method
active
method:feature-conditioned-hypothesis-generationFeature-Conditioned Hypothesis Generation
Finds data clusters where response concept is chosen/rejected leaning.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Pipeline from preference data to hypotheses to interventions.
Methods (3)
method
- Prompt-Conditioned Hypothesis Generationrelated_toFinds response-delta clusters shifted for prompt clusters.
- Cluster features by activation statistics to get higher-level behaviors.
- Statistic ranking data-feature pairs, filtered by sign consistency across halves.
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.
- Argues against hypothesis-generation defense.
- Models exploit non-robust but predictive patterns.
- LLM examines max-activating examples, writes rubric, judges test set.
- Hypothesis that neurons form privileged bases to encode information; consistent with constructive abstraction
- Synthesizing inputs maximizing component activation.
- The claim in RL that any goal can be expressed as maximizing the expected cumulative sum of a scalar reward signal.
- Quantitative relationship between concept frequency and feature presence.