“OpenAI considers unsupervised learning and reasoning to be complementary paradigms with feedback loops on each other.”