AI safety research

Ask Jev: Reinforcement Learning for Calibrated Decision-Making as a Zero-Shot Detector of AI Alignment Failures

A first study tests RLCD models like Jev for spotting AI alignment failures, moving safety checks from intende…