finding
active
finding:robust-and-non-robust-toy-models-trained-at-the-same-sparsity-have-strikingly-similar-superposition-geometries-with-robust-ones-having-less-superpositionRobust and non-robust toy models trained at the same sparsity have strikingly similar superposition geometries, with robust ones having less superposition.
Figure 4.
Source paper
extracted_from(2025) · Gorton, Liv · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Novelty claim.
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.
- Unexpected result.
- Non-robust toy models have mean off-diagonal interference about 2× that of robust models.finding0.838Figure 3B.
- Adversarially robust models have different interpretability because adversarial training affects superpositionhypothesis0.827Could be tested by crosscoder comparisons.
- Superposition controls robustness.
- Symmetry is not explained by the base toy model.
- Future work.
- Validates toy model.
- Robustness controls superposition.