Under the educational purpose of teaching students in accordance with their aptitude, in the wave of large-scale models, Netease Youdao has set up a model of vertical large-scale models for the industry by virtue of its in-depth understanding of educational scenarios and rational thinking about commercialization.
On July 26, NetEase Youdao, an educational technology company, held a release conference on the application results of the "powered by Ziyue" educational large model. At the meeting, the first vertical large-scale model "Ziyue" in the field of education in China was launched, and six innovative applications based on the large-scale model of "Ziyue" were released - "LLM translation", "Virtual population language coach", " AI Composition Guidance", "Grammar Lecture", "AI Box" and "Document Q&A".
"Whether a good technology has value and whether it can play a huge role depends on the selection of scenarios and applications and the polishing of details. Through the combination of software, hardware, and AI technology, making high-quality products is what we are doing now." . At the press conference, Netease Youdao CEO Zhou Feng repeatedly emphasized the importance of "scene pull". He said that the application of the "Ziyue" large model in the education industry can not only help students learn better, but also help teachers teach better, so as to realize the educational ideal of teaching students according to their aptitude.
The first large-scale education model in China "Ziyue": Leading the education personalized revolution
As a leading educational technology company in China, NetEase Youdao deeply understands that AI big models are bringing new possibilities to all walks of life and will revolutionize the face of education.
What is the greatest opportunity for education that the advent of large models can bring?
Zhou Feng said at the press conference: "I think it is to help teach students in accordance with their aptitude." According to the introduction, the reason why it is called "Zi Yue" is because Confucius is the sage of education in our country and the founder of the educational concept of teaching students in accordance with their aptitude. 'A master teaches people according to their aptitude', we hope that the master model of Ziyue can move towards such an educational ideal. "
Zhou Feng shared with the audience the three advantages of "teaching students in accordance with their aptitude" of large models. Firstly, the large model can provide students with personalized analysis and guidance; secondly, the large model can realize guided learning, like teachers, ask questions and guide students to explore the answers by themselves; finally, the large model has the ability to integrate general knowledge. By connecting multimodal knowledge bases and integrating knowledge content across disciplines, the large model can meet the dynamic needs of students at any time and help children develop more comprehensive abilities.
Compared with general-purpose large-scale models, the "Ziyue" large-scale model has been positioned as a "scenario-first" educational large-scale model from the very beginning. It can serve as a base model to support many downstream tasks, and provide basic capabilities such as semantic understanding and knowledge expression to all downstream scenarios. Based on this, the Youdao R&D team has designed customized models for different learning scenarios on the basis of the "Ziyue" large model, so as to achieve a high degree of fit between the model and the scenario.
"Our goal is very clear, which is to be driven by actual educational scenarios, and to use technological innovation to facilitate educational innovation." Zhou Feng said that he hopes that the deep integration of technology and products can provide personalized education according to students' different abilities and needs. personalized teaching.
Six innovative achievements were released, and the virtual human language coach was amazed at the scene
At the press conference, Netease Youdao demonstrated the application results of the "Ziyue" large model in multiple scenarios, covering six major education subdivisions, including oral training, composition correction, and answering questions, fully demonstrating the "Ziyue" The technical strength of large models in the field of natural language processing and the broad application prospects in the field of education.
The most eye-catching thing is the artifact of oral English practice in the era of large-scale models-Hi Echo, a virtual human language coach.
At the press conference, the person in charge of Youdao dictionary business conducted several random conversations with Echo. Echo quickly understands the scene and context, and gives immediate feedback. Her facial expressions and mouth shapes are vivid and natural, and her pronunciation is also very authentic. Details such as stress, weak pronunciation, and rising and falling tones are handled very well, and the audience praised her.
During the conversation, Echo can be as persuasive as a real teacher, guide the conversation heuristically, and provide real-time feedback. After the conversation is over, Echo will give suggestions and polish from the perspectives of pronunciation and grammar, which can effectively solve the long-term problems of oral English learners who have nothing to say, don't know where to start, and are afraid of making mistakes.
"When Chinese people speak English, they often face the dilemma of being difficult to speak, afraid to speak, and not knowing where to start. The key lies in the lack of language environment." The person in charge of Youdao Dictionary said that Echo can bring users Come to this kind of "context" that really fits the reality and help them practice oral English better.
