From September 19 to September 21, the 2024 Yunqi Conference, themed “Cloud Intelligence, Industrial Transformation,” was held at Yunqi Town in Hangzhou. TAL Education Group CTO Tian Mi was invited to attend the Yunqi Conference AI Native Architecture Summit and delivered a keynote speech titled "Educational Large Models Make 'Personalized, High-Quality, and Large-Scale Teaching' Dream Come True," revealing the progress of MathGPT.

MathGPT has already acquired the abilities of multi-disciplinary problem-solving, conversation, grading, explanation, and recommendation.

In his speech, Tian Mi first shared the "impossible trinity" theory in the education industry, meaning it’s difficult to achieve personalization, high quality, and large scale simultaneously. "To break the impossible trinity, we can only rely on the development of AI technology. The emergence of large model AI technology has given us hope," Tian Mi said. He explained that the education industry is one of the recognized prime fields for the application of large models. In 2023, Xueersi officially introduced MathGPT. The Xueersi team enhanced the model based on the industry's excellent open-source base models, including fine-tuning and reinforcement, to continuously improve its capabilities, allowing MathGPT to achieve lower costs and better results in relevant scenarios compared to general large models.

Recently, OpenAI's new model o1 has grabbed the industry's attention and acclaim. Trained through reinforcement learning, o1 has shown superior performance in reasoning tasks involving physics, mathematics, and other disciplines. Tian Mi mentioned o1 while sharing MathGPT training details and stated, "MathGPT had already undertaken a lot of reinforcement work more than half a year ago, with significant innovations at this stage, greatly enhancing its effectiveness."

Currently, MathGPT has been applied in multiple educational scenarios. According to Tian Mi, MathGPT now possesses capabilities in multi-disciplinary problem-solving, conversation, grading, explanation, and recommendation, successfully implemented in various practical applications.MathGPT has been embedded in multiple apps, smart hardware, and learning services under Xueersi. For example, grading for Chinese and English essays, and entire math papers has been realized in Xueersi Xpad, providing substantial help to parents; English conversation capabilities are applied in the learning machine and "ABC English Corner"; “MathGPT Anytime Ask” is available on apps and mini-programs, functioning like a one-on-one Math AI tutor for multi-round conversations and heuristic explanation; the adaptive recommendation ability tags and breaks down steps of questions to recommend targeted practice and personalized Q&A guidance for users.

Tian Mi stated that achieving success with the large model at Xueersi is inseparable from the support of the entire industry ecosystem and strong partners like Alibaba Cloud. At the same time, Xueersi has officially opened up the capabilities of MathGPT in problem-solving, explanation, and grading to the entire industry, and MathGPT will soon be available on the Alibaba Cloud BaiLian Platform to support and serve more industry users.

It is reported that this year’s Yunqi Conference has invited top global tech experts, academic elites, industry leaders, and innovation pioneers, holding 400 forums and special discussions, focusing on the integrated development of cloud computing and AI, cutting-edge achievements of large models, AI infrastructure, large model service toolchains, and in-depth industry practices.

Original article from: https://new.qq.com/rain/a/20240921A001R100