Are you sure that they are facing different prior information? If the sim is a good one, then the TDT agent won’t be able to tell whether it is the sim or not. However, you are right that one solution could be that there are multiple TDT variants who have different information and so can logically separate their decisions.
I mentioned the problems with that in another response here. The biggest problem is that it seriously undermines the attraction and effectiveness of TDT as a decision theory if different instances of TDT are going to find excuses to separate from each other.
Are you sure that they are facing different prior information? If the sim is a good one, then the TDT agent won’t be able to tell whether it is the sim or not. However, you are right that one solution could be that there are multiple TDT variants who have different information and so can logically separate their decisions.
I mentioned the problems with that in another response here. The biggest problem is that it seriously undermines the attraction and effectiveness of TDT as a decision theory if different instances of TDT are going to find excuses to separate from each other.