The beauty of the approach is that it’s agnostic about such things. It can support your view (let’s call it physical continuity) the same way it supports SSA and SIA. More precisely, you can give UDT a fixed measure of care about descendants that is 33% SSA, 33% SIA and 33% physical continuity. Then the post shows that UDT will make decisions as though it was adjusting these weights after each observation, so from the perspective of each view it will look like UDT is learning that view with high probability. That was the goal of the post—to give a crisp model of updating in favor of this or that anthropic assumption.
The beauty of the approach is that it’s agnostic about such things. It can support your view (let’s call it physical continuity) the same way it supports SSA and SIA. More precisely, you can give UDT a fixed measure of care about descendants that is 33% SSA, 33% SIA and 33% physical continuity. Then the post shows that UDT will make decisions as though it was adjusting these weights after each observation, so from the perspective of each view it will look like UDT is learning that view with high probability. That was the goal of the post—to give a crisp model of updating in favor of this or that anthropic assumption.