MD vectors outperform probe-based vectors in SJTs because they align with construct centroids rather than distorting direction via regularization

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

• Some steering vectors produce more salient perturbations than others, perhaps based on shared semantic or qualitative factorsclaim0.775
  Observation from 100% accuracy on specific concept-layer-strength combinations suggesting concept-specific detectability
• Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vectorfinding0.770
  Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
• Synthetic MFT SJTs achieve 77.71%-83.84% alignment with Clifford et al. human-composed MFT vignettesfinding0.762
  Moderate-to-high alignment validating SJT synthesis for moral foundations domain
• Steering vectors discover effective triggers such as 'However' and 'Otherwise', consistent with prior reported reflection datasetsfinding0.758
  Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
• In Qwen-2.5-9B, only v1 has meaningful cosine similarity to DIM direction; all additional basis vectors have cosine similarities ~1e-9finding0.758
  Appendix E replication of DIM alignment finding in Qwen model
• MDS injections align with the Linear Representation Hypothesis: target trait varies near-linearly with alpha in open-ended generationclaim0.755
  Theoretical alignment claim backed by OLS R2 analysis showing 96.15% of trends have R2>=0.75
• Why were interventions with mass-mean probe directions extracted from the likely dataset so effective, despite these probes not being accurate at classifying true/false statements?question0.755
  Open question raised in §7.1 about an unexplained anomalous result
• Steering vectors from µ(0→2) slightly outperform µ(1→2) for instruction discovery across datasets and modelsfinding0.755
  Shows that contrasting No Reflection with Triggered Reflection provides a stronger signal than Intrinsic vs Triggered.