finding
active
finding:linear-model-learns-top-m-most-important-features-regardless-of-sparsity-n-20-m-5Linear model learns top-m most important features regardless of sparsity (n=20, m=5)
Baseline.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
- Main contribution for interpretability.
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.
- Trend observed in Experiment 2 results.
- Open question.
- Fig. 22.
- Concurrent work result showing emergent misalignment occurs in small models
- Hypothesis box in §3.2.
- A linear model predicts out-of-sample steering effects on queried concepts from manifold distancesfinding0.781Predictive geometry.
- Both empirical and theoretical phase diagrams.
- Motivation for using sparsity-based dictionary learning on language models