Since a binary system of grading a model's output as either right or wrong is supposedly fueling hallucination, the OpenAI researchers say that the AI industry must instead start rewarding models when ...