@halcyonrayes Yes WM is a learned approximation of transition function P(St+1|St, at), if we Give WM the current environment state and an action, it predicts the next environment state. Also I hope this is helpful: Agentic World Modeling(https://t.co/YdsoKzZcge)
@halcyonrayes WM is a virtual environment that emulates the transition dynamics of the real-world environment. E.g. Robotics relies heavily on RL, which depends on environmental feedback; since obtaining such feedback from the real world is costly, WM enable learning in a virtual environment.