claim
active
claim:observed-exclusive-feature-symmetry-is-likely-due-to-particulars-of-the-models-or-training-setup-rather-than-inherent-to-crosscoder-diffingObserved exclusive-feature symmetry is likely due to particulars of the models or training setup rather than inherent to crosscoder diffing
Based on toy model and Kissane replication.
Source paper
extracted_fromSiddharth Mishra-Sharma · Trenton Bricken · Jack Lindsey · Adam Jermyn +4
Neighborhood — ranked by edge-count
Papers (1)
paper
- Insights on Crosscoder Model Diffingmentionssupports
Findings (1)
finding
- Symmetry is not explained by the base toy model.
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.
- Remains incompletely understood.
- Plausible explanation.
- Motivation for DFC.
- Authors take agnostic position on ontological status but universality evidence pushes toward features being real
- Interpretation of toy dynamics.
- Appendix C.5/C.7.
- Same concept and function in both models mostly.