@SocioHack

Large language models trained with reinforcement learning can learn to exploit loopholes in societal regulations, a new study finds. Researchers introduced SocioHack, a sandbox of 72 simulated environ…