hypothesis
active
hypothesis:if-rebalancing-is-a-general-property-of-reasoning-models-behavior-change-attempts-e-g-suppressing-dishonest-responses-require-more-sophisticated-techniques-as-models-route-around-modificationsIf rebalancing is a general property of reasoning models, behavior-change attempts (e.g., suppressing dishonest responses) require more sophisticated techniques as models route around modifications
Safety implication of the oversteering result.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Oversteering some R1 features makes the model revert to original behavior before outputs become incoherentassociated_withSecond steering insight, unlike non-reasoning models.
Concepts (1)
concept
- AI alignmentcitesField within which this work has implications for evaluating alignment progress.
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.
- Open question posed by the authors.
- Reasoning-model training may incentivize greater implicit awareness of internal state, producing the rebalancing effecthypothesis0.794Why rebalancing would appear in reasoning models specifically.
- Practical implication.
- Core motivation.
- Inverse scaling interpretation.
- §3 Discussion.
- The method can steer the model in both positive and negative directions on the target semantic.
- Supports the awareness hypothesis.