finding
active
finding:steering-vectors-computed-over-multiple-shots-and-norm-normalized-have-weaker-effect-than-0-shot-vectorsSteering vectors computed over multiple shots and norm-normalized have weaker effect than 0-shot vectors
Counter-intuitive result.
Source paper
extracted_from(2025) · Eric Bigelow · Daniel Wurgaft · YingQiao Wang · Noah D. Goodman +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Questions (1)
question
- Why do steering vectors computed over multiple shots have a weaker effect than single-query vectors?answered_byLeft to future work.
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.
- Many-shot vector geometry.
- Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
- Priority claim establishing novelty of the Head Cor intervention approach
- Steering Vector Control maintains low unexpected rate of 0.08 in Experiment 1, comparable to baselinefinding0.814Shows that inducing deception via steering vectors preserves semantic coherence and does not cause random errors
- The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.
- Core empirical claim comparing steering approaches on cyclic concepts.
- Mechanistic interpretation of how activation steering induces deception through the model's reasoning process