curate

New research argues AI alignment can't be judged by model-level tests alone

A new paper argues that evaluating AI alignment solely at the model level is insufficient for understanding its real-world deployment. The research highlights that current benchmarks lack user-facing verification and pr…