claim
active
claim:intervening-on-features-can-steer-model-behaviour-in-a-precise-quantitative-wayIntervening on features can steer model behaviour in a precise, quantitative way
Central claim motivating the product.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Listed achievement in TL;DR.
Questions (1)
question
- Open problem flagged by authors.
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.
- Clamping feature activations causally alters model behavior in interpretable ways.
- The method can steer the model in both positive and negative directions on the target semantic.
- Central interpretive claim organizing the entire paper's results
- Core question motivating the shift from linear to geometry-aware steering; answered via manifold alignment analysis.
- features serve as oversight signals to intentionally design models with desirable capabilitiesquote0.801Core vision statement.
- Explanation of brittle steering.
- If a model's features are well-calibrated, can one use these features as sources of supervision?question0.796Motivating question.