资讯

OpenAI researchers say they've discovered hidden features inside AI models that correspond to misaligned "personas," or types ...