← back
arXivDivya Appapogu, Freya Behrens, Yonatan Belinkov, Aaron MuellerFri, Sep 11, 2026, 10:08 AM PDT
score 15.5

New benchmark tests how AI models represent non-binary concepts

Original: MAxBench: A Multinomial Concept Recovery Benchmark

Source: arxiv.org

Writing ELI5 summary…