The SLAM Problem and the Atlas Framework, Right?

In the SLAM problem, people have found usually that as the space gets bigger, you start having errors in a global map. And so one of the optimal solutions is to split the map into smaller chunks and then use certain transformations for going across those chunks. That like minted SLAM and also the Atlas framework proposes these kinds of ideas. But none of them have said anything about reusing those chunks to kind of predict what the rest of the environment might look like or what other environments might look like across environments. It's interesting. The fact that we generalize indicates that we're probably using a lot of the stuff that we're learning and combining or composing it in various ways in order