标签
Introduces an analytically exact framework for controlled behavioral evaluation of LLMs, using fully crossed factorial experiments and exact token-level probability mass functions to isolate causal biases that aggregate benchmarks obscure.
本文提出将ZCA白化作为WEAT的几何预处理步骤，以解决嵌入各向异性问题，结果表明校准会改变超过30%结果的显著性状态，未经校准的偏差测量可能不可靠。