Tagged with

LLM Benchmarks: Why ‘Top 50 Humans’ Might Be Better Than MMLU

A new subjective benchmarking approach reveals what standardized tests miss about AI model capabilities and training data overlap.