question
active
question:what-is-the-relationship-between-the-features-we-extract-and-the-actual-underlying-computational-differences-between-modelsWhat is the relationship between the features we extract and the actual underlying computational differences between models?
Open question for safety applications.
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
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.
- Core framing hypothesis; analogy not to be taken literally.
- Explicitly identified research gap: anecdotal evidence exists but rigorous characterization is absent
- Core claim distinguishing this paper's contribution from looser representational similarity arguments.
- Authors argue features are model properties because logit effects and ablations are consistent with feature interpretations
- Second of three speculative claims asserting that subgraphs of neural networks are tractable and meaningful objects of study
- Fresh-traction question.
- Feature presence depends on concept frequency in training data, with a threshold scaling inversely with alive features.
- Primary unexpected empirical observation.