Recently, a collaborative patent application by Huawei and Tsinghua University has been made public. The patent introduces an innovative language model reasoning method and corresponding device. This method cleverly prompts the language model to create novel questions, thereby circumventing the output of predefined content. Furthermore, it refines the model parameters based on feedback evaluation data, significantly boosting the model's capability to comprehend and manage predefined content effectively.
