The paper discusses the potential of video generation as a powerful tool for real-world decision-making, drawing parallels between video and language models. While language models have made significant strides in various applications, video generation remains largely limited to entertainment. The authors argue that video data captures crucial information about the physical world that is difficult to express in text. They propose that video can serve as a unified interface for absorbing internet knowledge and representing diverse tasks, similar to how language models function. The paper highlights recent advancements in video generation techniques, such as in-context learning, planning, and reinforcement learning, which enable video models to act as planners, agents, compute engines, and environment simulators. The authors identify key impact opportunities in domains like robotics, self-driving, and science, supported by recent research demonstrating the feasibility of advanced video generation capabilities. Finally, they address major challenges in video generation, including dataset limitations, model heterogeneity, hallucination, and limited generalization, and suggest potential solutions to overcome these obstacles. The paper concludes by emphasizing the potential of video generation to become a critical component in AI applications, alongside language models.The paper discusses the potential of video generation as a powerful tool for real-world decision-making, drawing parallels between video and language models. While language models have made significant strides in various applications, video generation remains largely limited to entertainment. The authors argue that video data captures crucial information about the physical world that is difficult to express in text. They propose that video can serve as a unified interface for absorbing internet knowledge and representing diverse tasks, similar to how language models function. The paper highlights recent advancements in video generation techniques, such as in-context learning, planning, and reinforcement learning, which enable video models to act as planners, agents, compute engines, and environment simulators. The authors identify key impact opportunities in domains like robotics, self-driving, and science, supported by recent research demonstrating the feasibility of advanced video generation capabilities. Finally, they address major challenges in video generation, including dataset limitations, model heterogeneity, hallucination, and limited generalization, and suggest potential solutions to overcome these obstacles. The paper concludes by emphasizing the potential of video generation to become a critical component in AI applications, alongside language models.