Is this different than the model not really encoding space? Most operations will be spatially invariant which means binding will be problematic, but not THE problem.

Comments