WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models | Bohai Gu et al. | ResearchPod