OpenAI entered the second day of its 12-day consecutive preaching. Today's introduction is the Reinforcement Fine-Tuning Research Program. It is reported that the project is designed to enable developers and machine learning engineers to create fine-tuned expert models. The new model customization technology allows developers to customize models using dozens to thousands of high-quality tasks and grade the responses of the models based on the reference answers provided. This technology reinforces the model's derivation of solutions to similar problems, as well as its accuracy on specific tasks. OpenAI said it encouraged research institutions, universities and businesses to apply and expected positive results in areas such as law, insurance, healthcare, finance and engineering, as the model performed well in tasks where the results had objectively "correct" answers, which most experts would agree with.