Scenario-driven and technology-empowered, Youdao not only focuses on the pain points of users who are afraid to speak English in English learning, but also covers a variety of learning scenarios. With the support of the "Ziyue" large model, it provides users with a variety of efficient learning plan.
For example, when writing English homework, students not only have the need to solve specific problems, but also need to learn to infer other cases from one instance. The "Grammar Lecture" function empowered by the "Ziyue" large model can provide students with targeted ideas and methods for solving problems, and can also recommend similar types of test questions to help students understand the test points in the test syllabus by analogy.
In addition, the "AI Composition Guidance" application launched by Youdao at the press conference not only has the "Composition Correction" function, but also has the "Composition Guidance" function. According to reports, the app aims to solve the problems of "students can't write" and "teachers don't have time to change". For the problems faced by students in the process of writing, before, during and after, such as difficulty in determining the theme of the topic, lack of writing materials, etc., the application can provide guidance and help students "write like a god". In the correction process, AI composition guidance will also provide comprehensive improvement suggestions from the four dimensions of expression, structure, content depth, and emotional richness.
Scenario pulling, "AI+education" technology precipitation and innovation
It is not difficult to find that NetEase Youdao's strategic key word is "scene pull" - relying on the deep cultivation in the field of intelligent learning, it has accumulated a lot of business data and understanding in the field of education. Through in-depth research and analysis of user needs in different scenarios, NetEase Youdao has successfully used the power of large models to create a wealth of solutions in the field of education, such as virtual population language coaches, grammar intensive lectures, and AI writing guidance. This strategy not only reflects NetEase Youdao's in-depth understanding of education scenarios, but also provides users with a more personalized and efficient learning experience.
Why is Youdao able to make the first vertical large-scale model in the education industry and become the first company in China to implement AIGC technology in the education scene? Behind the highlights are years of technology accumulation and product layout.
As early as 2008, Youdao launched the first domestic statistical machine translation online engine independently developed. After 15 years of technological iteration, Youdao Neural Network Translation (NMT) has evolved into the industry-leading "most powerful brain". According to the latest data from QUESTMOBILE, Youdao Dictionary has more than 100 million monthly active users, ranking first in the domestic dictionary translation market.
Since 2016, Youdao has collaborated to build AI basic capabilities, and simultaneously formed teams for language, vision, and sound. Currently, Youdao has accumulated four AI technologies in neural network translation (NMT), computer vision, intelligent voice AI technology, and high-performance computing (HPC). Large underlying technical capabilities.
Since 2017, Youdao has been "going two-way" with the mainstream technology Transformer, unifying AI capabilities under the large model, and paying special attention to the application on the end side. Youdao Dictionary Pen 2 will be equipped with offline Transformer NMT for the first time in 2019. In 2022, Youdao Dictionary Pen P5 will be equipped with self-developed offline ASR, which has also been upgraded to Transformer technology. The continuous empowerment of technology has established the leading position of Youdao learning hardware in the industry.
While the underlying technology is constantly innovating, Youdao is also constantly developing "black technology" in subdivided scenarios. For example, the virtual human language coach has achieved breakthroughs in many technical capabilities such as speech recognition ability, virtual human driving technology, content generation and dialogue ability.
For example, the virtual human language coach has made great innovations in speech recognition capabilities. It supports multilingual streaming low-latency speech recognition technology, allowing Echo to do well in Chinese English, English, and Chinese-English mixed scenarios; acoustic noise reduction, Echo cancellation, automatic voice detection, automatic sentence segmentation and other technologies make it like a real listener and communicator. It can not only judge the beginning of the user's speech, but also allow the user to interrupt at any time and intelligently trigger the follow-up process.
In terms of AI virtual human driving, Youdao, based on the self-developed emotion recognition algorithm and real-time rendering driving engine, conducts in-depth analysis of the played voice data, and drives the virtual human’s facial expressions and voice-synchronized lip changes in real time, making the virtual human It can be close to real people and interact with users one-on-one in a more natural and vivid way, thereby significantly improving the realism of the conversation and user experience.
In the current wave of large-scale models, NetEase Youdao has set a model for the industry with its in-depth understanding of educational scenarios and rational thinking about commercialization.