Tag
This Systematization of Knowledge paper proposes a unified Multi-Trait Multi-Method (MTMM) geometric framework for evaluating Large Language Models, unifying disparate metrics into a shared latent coordinate space to address construct validity issues in current benchmarks.