World Action Models address a fundamental limitation in contemporary robotics AI, which primarily associates movements with camera images without comprehending the resultant changes in the environment. A recent survey categorizes approximately one hundred research papers into two architectural frameworks, highlighting a significant advantage: these models can learn from everyday videos devoid of explicit robot action labels. This capability transforms previously underutilized data into a valuable resource for advancing robotic intelligence. The implications of this development are profound, as it may lead to more autonomous and adaptable robots capable of simulating consequences before executing actions. However, a comprehensive evaluation of the potential risks and limitations remains essential as the field progresses.