After a year of rapid iteration, the industry already has answers as to which industries generative AI will be the first to be implemented.
The education field is not only listed as one of the key applications by OpenAI, but has also become a key direction for the layout of large domestic model manufacturers.
Advances in generative AI technology demonstrate increasing human interaction with large models. The large model continuously learns the human way of thinking through training, and humans also gain new inspiration and knowledge in the dialogue with the large model.
This process is the ideal state pursued by education and is also the embodiment of "teaching and learning". Generative AI is highly compatible with education, making education an excellent application field.
After determining the landing place, the more important thing is how to combine technology and education to achieve the role of empowering the latter.
What achievements has AIGC empowered education achieved in the past year? How will AIGC disrupt education? In the future, what new trends will emerge in education due to the development of generative AI?
In the "AIGC Education Industry Panoramic Report", the Qubit Think Tank gave a systematic review after exchanges within and outside the industry.
Core viewpoints include:
......
Let's look at the specific details one by one .
Generative AI leads us into the AI2.0 era. By learning massive amounts of data, AI has demonstrated amazing generative capabilities. The answers it generates are more in line with human language communication and can stimulate people's thinking consciousness to a certain extent.
In the field of educational technology, the most intuitive change is the “AGI” product. From learning machines to APPs to smart teaching, the educational technology field has access to GPT products to a certain extent.
During May-June 2023, the GMV of the AI learning machine equipped with iFlytek’s Spark cognitive large model increased by 136% and 217% year-on-year respectively. The AI learning machine also became the first choice for JD.com and JD.com during last year’s Double Eleven. Tmall dual-platform sales champion.
The learning machine is a unique educational technology product in the domestic market. With the support of large models, the learning machine has achieved a rebound in sales by providing incremental value to users.
From a product perspective, when doing AIGC education in China, most large companies choose to equip learning machines with large models to increase product sales by providing incremental value to users.
The learning machine track is currently highly competitive, and the emergence of generative AI is expected to build new competition barriers for learning machines.
From a user perspective, the audience for learning machines is K-12 people. The willingness to pay for this group of people comes from their parents.
In primary school, parents value the development of their children’s interests; while in middle school, parents value whether products can improve their children’s academic performance.
Language learning APP has performed well in overseas markets, and leading products have been integrated GPT4. Domestic language APPs also have access to large models, and the interactive question and answer with generative AI technology has been substantially improved compared to previous generation products.
From a product perspective, companies that enter the AIGC education track through software access general large models and use their own accumulated education data for fine-tuning training, focusing on APP products.
This type of product has the most positive market feedback among large-model products used in education due to its upgraded dialogue experience.
From the perspective of audience demand, currently it is mainly college students and office workers. Among this group of people, college students also have the need to improve their grades, while office workers do not have the need to improve their grades. Learning languages is more out of work needs or interest. For these two groups of people, user retention rate is the key. Based on this, product design will pay more attention to user experience and marketing strategies.
The combination of education and technology has a long history. With the rise of the Internet in the 1990s, the era of online education began. Entering the 21st century, adaptive learning has become a hot topic in overseas education through the introduction of intelligent systems. After 2010, with the development of artificial intelligence technology, technologies such as knowledge graphs were used in the field of education.
At every stage, education will be combined with current technology to solve the problems faced in education.
In the era of generative AI, the integration of technology and education has had a subversive effect on teachers and students, the two core roles of education. AI has been integrated into the daily lives of teachers and students in a new form - Agent, changing their learning and working status on the basis of providing teaching and learning assistance.
The combination of AI and education has also made technology companies account for an increasing share of the education market, and generative AI capabilities have become a new competitive point.
AI intelligence provides students with personalized learning Specialized teaching: mainly reflected in the after-school stage, which can effectively achieve one-to-one learning and tutoring.
In the era of generative AI, the biggest opportunity for educators is the universal benefit of adaptive education.
Before generative AI, the use of personalized learning in the education field was an extremely costly matter. It put forward high requirements for schools and families, requiring advanced hardware equipment and more teachers. and teaching venues.
After the emergence of generative AI, the development of AI agents is within reach. AI agents can serve as each student's personal assistant, accompanying students in their studies anytime and anywhere.
In terms of tool use, AI intelligent agents reduce teachers’ learning costs and provide It provides almost zero-threshold usage. In the work process, AI agents help teachers generate courseware more efficiently and provide teaching design ideas, etc.
