question
active
question:why-are-exclusive-features-symmetric-between-models-in-real-diffsWhy are exclusive features symmetric between models in real diffs?
Remains incompletely understood.
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 Diffingassociated_withmentions
Findings (1)
finding
- Near-identical numbers of exclusive features are found for both models in base vs finetuned diffssupportsSymmetry of exclusive features across models.
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.
- Primary unexpected empirical observation.
- Based on toy model and Kissane replication.
- Mixed picture for diffing.
- Central interpretation.
- Symmetry is not explained by the base toy model.
- Authors take agnostic position on ontological status but universality evidence pushes toward features being real
- Plausible explanation.
- Similar number and nature of exclusive features across models results from oversampling of Human/Assistant chat transcriptshypothesis0.743Motivated by Bricken et al; tested and not supported.