“OpenAI is developing reinforcement fine-tuning techniques that will allow developers to create their own domain-specific tasks and graders to train models on correct tool-calling paths.”