finding

active

finding:steering-vector-constructed-from-all-16-contrastive-pairs-outperforms-most-single-pair-vectors-best-4-pair-vector-outperforms-full-16-pair-vector

Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vector

Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.

Source paper

extracted_from

Steering Evaluation-Aware Language Models to Act Like They Are Deployed

(2025) · Hua, Tim Tian · Qin, Andrew · Marks, Samuel · Nanda, Neel

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Four best contrastive prompt pairs outperform full 16-pair average steering vector for type hint suppressionfinding0.881
Optimization result for steering vector construction.
Contrastive Steering Vector Constructionmethod0.824
Method for computing steering vectors as mean activation differences between reflection levels at a given layer.
Steering vectors discover effective triggers such as 'However' and 'Otherwise', consistent with prior reported reflection datasetsfinding0.819
Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
Some steering vectors produce more salient perturbations than others, perhaps based on shared semantic or qualitative factorsclaim0.812
Observation from 100% accuracy on specific concept-layer-strength combinations suggesting concept-specific detectability
Steering vectors capture latent dimensions of reflective behavior more faithfully than surface-level embedding similarity.claim0.809
Supported by the instruction discovery experiments comparing steering vs. embedding baselines.
Steering at 6 layers (strength 0.6 each, total 3.6) outperforms single-layer steering at equivalent total strength for type hint suppressionfinding0.806
Demonstrates distributed steering is more effective and less accuracy-damaging than concentrated steering.
Distributing steering strength across multiple layers (6 layers at 0.6 each) is more effective and less accuracy-damaging than concentrating the same total strength in one layerclaim0.798
Practical finding for optimizing steering setup.
Steering vectors enable systematic discovery of reflection-inducing instructions beyond trial-and-error prompt design.claim0.794
Core applied contribution claim, supported by top-k accuracy comparisons.