Outperforming models from OpenAI and Microsoft, Mathpresso’s large language model, MathGPT, has broken a new world record in mathematics. Mathpresso is the proud inventor of QANDA, the top AI-driven learning platform in Asia.
In benchmark assessments of mathematical proficiency, such as the ‘MATH’ benchmark (12,500 difficult math problems) and the ‘GSM8K’ benchmark (8,500 primary school arithmetic problems), MathGPT has achieved the highest ranking.
Notably, it beat the previous record-holder, Microsoft’s “ToRA 13B,” and it even beat OpenAI’s GPT-4 in the MATH benchmark.
This accomplishment is the result of KT, Upstage, and Qanda working together since November of the prior year. Through the provision of crucial learning data to Upstage, Qanda enabled the creation of MathGPT.
The success of MathGPT contrasts with that of models such as ChatGPT, which frequently produce inaccurate responses (a phenomenon called hallucination) as a result of training on large amounts of text data.
Precision and dependability are essential in educational settings, and MathGPT’s exceptional accuracy—especially in mathematical domains—places it at the forefront of AI-driven education.
In their statement, Qanda pledges to keep improving MathGPT’s precision and functionality. The final objective is to include it in their learning interface and launch an “AI Tutor” that acts as a teaching assistant. The goal is to revolutionize the education business by bringing AI tutoring services to educational websites throughout the world.
The CEO of Upstage, Kim Seong-hoon, emphasized the value of working together to create a top-notch, math-specific language model.
Declaring, “We will lead generative AI innovation,” he demonstrated their dedication to developing artificial intelligence. With substantial backing from organizations like Google, TikTok, and Softbank Ventures Asia, Qanda is still making a big impact on the education landscape by using cutting-edge AI technologies.