AI agents release teachers’ productivity to a certain extent. More teachers’ energy can be used to improve students’ literacy. Teachers’ responsibilities gradually shift from teaching and solving doubts to educating people, and the focus expands from focusing on student performance to Pay attention to students' mental state.
Educational large models mainly face three types of challenges during the implementation process:
1) Data quality; 2) Illusion Question; 3) Values alignment.
Low-quality text data, including too much repeated content, low-quality articles, etc., not only affects the model training effect, but also causes It must be a waste of computing power.
The limited access to educational data, especially teaching behavior data, and the lack of high-quality educational annotation data result in limited trainable educational artificial intelligence models. Therefore, data quality becomes key to the development of large educational models.
The most critical thing about large models in the field of education is to ensure the accuracy of model answers gender and value orientation.
Accuracy is reflected in mathematical problems. The current accuracy of large education models on mathematical problems is low and cannot yet reach a level that can be applied on a large scale. In answering some high-level mathematics questions, the accuracy rate is generally below 50%.
In the field of education, the problem of hallucinations can be solved through question and answer reasoning based on search-augmented generation (RAG) and question and answer reasoning based on prompt word engineering.
For value orientation, it mainly refers to whether there are extreme and inconsistent remarks in the answers of the big model. , and even some error guidance occurred.
In terms of value alignment, the commonly used method is RLHF (Human Feedback Reinforcement Learning), which allows a language model trained on a general text data corpus to be aligned with complex human values.
In addition, since the target group is students, the review of large education models will be more stringent. AI-generated content will also be reviewed manually to further filter out low-quality content and ensure that there are no misdirections in the generated content.
In the domestic market, self-built large models have become an essential capability for industry players.
In the field of education, companies entering AIGC all choose to build their own large models. The difference lies in the capabilities of the models. Some companies choose to provide comprehensive tutoring, while others choose to make breakthroughs in a single subject first. At present, the enterprise's self-built large-scale education model has the ability to integrate interdisciplinary knowledge. In the future, interdisciplinary understanding will become the embodiment of high-level capabilities in self-built large models.
Interdisciplinary large models are a technical bridge between connecting knowledge and solving practical problems.
Interdisciplinary large models have higher value among student groups who need to master a rich knowledge system, such as middle school students and college students. For middle school students, there are many subjects to study. Students' usual learning method is to practice a large number of questions and memorize them repeatedly, but there is no "connection-transfer-through" between knowledge points. Therefore, it is easy to forget after taking the test and cannot be combined with practical applications. This is also a problem that the education field has always wanted to solve.
The interdisciplinary large model solves the problem of helping students truly integrate knowledge points and be able to use knowledge to solve real-life problems.
Multi-modal large models integrate information from different modalities, which can promote more accurate and comprehensive understanding and reasoning .
In the field of education, by analyzing data such as students’ writing, painting, and language expression, agents can better understand students’ learning status and needs, and provide personalized guidance and suggestions. By analyzing teachers' teaching data and classroom conditions, it can help teachers manage students in real time and complete teaching work efficiently.
The development of large multi-modal models will be accompanied by the continuous upgrading of hardware, and the product form of AI agents will also be iterated accordingly. As the data dimension increases, AI agents will gradually become users’ “exclusive” partners.
AI agents will develop toward concreteness, moving from Internet AI to Embodied AI.
Currently, AI agents have become the mainstream form of large model applications, completing tasks through "human-machine collaboration", with AI agents undertaking most of the work.
AI intelligence now exists in the form of software in various smart devices, and some products already have digital human images to enhance the realism of interactions.
In the future, Agent A will further evolve from the virtual world to the real world and have a physical image. Embodied intelligence that "combines soft and hard" will further expand the scope of the agent's capabilities. At this stage, the AI agent, in the form of a robot, empowers educational practice scenarios and has stronger companionship attributes.
In the brain-computer era, the interaction between AI agents and humans will be more in-depth. AI agents will be more accurate in understanding human intentions and analyzing states, and will be able to intervene in people with learning disabilities. The integration of man and machine will reach new heights.
Specific report details and a collection of representative player cases can be found in the comment area below.
The above is the detailed content of AIGC education industry panoramic report: AI teaching assistants and tutors come true, and learning machines usher in new opportunities. For more information, please follow other related articles on the PHP Chinese website!
![[Web front-end] Node.js quick start](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
