finding
active
finding:phase-boundary-jointly-modulated-by-context-length-and-steering-magnitude-reconstructed-by-the-bayesian-model-with-high-fidelityPhase boundary jointly modulated by context length and steering magnitude, reconstructed by the Bayesian model with high fidelity
Prediction 3 confirmed.
Source paper
extracted_from(2025) · Eric Bigelow · Daniel Wurgaft · YingQiao Wang · Noah D. Goodman +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (2)
claim
- ICL and activation steering both control LLM behavior by updating belief in latent conceptsassociated_withCore thesis.
- Prediction 3.
Hypotheses (1)
hypothesis
- Eq. 10.
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.
- Concurrent work result showing emergent misalignment occurs in small models
- Source paper for Bayesian model reduction methodology used in structure learning
- Promising future research direction about the internal mechanism of error detection.
- Proposed future application of the Assistant Axis
- A linear model predicts out-of-sample steering effects on queried concepts from manifold distancesfinding0.755Predictive geometry.
- The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.
- Limitation acknowledgment about the adequacy of the linear representation assumption
- The method can steer the model in both positive and negative directions on the target semantic.