I'm somewhat skeptical of that approach. It's not even clear if "world models" are significant for our subcouscious decision making as human beings. Why assume it's necessary for AI?
But I'm not at all an expert in the domain and I might have misunderstood something please feel free to contradict me!