< Explain other AI papers

SAND-Math: Using LLMs to Generate Novel, Difficult and Useful Mathematics Questions and Answers

Chaitanya Manem, Pratik Prabhanjan Brahma, Prakamya Mishra, Zicheng Liu, Emad Barsoum

2025-07-29

SAND-Math: Using LLMs to Generate Novel, Difficult and Useful
  Mathematics Questions and Answers

Summary

This paper talks about SAND-Math, a system that helps AI models create new and challenging math questions along with their answers to improve their problem-solving skills.

What's the problem?

The problem is that existing math problems used to train AI models are often limited in difficulty and variety, which stops the models from getting better at solving more complex math questions.

What's the solution?

SAND-Math solves this by using large language models to generate new math problems and then gradually increase their difficulty. This process creates a diverse set of tough math questions that help train the AI more effectively.

Why it matters?

This matters because better training with hard and diverse math problems makes AI models smarter and more capable at solving advanced math tasks, which can be useful for education, research, and other technical fields.

Abstract

SAND-Math, a pipeline for generating and elevating the complexity of mathematical problems, significantly enhances the performance of Large Language Models on the AIME25 benchmark.