent hardware products are a business ecology for the coordinated development of hardware, software and content. In today's rapid development of the consumer electronics industry, hardware configuration tends to be homogenized, and software and content have become an important standard to measure the advantages and disadvantages of educational intelligent hardware products. Software determines the consumer's experience, content determines the information obtained by the consumer, and the effect of product use is determined by the software and content. The difference in software and content will become the key for brands to break through the barriers of homogenization competition and acquire customers in the future.
Attracting manufacturers to apply AI models to educational products is not only related to the serious homogenization of educational intelligent hardware products and the pursuit of more innovation, but also related to the surge in demand for educational hardware products and the stimulation of market prospects after the "double reduction" policy.
According to the "2022 China Education Intelligent Hardware Market and User Insight Report" released by iResearch, parents still maintain a positive attitude towards educational intelligent hardware products under the background of the impact of the epidemic and consumption downgrade. The proportion of parents spending more than 30% on smart hardware in education will increase from 16.3% to 29.5% in the future. According to data from consulting firm IDC, from the fourth quarter of 2021 to the second quarter of 2022, the cumulative shipment volume of China's learning tablet market reached nearly 4 million units, a year-on-year increase of 18%.
However, with the resumption of offline learning of primary and secondary school students since the beginning of this year, the end of the "epidemic dividend", and the more cautious consumer spending this year, in Xu Ying's view, this year's learning tablet market is difficult to have such a high growth as in previous years, and more is back to the tools to assist after-school learning.
"At present, we do see that some players in the head are actively trying to land AI in educational scenarios and combine it with hardware products to improve the learning efficiency of users by improving the ability of human-computer interaction, but smart tablets, especially those that mainly promote language interaction, require further training, development and application capabilities of manufacturers in terms of language models, and the market effect of AI-assisted tablets remains to be seen." Xu Ying said.
The reporter learned that the application of large models in the learning track also has difficulties in landing.
"Whether manufacturers can show a sustainable development trend in the investment in new technology research and development and later landing, whether the application brought by the large model is intelligent enough, whether the answer to the question is accurate and high-quality, and whether the development ability and maturity of the model are capable of making consumers willing to pay for it are all difficulties for the application of large models in learning scenarios and popularization."
Ding Daoshi, president of the Speedway Research Institute, told reporters that the content services provided by traditional learning machines are generally reviewed and checked by content reviewers, but the generative AI content based on large models is newly generated, and who will check the content, this problem has not yet been solved, which is also one of the difficulties in its landing.
In Ding Daoshi's view, he does not recommend that large models involve the learning field too early, "At present, even the latest GPT-4 large models often have some obvious errors, which are unacceptable for children and students." ”