claim
active
claim:this-is-the-first-study-to-mitigate-the-performance-trade-off-in-steering-vectors-derived-from-the-difference-in-means-approachThis is the first study to mitigate the performance trade-off in steering vectors derived from the difference-in-means approach
Priority claim establishing novelty of the Head Cor intervention approach
Source paper
extracted_from(2026) · Yoshihiro Izawa · Gouki Minegishi · Koshi Eguchi · Sosuke Hosokawa +1
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Used to extract trait and style directions.
- Steering vectors computed over multiple shots and norm-normalized have weaker effect than 0-shot vectorsfinding0.820Counter-intuitive result.
- Contrast with probe steering.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
- Steering vector from mean activation difference between contrasting datasets.
- Why do steering vectors computed over multiple shots have a weaker effect than single-query vectors?question0.801Left to future work.
- Steering vectors used to reduce eval awareness can inadvertently introduce alternative user personasfinding0.800A side effect observed when applying activation steering: the model's response persona changed unexpectedly.
- Observation from 100% accuracy on specific concept-layer-strength combinations suggesting concept-specific detectability