The MathGPT large model has officially entered the public beta stage and can handle hundreds of billions of mathematical problems

On August 24, this site learned that during TAL’s 20th anniversary live broadcast, CTO Tian Mi announced that MathGPT, TAL’s self-developed hundreds of billions-level model in the field of mathematics, was officially launched and open to public beta. From now on, users can apply for a free trial experience by registering an account through the official website (www.mathgpt.com).

In May of this year, TAL announced that it was developing a large self-developed mathematical model, named MathGPT. MathGPT is a large-scale model in the vertical field of mathematics with problem-solving and problem-telling algorithms as its core for global mathematics enthusiasts and scientific research institutions. It is also the first large-scale model specifically built for mathematics in China.

The usage is also very simple. When users use MathGPT, they can upload math questions in the form of text or pictures to get conversational answer feedback. They can also use the "Random question" button to randomly generate math questions and have the system give answers.

MathGPT currently provides Chinese and English versions for PC and mobile experience

Leading mathematical problem-solving capabilities

MathGPT brings together TAL’s years of education, teaching and research data accumulation, focusing on the field of mathematics. The training, inference, and deployment framework of hundreds of billions of large models endows the model with powerful capabilities. Through high-quality education data, we can achieve continuous training and supervised fine-tuning of multi-tasks such as question calculation, explanation, question and answer, etc., showing excellent performance. In addition, with the help of human feedback alignment, the comprehensive quality of the model will be further improved. MathGPT has obvious advantages in problem-solving accuracy, stability and user experience.

It is understood that MathGPT’s mathematical calculation capabilities have covered mathematics problems in elementary school, junior high school, and high school. The types of questions cover calculation problems, application problems, algebra problems, etc. You can also ask follow-up questions about the topic. However, Q&A interactions other than mathematics are not yet open.

MathGPT Technical Report

What is the specific effect? Among the test results of 6 public mathematics assessment collections, including CEval-Math, AGIEval-Math, APE5K, CMMLU-Math, College Entrance Examination Mathematics and Math401, MathGPT achieved the highest scores in multiple tests. At the same time, MathGPT also performed well on C-Eval’s general test collection for middle and high schools.

MathGPT’s C-Eval rankings of middle and high school scores in various subjects

##In addition, In terms of problem-solving stability and explanation friendliness, MathGPT conducts model training based on massive data from famous teachers’ problem-solving processes, and the model’s problem-solving steps are professional and clear.

Let's take a sequence question as an example. The answer given by MathGPT includes three parts: "analysis", "detailed explanation" and "key points", which is rougher than the general large model. The explanation method is more detailed. Among them, "Analysis" provides the problem-solving ideas and thinking methods to help users better understand the topic; "Detailed Explanation" provides specific calculation methods and answers; and the final "Finding Points" link examines the test points, difficulties, and key points of the topic. Click for prompts to help users review and reflect on the intention of setting the question and draw inferences from one example.

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 to use 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 for global artificial intelligence experts and mathematics enthusiasts to experience and evaluate. . TAL hopes to make MathGPT play a greater role in the field of mathematics education, and is willing to share its R&D experience and methods of hundreds of billions of large models based on large-scale, high-quality content with the industry, and make progress together with the industry.

TAL Future AI’s accumulated experience

Driven by the AI wave, since this year Many technology companies have announced the launch of general-purpose large language model products, but TAL has chosen another direction. It is not based on fine-tuning and interface calls of existing large language models, nor making general-purpose large language models, but in-depth research and development in the vertical field of mathematics. Big Model is committed to creating independent, stable, sustainable, and high-quality mathematical solutions.

General large model "emphasizes text but ignores theory", and has obvious shortcomings in solving, explaining, answering and recommending mathematical problems. On another level, on the road to general artificial intelligence, mathematical reasoning ability is very important, and there are many large companies around the world doing research in this area.

“TAL has 20 years of accumulation in mathematics data and business. It has accumulated a large amount of educational data and has the ability to continuously produce educational data, so we chose to do this. Difficult and correct things." Tian Mi said that TAL hopes to use its many years of accumulation in mathematics and AI to do a good job in the mathematical foundation of the AI large model era.

In fact, TAL established an AI lab as early as 2017. With the help of the smart education artificial intelligence open innovation platform, TAL AI lab has won 16 championships and 6 runner-ups in various top academic conference competitions, and published nearly 100 high-level academic papers in international journals and conferences.

In 2019, the Ministry of Science and Technology announced that relying on TAL to build a new generation of national artificial intelligence open innovation platform for smart education, TAL became the first and only artificial intelligence "national team" in the education industry. ” Member, with many years of in-depth research in the field of artificial intelligence. Over the years, TAL has been driven by the major needs of the education industry to build a national education science and technology innovation platform with education-oriented artificial intelligence algorithm capabilities, application solutions, basic software and hardware systems, and open source open services.

TAL Future is also actively involved in promoting the construction of the large model standard system. As a core unit, it has participated in the large model series of national standards organized by the National Artificial Intelligence Standardization Group, China The "Large Model Pre-training Model Technology and Application Assessment Methods" series of group standards led by the Academy of Information and Communications Technology, and the "Education General Large Model" series of standards led by the Education Information Technology Standards Committee of the Ministry of Education and the National Information Technology Standardization Technical Committee.

## Recently, TAL is taking the lead as a leading unit to compile large education models with industry-leading scientific research institutions, universities, and enterprises such as China Academy of Information and Communications Technology, Fudan University, iFlytek, Baidu, etc. Group standards comprehensively evaluate the capabilities of large education models from the dimensions of coverage scenarios, application effectiveness, and service reliability, and provide reference and guidance for the implementation of large education model applications.

Use AI to implement large-scale teaching in accordance with aptitude

With the rise of large language models, How to use AI technology to serve all walks of life is the focus of social attention. The education industry is one of the first industries to start deploying AI, and the changes that AI can bring to the education ecosystem have always attracted much attention.

"AI has brought the opportunity to redefine the education industry, and large-scale model technology has made it possible to teach students in accordance with their aptitude on a large scale." Tian Mi introduced that in the past 20 years, TAL has been exploring personalized learning, from offline small classes to online large classes, and then to AI classes. The formats are constantly evolving, but the teaching content is always fixed, there is less interaction between students and teachers, and the granularity can only reach the question level. .

#Tian Mi believes that the essence of large models is a more efficient way to learn knowledge from data and apply it. With the support of AI capabilities, the new learning method of "students self-study AI Q&A" has become widely possible. The threshold and cost for learners to obtain high-quality teaching content are reduced, and the degree of personalization and refinement of the teaching content they obtain continues to increase. AI teaching and question-answering guidance can be realized for thousands of people, and each student can get the learning that is most suitable for him or her. content.

Based on MathGPT, TAL will continue to explore learning methods in the AI environment to better serve learners and mathematics enthusiasts around the world, and transfer experience in a timely manner Share with the industry and promote positive changes in educational technology through AI technology.

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 being accelerated and will be released in the near future.

The above is the detailed content of The MathGPT large model has officially entered the public beta stage and can handle hundreds of billions of mathematical problems. For more information, please follow other related articles on the PHP Chinese website!