claim
active
claim:crosscoder-causal-graph-may-differ-from-the-model-s-actual-mechanism-even-if-activations-are-isomorphic-so-interventional-experiments-need-careCrosscoder causal graph may differ from the model's actual mechanism even if activations are isomorphic, so interventional experiments need care
Downside of abstraction.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Claims (1)
claim
- Closing position.
Questions (1)
question
- Open issue in discussion.
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.
- Decoder cost dominates.
- Scaling causal interventions on codes will move from qualitative alignment to mechanistic claims.hypothesis0.779Future work.
- Central thesis of the paper
- Circuit simplification theory.
- Cited as enabling precise behavioral control through SAE features, extending the same methodological line
- Motivating problem.
- Methodological lesson.