method
active
method:10m-feature-weakly-causal-crosscoder-dictionary10M-feature Weakly Causal Crosscoder dictionary
Dictionary used for all main-text features.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Weakly causal crosscoder (10M features)related_toSource of the features used in attributions.
- Gradient-based technique using SAE features to estimate causal effects on completions; used to corroborate NLA findings.
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.
- Naive implementation results.
- Cited as enabling precise behavioral control through SAE features, extending the same methodological line
- Train a single SAE encoding and decoding activations from two models simultaneously to find shared and exclusive features.
- Decoder cost dominates.
- Sparse dictionary learning variant that reads and writes to multiple layers or models, yielding shared features.
- Figure 5.
- Extension of SAEs that jointly learns latents across representations from different models; proposed as alternative for extended fine-tuning
- Appendix C.4.