TAL launches self-developed large-scale mathematics model MathGPT to achieve personalized teaching through AI

On August 24, 2021, Tian Mi, TAL’s Chief Technology Officer, announced that MathGPT, a 100-billion-level large-scale model in the field of mathematics independently developed by TAL, has been officially launched and has begun public testing. Starting today, users can apply to register an account through the official website, try out and experience the model for free

TAL announced in May this year that it was developing a large self-developed mathematical model, which was named MathGPT. According to reports, MathGPT is a large-scale model for mathematics enthusiasts and scientific research institutions around the world. Its core is problem-solving and problem-telling algorithms, focusing on the vertical field of mathematics. This is also the first large-scale model in China specially built for the field of mathematics

When users use MathGPT, they only need to upload math questions through text or pictures to get conversational answer feedback. In addition, you can also click the "Random question" button, and the system will randomly generate math questions and give answers. Currently, MathGPT already supports Chinese and English versions of PC and mobile experience

MathGPT is a project focusing on the field of mathematics. It brings together TAL’s education, teaching and research data accumulated over the years. By utilizing these high-quality educational data, MathGPT is able to perform continuous training and supervised fine-tuning on various tasks such as problem calculation, explanation, and question answering, thereby demonstrating excellent performance

The official website of MathGPT shows that MathGPT’s mathematical calculation capabilities have been extended to various mathematics problems in elementary school, junior high school and high school. These question types include calculation questions, application questions and algebra questions. In addition, MathGPT can also ask questions on questions, but it is not yet open for question and answer interaction with questions other than mathematics

Mathematical GPT Technical Report

The mathematics GPT technical report shows that among the test results of six public mathematics evaluation collections, including CEval-Math, AGIEval-Math, APE5K, CMMLU-Math, college entrance examination mathematics and Math401, TAL's MathGPT has achieved multiple test results. highest score. At the same time, MathGPT also performed well on C-Eval’s general test collection for middle and high schools.

The content that needs to be rewritten is: MathGPT’s scores in various subjects of middle and high schools on the C-Eval list

MathGPT's problem-solving stability and explanation friendliness are based on model training based on massive problem-solving process data of famous teachers. The model's problem-solving steps are professional and clear

Taking a sequence question as an example, the answer of MathGPT is divided into three parts: "Analysis", "Detailed Explanation" and "Key Points". Compared with simple explanations of general large models, MathGPT's answers are more detailed. The "Analysis" section provides problem-solving ideas and ways of thinking to help users better understand the topic. The "Detailed Explanation" section gives specific calculation methods and answers. Finally, the "Key Points" link prompts the test points, difficulties and key points of the question, helping users review and think about the intention of the question, and be able to draw inferences from one instance to other cases

For users, studying mathematical problems is not only about getting the answer itself, but also about the problem-solving principles and logic behind the answer. Compared with other general-purpose large models, MathGPT can achieve higher accuracy problem solving, and can also analyze and explain answers more clearly, better meeting the core needs of users using AI products to answer mathematical problems.

At the same time as MathGPT was released, TAL also updated a representative and challenging mathematical task evaluation set on the official website for global artificial intelligence experts and mathematics enthusiasts to experience and evaluate

With the rise of large language models, how AI technology can provide services to all walks of life has become the focus of social attention. Tian Mi believes that AI brings opportunities to redefine the education industry, and large-scale model technology makes it possible to teach students in accordance with their aptitude on a large scale. The essence of large models is a more efficient way to learn knowledge from data and apply it. With the support of AI, the new learning method of "student self-study and AI question answering" has become widely feasible. Learners can lower the threshold and cost of obtaining high-quality teaching content, and at the same time obtain increasingly personalized and refined teaching content, realize AI teaching and Q&A guidance for thousands of people, and ensure that each student can get the most suitable course for them. Learning Content

According to Tian Mi’s introduction, with the smooth progress of the public beta, MathGPT’s problem-solving capabilities will continue to improve, and product-level applications based on MathGPT are also accelerating the development process and are expected to be released in the near future

The above is the detailed content of TAL launches self-developed large-scale mathematics model MathGPT to achieve personalized teaching through AI. For more information, please follow other related articles on the PHP Chinese website!