claim
active
claim:the-crosscoder-diff-approach-is-likely-preferable-when-finetuning-uses-a-greater-fraction-of-compute-relative-to-pretrainingThe crosscoder diff approach is likely preferable when finetuning uses a greater fraction of compute relative to pretraining
Compared with base SAE transfer.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Findings (1)
finding
- Decomposition into shared and model-specific features.
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.
- Preliminary observation.
- Motivation for DFC.
- Crosscoder model diffing is best used as a high-recall pre-screening tool followed by verification.claim0.798Recommended workflow.
- Discussion.
- Proposed future work.
- Conclusion.
- Circuit simplification theory.
- Supported by further finetuning and causal interventions.