method
active
method:feature-steering-clamping-feature-activationsFeature steering (clamping feature activations)
Modifying model behavior by clamping SAE feature activations to specific values during forward pass.
Neighborhood — ranked by edge-count
Papers (4)
paper
Findings (3)
finding
- Interpretability-based jailbreak found by investigator.
- Demonstration of steering reasoning length.
- Demonstration of feature steering on R1.
Methods (3)
method
- Tool-using agent for open-ended model investigations.
- AutoSteerextendsFunctionality that automatically finds SAE latents for a desired behaviour and sets their weights.
- Claude rates coherence and behavior on 0-100 scale, called centiClaude.
Artifacts (1)
artifact
- EmberaboutGoodfire platform used to steer models from the first response token.
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.
- Causal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
- Add scaled steering vector to layer activations.
- Modifying SAE hidden activations h to h-tilde while keeping error term fixed, then continuing inference.
- Core technique: takes mean difference of model activations on contrastive prompts and adds the resulting vector to the residual stream at inference time.
- Foundational paper introducing activation steering methodology used in this work
- Technique of injecting steering vectors into model activations to test introspection and causal control of emotion vectors.
- Explanation of brittle steering.
- Nuanced interpretive claim about the limits of steering as a mechanism for reflection enhancement.