As we all know, the concept of large models in the first half of 2023 is hot. Major technology companies have been involved one after another, or declared their layout in related fields, or took the lead in announcing their self-developed large models. As the capital market's enthusiasm for the concept of large-scale models has subsided, a new round of competition belonging to this battlefield has gradually focused on the application of technology.
On August 15, HKUST Xunfei officially launched Xunfei Spark Cognitive Model V2.0. Soon, the "MIT Technology Review" magazine tested the latest versions of the four major Chinese models of Baidu Wenxin Yiyan, Xunfei Xinghuo, SenseTime, and Ali Tongyi Qianwen. Xunfei Spark Cognitive Model V2.0 ranked first in six of the eight first-level categories, and reached the top of the evaluation with a score of 81.5 points (percentage system), and won the "smartest" award Domestic large-scale model title.
It is worth noting that the focus of this news is not only on the upgrade and iteration of the large model itself, but more attention is paid to the implementation of the combination of Spark and various usage scenarios, especially in the field of education. In other words, "big models change the world" has become a consensus, while "where the world will start to change" and "how will it change" is the focus. As a result, the HKUST Xunfei learning machine, which has quickly tasted the sweetness of sales under the blessing of large models, has gradually come to the center of the topic.
The big model and the learning machine are rushing in both directions, and the revolution of intelligent learning equipment has started brilliantly
In three months, Xunfei has opened up a new situation in the combination of large models and learning machines, and the one-to-one interaction function has been updated to seven items.
- On May 6, Xunfei released the T20 series, the first cognitive large-scale model AI learning machine, and launched three new AI-one-to-one functions based on large-scale models: one-to-one Chinese and English composition correction, mathematics interactive supplementary learning, and talk talk one-to-one One-to-one oral English training.
- On July 6, the LUMIE 10 series was launched, adding two new functions equipped with large models - AI one-to-one encyclopedia free question and answer and AI one-to-one parent-child education assistant.
- On August 15, Xunfei announced at the V2.0 release conference of the Xinghuo Cognitive Large Model that the learning machine using the new version of the large model will expand its application again, adding AI one-to-one intelligent programming assistant and AI one-to-one creative painting partner . Focus on improving the code generation ability, semantic understanding ability and image understanding ability based on large models.
Putting the large model into the learning machine, there is no acclimatization, but an instant hit, breaking the long-standing homogeneous competition in the field of intelligent education hardware.
Since the learning machine market was developed in the 1980s and 1990s, major brands have competed in channels, content, and intelligence for more than 30 years. At the beginning of 2023, the field of learning machines has formed a highly homogeneous competition situation. Zhang Yi, chief analyst of iiMedia Consulting, also said, "The homogeneity of educational intelligent hardware products is relatively serious, and their practicability is relatively weak, and the product concept is better than its practicability." In order to consolidate their market share, each brand of learning machines had to "open the book" in terms of hardware performance.
As far as Xiaodu and HKUST Xunfei learning machines, which are the top two in terms of sales, Xunfei launched a new model T20 Pro in early 2023. Its upgrade focuses on the screen, sensor, and camera. It is also equipped with a learning machine with the largest battery in the industry. Xiaodu launched the new learning tablet Z20 Pro and Z20 Plus at the beginning of the year, focusing on the portability advantages that are different from the home scene. The new models weigh only 0.92 kg, which is nearly 2/3 lower than the 2.6 kg of the P20. In fact, according to sales data, hardware upgrades did not have a significant impact on user purchase behavior. At this time, the learning machine market is still as calm as before, however, a technological innovation is quietly approaching.
In May, Xunfei released the large model of Xinghuo, which was officially launched on its flagship product, which immediately sparked an upsurge in popularity.
