Given the linear representation hypothesis and binary linguistic features, 1D DII is sufficiently expressive for controlling model behaviour in CausalGym

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

• Linear representation hypothesis: neural networks represent meaningful concepts as directions in their activation spaces.hypothesis0.765
  Foundation for interpreting features as linear directions.
• Trainable intervention (DAS) finds sparser gender representations than linear probing, suggesting probing overestimates causal coverageclaim0.763
  Interpretive claim from Case Study II about the distinction between correlational probes and causal interventions
• Representation geometry causally shapes behavior; activation and behavior manifolds are approximately isometric.claim0.760
• All three Claude models show high boundary_awareness and low aesthetic_response relative to own means — distinctive Constitutional AI signaturefinding0.758
  Constitutional AI fingerprint in dimension profile; training that makes models self-observant also makes them polished at cost to aliveness
• Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models (Marks et al., 2025)concept0.757
  Cited as enabling precise behavioral control through SAE features, extending the same methodological line
• DAS consistently finds the most causally-efficacious features across all pythia model sizes in CausalGymfinding0.756
  Main benchmark result showing DAS superiority over probing, diff-in-means, PCA, k-means, LDA, and random
• We hypothesize that representation geometry drives model behavior — the geometric structure of internal representations causally shapes what models do externally.hypothesis0.754
  The causal hypothesis motivating the use of causality (intervention) as the lens connecting representation and behavior geometry.
• Does the geometric structure of neural representations causally shape model behavior?question0.754
  The motivating research question of the paper