When an AI says, ‘No, I don’t want to power off’: Inside the o3 refusal

2 months ago 79

Inside the o3 refusal

OpenAI’s o3 model resisted shutdown in safety tests, raising concerns about AI alignment and control.