MIRA-Math: A Benchmark for Minimal Information Requesting and Mathematical Reasoning
A new benchmark testing whether LLMs can identify and retrieve critical missing data to solve complex math problems.
Mathematical reasoning benchmarks typically provide all facts needed to solve each problem, while interactive benchmarks often mix reasoning with tools, retrieval, and long-horizon dialogue. We introduce MIRA-Math, a benchmark for a narrower diagnostic capability: solving mathema...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
MIRA-Math shifts the focus from mere computation to the strategic acquisition of missing information, addressing a critical gap in how LLMs handle partial knowledge. By enforcing strict constraints on fact-requesting, the benchmark reveals whether models truly understand logical dependencies or are simply guessing based on familiar patterns.
Get the day's AI news in one short email. Free.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.