Paper proposes new taxonomy for assessing LLM mathematical creativity

A new paper argues that assessing the mathematical capabilities of large language models is currently underspecified. The author proposes a taxonomy of mathematical creativity, distinguishing between modes like reflexive introspection, analogical import, problem-driven construction, and bridging distant domains. Current transformer-based systems are believed to excel at recombination and search, potentially limiting their ability to perform other modes of mathematical creativity in principle. As AI improves at generating proofs, the paper suggests that mathematical value is shifting towards these less accessible modes, and evaluations should reflect this. AI

IMPACT Suggests a new framework for evaluating LLM mathematical abilities, potentially guiding future research and development.

RANK_REASON Academic paper published on arXiv discussing LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• transformer-based systems

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →