OpenAI pulled access to a GPT-4o powered toy caught giving inappropriate responses, but it apparently still thinks these ones ...