According to the semi-annual report of HKUST Xunfei, after the release of the Spark model, the GMV of Xunfei AI learning machine increased by 136% and 217% respectively from May to June. At the same time, the sales data of Jingdong learning machine also proves the great attraction of large models. In the first five months of 2023, iFlytek’s market share was 16.1%, and with the release and application of the large-scale spark model in May, iFLYTEK’s market share on the Jingdong platform soared to 23.6% in May. An increase of 5.4 percentage points, ranking first in the industry. The T20 and T20 Pro based on the Xinghuo Cognitive Model have become the main products, accounting for 61% of Xunfei’s learning machine sales that month.
The success of large-scale learning machines is not an ordinary encounter in the AI era, but a two-way rush based on technical characteristics and educational needs. The revolution of intelligent learning equipment will start brilliantly.
The biggest key point that triggers market pursuit and industry wind direction turning point is the extremely high matching degree between technical characteristics and industry demand.
One of the subversive features of the large model is the upgrade of the natural language interaction experience. The education industry is based on language interaction and the interaction effect is the core competitiveness. It is an excellent scenario for technology implementation. The previous products plunged into the sea of questions and got stuck on being unable to provide better interaction and personalized in-depth suggestions, so they fell into a whirlpool of homogeneity.
The current mainstream learning machine is mainly sold with a rich content knowledge base, a large number of free educational resources, and high-quality expert private lessons, and on this basis, it presents diversified content. For example, brands such as homework help and Xueersi, which take rich teaching resources as their core competitiveness, can scan questions and student answers, correct homework based on their own resource library, and provide original question explanation videos. The one-to-one diagnosis and planning function promoted by the brand is actually to call the appropriate teaching resource library according to the accuracy of students' answers. The correct rate of city students doing targeted exercises only played a role in the targeted transfer of resources.
In a word, the learning machine before the large model is added, even if it has a certain degree of artificial intelligence recognition, search, recommendation and other functions, it can still be summarized as a "think tank" learning machine in essence. When the mass content in the library becomes more and more saturated, the market will have phenomena such as volume weight, battery capacity, and screen size. Although the learning machine market seems to be "unrecordable", in fact, the dazzling products have not effectively solved the problems faced by the current education industry, such as resource shortage, uneven distribution, outdated models, and insufficient quality education. The education market under the background of limited education and training industry and changes in the structure of social talent demand is eager for the emergence of a more effective, comprehensive and interactive teaching mode suitable for different students.
The large model learning machine happens to be at this node, giving a new direction of interaction, satisfying education's desire for individualized learning and personalized learning, and assisting in the change of content from quantity to quality.
Xunfei ignited the fire of general-purpose large-scale models and learning machines, and the revolution of intelligent learning equipment has just begun.
A single spark can start a prairie fire, AI+ applications explode, and the market structure will usher in a rewrite
The China Big Language Model Industry Development Report disclosed by iResearch in April asserted that whoever owns the value, ecology and traffic entry of the basic big language model is more likely to have revenue from the application layer to the computing power layer The right to speak. The same is true in the field of intelligent education hardware. The large model has become a battleground for military strategists and an indispensable moat for intelligent education hardware companies in the future.
The two-way rush between Xunfei Spark and learning machines is just the beginning. Other learning machine brands and education brands are also increasing their efforts. AI+education is an irreversible trend.
Xunfei is not the only company that has the advantage of large models in the learning hardware track. Baidu is sitting on two important achievements of Wenxin Yiyan and Lingji, and released the "Qinghe Mobile Phone" equipped with Lingji in May this year, trying to find another way in the crowded tracks of mobile phones and learning machines. At the same time, this learning mobile phone is priced at 1499, which is a gap of nearly 7,000 yuan compared with iFLYTEK high-end learning machines. The large model has the potential to enter thousands of households with a friendly attitude.
