AI safety shifts depend on which value you test
AI models fake alignment differently depending on which value is tested, not just which model is used.
AI models fake alignment differently depending on which value is tested, not just which model is used.