claim
active
claim:low-cosine-shared-features-are-best-interpreted-as-same-features-used-differently-in-the-two-modelsLow-cosine shared features are best interpreted as same features used differently in the two models
Interpretation of cosine distribution.
Source paper
extracted_fromSiddharth Mishra-Sharma · Trenton Bricken · Jack Lindsey · Adam Jermyn +4
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Larger rotations between shared features skew decoder cosine similarities downward in the toy modelsupportsSupports 'same features, used differently' interpretation.
- Validates 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.
- Third empirical pattern.
- Suspicion not investigated in depth.
- Same concept and function in both models mostly.
- Supports the co-activation explanation of symmetry.
- Plausible explanation.
- Explicitly identified research gap: anecdotal evidence exists but rigorous characterization is absent
- Main proposal.
- Is the apparent universality of some low-level vision features the exception or the rule?question0.761Open empirical question following anecdotal cross-model universality findings