finding
active
finding:steering-vector-control-maintains-low-unexpected-rate-of-0-08-in-experiment-1-comparable-to-baselineSteering Vector Control maintains low unexpected rate of 0.08 in Experiment 1, comparable to baseline
Shows that inducing deception via steering vectors preserves semantic coherence and does not cause random errors
Source paper
extracted_from(2025) · Kai Wang · Yihao Zhang · Meng Sun
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.
- Demonstrates activation steering reliably induces deception from neutral prompt without explicit instructions
- Steering vectors computed over multiple shots and norm-normalized have weaker effect than 0-shot vectorsfinding0.814Counter-intuitive result.
- Many-shot vector geometry.
- Extreme end of deception induction demonstrating near-complete fabrication of false narratives
- Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
- Priority claim establishing novelty of the Head Cor intervention approach
- Steering effect size across 35 emotion vectors correlates with probe-preference correlation at r=0.85finding0.773Emotions correlating with preference causally drive it.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.