During the final exam preparation season, college students began to popularize the AI final review

Mondo Education Updated on 2024-02-01

Large-scale model products are making quality education the standard.

The field of education is one of the recognized fertile ground for the landing of AI large-scale model products.

Since ChatGPT started the road of large-scale model hurricane, the war in many subdivisions such as education, educational hardware, smart classrooms, and digital campuses has also burned into the field of large-scale models. At present, many education companies, including Good Future, Ape Tutoring, NetEase Youdao, Homework Gang, Reading Lang, etc., have released vertical education large model applications, and AI giants such as Ali and iFLYTEK have also completed the docking of their own large model products with the education market.

Behind the large model bet by many players is the imaginative prospect of AI+ education. From AI translation, AI homework correction, and AI precision learning, to today's AI Q&A, AI oral practice, AI real-time note-taking, and even AI lectures, the ever-changing large-scale model technology and application are reshaping the concept, model, and ecology of the education industry.

At a time when large-scale models are helping to promote high-quality education as the standard,Learn to go AIIt has also become a compulsory course for everyone.

While final exam week is underway, college students are in full swing to prepare. This year, a new class called"AI Final Review Method".The preparation method has become popular on social platforms and has attracted the attention of many college students.

Source: Internet.

The first step to mastering the AI Final Review Method is to have a suitable AI tool. In the era of large models, AI, as the technological driving force of production efficiency, is comprehensively transforming the services and experiences of various scenarios, and many practical application cases have been born at home and abroad in the dimension of learning efficiency improvement.

ChatGPT, the ancestor of large models, is one of them, and as a representative of the general large model, ChatGPT has become a "learning artifact" for many college students and graduate students. On social networks, college students share that they use ChatGPT to conduct literature search, data collation, content translation, plan formulation, and test question analysis.

Google's Socratic is a more vertical AI learning application that provides efficient learning solutions through technologies such as image recognition, speech recognition, and natural language processing, based on Google's AI capabilities. By taking photos and uploading assignments and questions, students can get **answers, analysis**, explanations**, learning materials and other resources.

Source: Screenshot of Google Play.

Of course, compared with foreign large models with higher access thresholds, domestic AI applications are more popular among young people.

Take Alibaba's Quark app as an example, this app equipped with a variety of AI capabilities has recently been regarded by many college students as a new "partner" for learning efficiency. A few days ago, Quark also launched a new large-scale model application "AI Topic Assistant" to help college students improve the efficiency and quality of self-learning with intelligent problem-solving ideas and simple explanations.

After the Yiou experience, it was found that after using the photo search tool of the Quark APP, the AI intelligent question function will appear in some result pages, and a virtual "Quark baby" will give the answers to the test questions and the solution ideas.

Source: Screenshot of Quark app.

Source: Screenshot of Quark app.

According to the official public information, at present, Quark's "AI Question Assistant" is the first to be launched in some scenarios of searching for English content, covering multiple choice questions, fill-in-the-blank questions, reading questions and other common test question types, and mainstream subjects such as mathematics will be added in the future, so that AI will gradually penetrate into all aspects of the learning scene.

AI can be a companion to students or a good helper for teachers to teach.

Bubu, a university teacher who teaches public English subjects, told Yiou that after using the Quark app many times, she found that the answers given by the "AI Topic Assistant" are very accurate and the explanations are very detailed and popular, and "it may not be better to teach students by yourself." Bubu also said that he is willing to recommend more intelligent tools like quark to students, "because there are many students in some majors, and the Q&A time between classes is particularly tight, so it is much easier to have AI."

Compared with traditional online courses and first-class explanations, intelligent tools such as Quark's "AI Topic Assistant" also have many advantages such as low cost, fast response, and simple language in the scenario of online self-study for college students.

Wang Zai, a sophomore in college, told Yiou that because he is more "socially phobic", he usually does not like to ask teachers, but he "either can't understand it or is afraid of making mistakes" when he finds answers on the Internet. At the recommendation of his classmates, he used Quark's "AI Topic Assistant" and found that AI was simply his "final savior", which could answer his questions at any time and improve his learning efficiency.

An excellent AI application is inseparable from the powerful empowerment of the underlying large model.

ChatGPT, which is popular all over the world, currently has GPT-3, a subsidiary of Open AI5 and GPT-4, two large language models, the former supports unimodal input and output, with 175 billion parameters, which can be called by users for free; The latter is a multi-modal large model with parameters as high as 18 trillion, requires a paid subscription to call.

Standing behind Quark's "AI Topic Assistant" is the full-stack self-developed, 100-billion-level parameters, multi-modal Quark large model released by Ali in mid-November last year. The data shows thatThe quark 100-billion-level parameter model topped the two authoritative lists of C-Eval and CMMLU. At the same time, in a number of learning evaluations, Quark achieved close to full scores in the college entrance examination, and passed the clinical practitioner qualification examination with a high level of 486 points, showing the strength of a real "scholar".

Source: Internet.

Quark has explained the advantages of its large model in four aspects: data, industry, knowledge correctness, and platformThe correctness of knowledge is the core ability of the large model to be implemented in the fields of education and medical care. Because there are many scenarios in these fields that require high rigor, scientificity, and accuracy of content, this poses more tests for the development and application of large models.

It is reported that in terms of general knowledge and writing enhancement, Quark has established a set of technical systems that can distinguish the authenticity of knowledge from content, search to reasoning, so it greatly reduces the hallucination rate of large models. The so-called hallucination rate, simply understood, is the frequency of "nonsense" of large models in testing. The data shows that in the field of health, quark large models can reduce the hallucination rate to less than 5%.

Shen Yang, a professor and doctoral supervisor at the School of Journalism at Tsinghua University, has a point of view that an important measure for large models to reduce the hallucination rate is to collaborate with search engines. Therefore, in terms of illusion elimination, large models with search engine backgrounds often have obvious advantages, so they can be better implemented in educational scenarios.

Some industry experts believe that one of the essences of AI transformation of education is to realize the scale, fairness, and personalization of high-quality educational resources. The learning materials generated by the application of large models are a supplement to educational content resources. In addition to intelligent tools, content resources will also be the focus of the competition on the AI+ education track.

The advantages of enterprises such as homework help, ape tutoring, and good future focus on teachers and real teaching, while Quark is based on the needs and insights of users in search scenarios, through self-built content and cooperation, it has accumulated a very rich professional information and learning materials, including more than 100 professional courses in universities, more than 500 popular majors and all kinds of practical information.

Cheng Fei, the person in charge of Quark learning products, said to Yiou:"The launch of the 'AI Topic Assistant' is an important step for Quark Learning to assist users in improving their learning efficiency, as AI products not only allow users to find answers, but also provide them with intelligent assistants to assist them in the self-learning process. We will continue to build a new learning experience by building a next-generation e-learning platform with the three advantages of search, content, and intelligent tools. ”

The transformation of scenarios such as test preparation and self-study by large-scale model applications is a microcosm of AI-enabled education. At present, the entire industry is paying attention to this promising track, and data shows that the scale of China's AI education market will exceed 799.3 billion yuan in 2024.

It is foreseeable that the huge market demand will give birth to more products, applications, scenarios, and platforms that break through imagination, so that AI can help human beings better acquire, inherit, and innovate knowledge.

Related Pages