Boundless DAS interchange interventions produce EMD exceeding natural-natural baseline

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

• SAE reconstructions on Llama-3-8B layer 25 produce intervened EMD exceeding the natural-natural baselinefinding0.763
  Empirical demonstration that SAE projections produce divergent representations in a real LLM
• Trainable intervention (DAS) finds sparser gender representations than linear probing, suggesting probing overestimates causal coverageclaim0.750
  Interpretive claim from Case Study II about the distinction between correlational probes and causal interventions
• DAS behavioral loss produces EMD along feature dimensions of 0.032±0.003 on synthetic 10-class datasetfinding0.748
  Quantitative baseline for divergence using behavioral DAS loss on synthetic dataset
• Representational divergence (as measured by EMD) can predict lower out-of-distribution intervention performanceclaim0.735
  Practical utility of reducing divergence demonstrated through regression analysis
• For small CL loss weights epsilon, IIA is maintained (potentially improved) while EMD decreases in Boundless DAS on a 7B LLMfinding0.734
  Empirical result showing the CL loss can reduce divergence without sacrificing interpretability accuracy
• DAS achieves substantial causal effect even on arbitrary input-output mappings where no causal mechanism should existfinding0.733
  Replication of Wu et al. 2023 finding; DAS expressivity concern validated in CausalGym setup
• Mean difference patching on Llama-3-8B layer 10 produces intervened EMD exceeding the natural-natural baselinefinding0.727
  Empirical demonstration that MDVP produces divergent representations in a real LLM
• DAS overcomes the localist limitation of prior causal abstraction by allowing individual neurons to play multiple roles via non-standard basesclaim0.727
  Central claim motivating DAS over prior methods.