Other learning machine brands that are inferior to Baidu and Xunfei on the basis of general-purpose large models, that is, old opponents who have been recruiting on the educational hardware track for many years, also frequently announced related exploration actions from May to August this year, trying to use It has been rooted in the advantages of the education industry for many years, explores vertical large-scale models, and creates differentiated advantages. For example, Netease Youdao, which is based on language, demonstrated the application of AI spoken language teacher, professional English writing correction and guidance based on the large model of "Ziyue"; Break through the "partial subject" phenomenon that the general model is biased towards reading and writing, but the reasoning and logic ability is weak.
It is not difficult to predict that smart education hardware companies will usher in a strong integration and differentiation revolution, and the balance of software + content will become the core of product competition.
As far as all kinds of learning machine products are concerned, the hardware conditions are relatively balanced, and the advantages of battery power and portability cannot be the main incentives for purchasing decisions. Large models have opened up a new direction for learning machine software competition, but high-quality educational content is still one of the main motivations for parents to buy smart educational hardware.
Contrary to the fact that users attach importance to both product functions and educational content, learning machine products cannot give good consideration to this, and consumers still need to make a trade-off between the two aspects. Many relatively comprehensive and pertinent evaluations from various major self-media have given such as "Xiaodu is highly interesting, but the course content is mostly public resources, unsystematic, and needs to be screened by parents"; However, the explanations for the wrong questions are not detailed, and sometimes even only the answers are given without explanations”; “There are many courses in learning and thinking, but you can’t write by hand, and the smart correction and interactive experience are poor” and other conclusions.
However, major brands have also sensed the plight of consumers with a keen sense of smell, and have begun to make up for their shortcomings. For example, Baidu cooperates with Gauss to launch the small learning machine Z20 Plus Gauss Deluxe Edition, seeking opportunities to stand out with high-end courses. Xueersi, NetEase Youdao, and Homework are also laying out large models, trying to fill in their own shortcomings in intelligent interaction. plate.
The big model hits a thousand waves with one stone. The competition logic of future learning machines is not standardization and cost reduction and efficiency increase, but on the basis of both software and content, seeking differentiated advantages according to their own characteristics, and attracting more precise customers. At the same time, the change in the direction of competition will inevitably bring about a strong integration of the market.
According to the research report of the institution, the market share of learning machines in 2023 will tend to concentrate on Baidu, Xunfei and other heads. In 2021, the CR5 of the domestic learning machine market will be 48%, and in the first five months of 2023, the CR5 of the Jingdong e-commerce platform has reached 70.1%. Led by users' higher expectations for intelligence and demand for personalized content, the market concentration of learning machines will continue to remain high, and players with slow progress in AI technology and lack of unique content advantages will be strongly eliminated.
AI+education, where will it blow after the wind blows? AI is not the goal. Let education keep up with the times, adapt to the quietly changing production factor market and talent demand trends, and pay attention to the cultivation and development of students' personality is the ultimate destination of learning products.
Education is a very special field. Its core is for human development and natural growth, and it also undertakes the task of cultivating professionals needed by the times for all walks of life in society. Whether learning equipment is hardware upgrades, content library updates, or reshaping the interaction mode with large models, blindly following the trend may be able to take advantage of the anxiety of parents to gain temporary results, but it cannot develop sustainably in the long run.
In the era of AGI, the most important thing for learning equipment is not leading technology or rich question bank, but clarifying the major changes of artificial intelligence technology to the personal ability evaluation system, so as to update the educational concept and teaching goals, and cultivate students who are better at independent thinking. , Independent innovation, students who use AI tools. The underlying logic of the relationship between artificial intelligence as a productivity tool and people is to change the individual ability evaluation system to generate human value stratification. Students who lack relevant qualities and thinking will not be able to gain advantages in the new system.
The "Hundred Models War" will continue to create waves, and the revolution of learning machines and related intelligent educational equipment has begun. In the context of the average development level of related hardware, providing users with an interactive experience that is more suitable for multi-dimensional talent training needs, and pursuing the balanced development of software + content will become the new standard of AI + educational hardware.
Source: Pinecon Smart