claim
active
claim:sparse-interpreter-models-may-represent-and-decompose-computational-mechanisms-differentlySparse interpreter models may represent and decompose computational mechanisms differently
Hinted by differences.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Difference between sparse and neuron analyses.
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.
- A better basis (e.g., factoring tokens into languages or parts of speech) may yield a sparse interpretable modelhypothesis0.835Same model can be dense in one basis and sparse in another.
- Sparse Autoencoders Find Highly Interpretable Features in Language Models (Cunningham et al., 2023)concept0.811Core methodology paper for SAE-based interpretable feature extraction
- Central claim of the paper: the method scales to state-of-the-art transformers.
- Core empirical hypothesis of the paper, supported by successful VPD decomposition yielding ~10,000 interpretable subcomponents across 24 weight matrices.
- Cited as enabling precise behavioral control through SAE features, extending the same methodological line
- Forward-looking prediction about scalability of the method to larger models
- Critique of activation-based interpretability methods.
- Articulates why a one-layer transformer with MLP is the appropriate starting target for mechanistic interpretability