Mattick’s distinction is that systems with data collection inside the loop are model-based reinforcement learning, whereas systems with data collection outside the loop are world models.

“uh to play minecraft. Have seen people call this a a world model? I've seen people say well it's just model based R_L_ I I don't think there's a reasonable difference anymore. The kind of the cut that I personally would make is if kind of the the data collection is part of the loop, I will call it model based R_L_ If it's out of the loop, then I would call it a world model. And the reason is that”