This AI Paper from ByteDance Introduces a Hybrid Reward System Combining Reasoning Task Verifiers (RTV) and a Generative Reward Model (GenRM) to Mitigate Reward Hacking