finding
active
finding:even-the-most-extreme-relative-norm-features-in-the-standard-crosscoder-have-non-zero-decoder-projections-to-the-other-model-mean-relative-norm-0-8884-llama-0-1652-qwenEven the most extreme relative-norm features in the standard crosscoder have non-zero decoder projections to the other model (mean relative norm 0.8884 Llama, 0.1652 Qwen).
Footnote 2.
Source paper
extracted_from(2026) · Thomas Jiralerspong · Trenton Bricken
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Motivation for DFC.
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.
- Appendix C.4.
- Figure 5.
- Decoder norm profile across 18 layers.
- Cross-layer transcoder features form a mix of local, global and intermediate features by decoder normfinding0.800Preliminary decoder norm analysis.
- Post-hoc metric for model exclusivity in standard crosscoders.
- Key design choice encouraging feature exclusivity.
- Table 1.
- Cross-judge validation of the primary ESR finding across OpenAI, Alibaba, Anthropic, and Google judge models