Xueersi, a Chinese education technology company, is developing a self-made large model for mathematics named MathGPT, which is aimed at math enthusiasts and scientific research institutions worldwide. The model is based on problem-solving and teaching algorithms in the field of mathematics and has achieved a phased result. Xueersi plans to launch product-level applications based on this self-made large model later this year.
MathGPT has been made a core project of Xueersi, led by CTO Tian Mi. The project started team building, data and computing power preparation, and technical development before the Spring Festival of this year. Xueersi aims to use its accumulated experience in mathematics and artificial intelligence to lay the foundation of mathematics in the era of large models for math enthusiasts and scientific research institutions worldwide.
Xueersi hopes that MathGPT can solve three problems of large language models (LLMs): first, ensuring the accuracy of problem-solving results; second, providing stable and clear problem-solving steps; and third, presenting interesting and personalized explanations for problem-solving.
Currently, there are several main schools of thought in the math field, such as Photomath, acquired by Google, Microsoft Math, Mathway, and WolframAlpha, which mainly use traditional AI technologies and databases to solve math problems.
Companies pursuing AGI, such as Google’s Minerva model and GPT4, try to improve the performance of LLMs on math tasks. Xueersi chose another less-traveled path, not based on existing LLMs but rather self-made MathGPT, a large model based on a professional field. Xueersi aims to create independent, stable, sustainable, and high-quality learning solutions.
As large language models continue to evolve, the choice of different technological routes is still subject to discussion and verification. It remains to be seen whether MathGPT, a self-made large model, can surpass general-purpose models in mathematical tasks and better match the math learning scenarios of different groups. As the industry deepens, and more talent participates in this field, more mature solutions are expected in the near future.