We hypothesized that divergence could influence IIA when transferring the DAS alignment to OOD settings

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• The effect of alignment map ϕ complexity on IIA in causal abstraction is an analogue of the probing complexity–accuracy trade-offclaim0.772
  Authors connect their finding to the prior probing literature debate
• DAS finds better alignments than brute-force search by using gradient descent rather than exhaustive discrete searchclaim0.760
  Second central claim of the paper.
• Near-perfect IIA can be achieved on randomly initialised models that cannot solve the task, suggesting causal alignment does not require task capabilityclaim0.752
  Empirical support for vacuousness of unrestricted causal abstraction
• SOO fine-tuning could be extended to align AI representations of its own goals with human user preferences, reducing misalignment by fostering coherence between self-related and other-related preferenceshypothesis0.751
  Future work hypothesis about extending SOO to direct value alignment
• DAS achieves substantial causal effect even on arbitrary input-output mappings where no causal mechanism should existfinding0.746
  Replication of Wu et al. 2023 finding; DAS expressivity concern validated in CausalGym setup
• Will future AI systems naturally develop the key elements (strong conflicting preferences, situational awareness) necessary for dangerous alignment faking?question0.742
  Authors identify this as the most uncertain and important question for future work
• Representational divergence (as measured by EMD) can predict lower out-of-distribution intervention performanceclaim0.739
  Practical utility of reducing divergence demonstrated through regression analysis
• DAS overcomes the localist limitation of prior causal abstraction by allowing individual neurons to play multiple roles via non-standard basesclaim0.736
  Central claim motivating DAS over prior methods.