method
active
method:dose-response-feature-steering-protocolDose-Response Feature Steering Protocol
Varying each feature's activation from -0.6 to +0.6, averaging over 10 random seeds per setting
Neighborhood — ranked by edge-count
Frameworks (1)
framework
- SAE Feature SteeringimplementsMethod of adding scaled versions of sparse autoencoder latent features during generation to causally modulate model behavior
Artifacts (1)
artifact
- Key paper finding structured first-person descriptions in LLMs claiming awareness or subjective experience during self-referential processing.
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.
- Modifying SAE hidden activations h to h-tilde while keeping error term fixed, then continuing inference.
- Consistent with Bigelow et al.
- Framework of using internal-state representations to control or steer generative models; conceptually parallel to manifold steering in language models.
- How do S and R respond to different mitigation strategies such as feature steering and in-training defenses?question0.717Future direction exploring whether persona-sensitive diagnostics can detect residual misalignment after interventions
- The method can steer the model in both positive and negative directions on the target semantic.
- Modifying model behavior by clamping SAE feature activations to specific values during forward pass.
- Scaling features inside QK circuits of selected heads to validate mechanisms.
- Steering changes behavior regardless of input x, as a constant shift in log posterior oddsclaim0.703Derived from LRH.