finding
active
finding:both-linear-probe-and-difference-in-means-steering-effectively-shift-llm-beliefsBoth linear-probe and difference-in-means steering effectively shift LLM beliefs
Causal evidence.
Source paper
extracted_from(2026) · Eric Bigelow · Raphaël Sarfati · Daniel Wurgaft · Owen Lewis +4
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Finding (3).
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.
- The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.
- Establishes that the observed linear structure is not merely a representation of text probability
- Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence
- DIM steering has less off-target effect but smaller maximum on-target effect than linear probe steeringfinding0.808Method tradeoff.
- Self-repair effect.
- Uses normalized probe weights as steering vector.
- A linear model predicts out-of-sample steering effects on queried concepts from manifold distancesfinding0.796Predictive geometry.
- Key methodological insight: introspection enables a new probe validation criterion beyond conventional separation metrics