OpenAI lays out its progress towards 'recursive self-improvement'—even as its chief scientist warns of the risks and says he hopes for a slowdown