got into an argument with our ML lead at 11pm yesterday about an eval methodology a PM had built off a framework she learned at an AI PM cohort. shes claiming a layered defense framework, hes saying the layers are statistically conditioned and her independence claim is wrong. they both have a point.
