In one such case, OpenAI has reported a breakthrough in which the company trained strong language models to produce text that is easy for weak language models to verify. The Microsoft-backed ...