Whether the steerability map transfers across models and survives fine-tuning serves as the future research avenue

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

• Does the steerability map transfer across models and survive fine-tuning?question0.956
  Primary future research question identified by the authors
• What a model exposes by default tracks the norms it was trained toward, while steering acts on deviations from those defaults rather than on the defaults themselvesclaim0.784
  Central interpretive claim organizing the entire paper's results
• Features can be used to steer large models.claim0.770
  Clamping feature activations causally alters model behavior in interpretable ways.
• Optimally steering model behavior requires isolating concept geometry and defining operators to navigate it.claim0.768
• SOO fine-tuning could be extended to align AI representations of its own goals with human user preferences, reducing misalignment by fostering coherence between self-related and other-related preferenceshypothesis0.768
  Future work hypothesis about extending SOO to direct value alignment
• Fine-tuning an emergently misaligned model on 120 secure code samples (35 steps, batch size 4) fully restores alignmentfinding0.767
  Demonstrates emergent re-alignment is achievable with minimal data from same domain
• Pairwise steering needs empirical maps rather than assuming vector addition will behave like ordinary semantic blendingclaim0.766
  Cosine similarity is useful but insufficient; highly similar traits can still destructively interact
• Our method enables bidirectional steering of model behavior.finding0.756
  The method can steer the model in both positive and negative directions on the target semantic.