method
active
method:crosscoder-based-steering-vector-transferCrosscoder-based steering vector transfer
Translates steering vector via weighted average of shared decoder vectors.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Chen et al. method to derive evil/hallucinating/sycophantic vectors.
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.
- A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
- Steering one concept direction while measuring introspection for a different concept, yielding a 4×4 steering-concept × measured-concept matrix to test concept-specific modulability
- Exploratory experiment computing vectors over multiple shots.
- Method for computing steering vectors as mean activation differences between reflection levels at a given layer.
- Learn a shared feature space across two models; basis of the work.
- MLP replacement with sparse features underlying the original attribution graphs.
- A combination of crosscoders and flexible decoding (Matryoshka, iterative codebook) could resolve the steering-classification tensionhypothesis0.770Proposed remedy.
- Supported by the instruction discovery experiments comparing steering vs. embedding baselines.