Coordinates of Capability: A Unified MTMM-Geometric Framework for LLM Evaluation

Adib Sakhawat, Tahsin Islam, Takia Farhin +3

The evaluation of Large Language Models (LLMs) faces a critical challenge in construct validity, where fragmented benchmarks and ad hoc metrics frequently conflate method variance,…