finding
active
finding:non-robust-toy-models-have-mean-off-diagonal-interference-about-2-that-of-robust-modelsNon-robust toy models have mean off-diagonal interference about 2× that of robust models.
Figure 3B.
Source paper
extracted_from(2025) · Gorton, Liv · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of Figure 3.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Figure 4.
- Unexpected result.
- Validates toy model.
- Superposition controls robustness.
- Symmetry is not explained by the base toy model.
- Appendix replication.
- Concurrent work result showing emergent misalignment occurs in small models
- Adversarially robust models have different interpretability because adversarial training affects superpositionhypothesis0.744Could be tested by crosscoder comparisons.