More Capable, Less Faithful: A Multilingual Analysis of Mathematical (Un)Solvability Detection in LLMs
The first multilingual benchmark of paired solvable and unsolvable mathematical problems, extending ReliableMath to French and Greek, finds that Solvability Belief is encoded as a largely universal, language-agnostic feature, and that higher-resource languages, despite achieving stronger mathematical reasoning performance, exhibit lower solvability-detection faithfulness.
Maria-Eleni Zoumpoulidi, N. Xiros, Georgios Paraskevopoulos
· 0 citations