ATG: Benchmarking Automated Theorem Generation for Generative Language Models

arXiv:2405.06677v1 Announce Type: new
Abstract: Humans can develop new theorems to explore broader and more complex mathematical results. While current generative language models (LMs) have achieved significant improvement in automatically proving theorems, their ability to generate new or reusable theorems is still under-explored. Without the new theorems, current LMs struggle to prove harder theorems that are distant from the given hypotheses with the exponentially growing search space. Therefore, this paper proposes an Automated Theorem Generation (ATG) benchmark that evaluates whether …

abstract arxiv automated benchmarking cs.ai cs.cl current explore generate generative humans improvement language language models lms results struggle theorem type while