concept
active
concept:contrastive-stimulus-designContrastive Stimulus Design
LAT methodology step constructing paired prompts that elicit divergent behaviors to extract steering vectors
Neighborhood — ranked by edge-count
Methods (1)
method
- Method for extracting deception steering vectors via PCA on contrastive activation differences; achieves 89% detection accuracy
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.
- Pairs of prompts at different reflection levels used to compute steering vectors.
- The property that living structures contain intense contrast—far more than one imagines helpful; true opposites which annihilate each other when superimposed, creating differentiation that gives birth to something; contrast unifies rather than separates when used correctly
- Supervised learning framework where system learns by observing contrast between current response and nudged improved response; requires weak additional forces from supervisor
- Method comparing brain activity in conscious vs. unconscious conditions.
- Pairs of statements with opposite truth values used as input to CCS; e.g., cities and neg_cities paired statements
- Core technique: takes mean difference of model activations on contrastive prompts and adds the resulting vector to the residual stream at inference time.
- Probe construction method: concept vector at each layer is L2-normalized difference between mean positive and mean negative representations from contrastive system prompts
- Unsupervised probing method from Burns et al. 2023 that identifies directions along which contrast pair representations are far apart