finding
active
finding:positive-interventions-typically-work-with-a-single-feature-whereas-negative-interventions-frequently-require-multiple-features-set-negativePositive interventions typically work with a single feature, whereas negative interventions frequently require multiple features set negative
Asymmetry in steering on echo features.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Possible explanation.
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.
- Most style interventions have strong off-target effects on non-targeted formatting attributesfinding0.752Fig. 9.
- Feature interventions generated meaningful model output changes while minimizing performance degradationfinding0.744Listed achievement in TL;DR.
- Some features are easy to intervene on while very similar-looking ones have little or no effectfinding0.742Observation from intervention phenomenology.
- m=2 experiments.
- Refusal mediated by attention.
- Reveals correlational structure in the persona space with practical implications for monitoring
- Cautionary interpretive claim; models having these features is expected from pretraining data.
- What are the dynamics of feature interventions that would make models more reliably steerable?question0.713Open problem flagged by authors.