finding
active
finding:pre-post-mlp-crosscoder-shows-a-trimodal-split-into-pre-only-shared-and-post-only-featuresPre/post MLP crosscoder shows a trimodal split into pre-only, shared and post-only features
Relative decoder norm structure.
Source paper
extracted_fromRelated 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.
- Encoders read pre-MLP space; decoders reconstruct pre-MLP and MLP output to find newly computed features.
- A post-MLP feature for words indicating uniqueness has pre-MLP inputs firing on specific such wordsfinding0.789Abstract features built from specific instances.
- Motivation for DFC.
- Interpretation of toy dynamics.
- Figure 5.
- May explain drift: MLPs relay features along different axes.
- Decoder norm profile across 18 layers.
- Decomposition into shared and model-specific features.