method
active
method:prompt-conditioned-hypothesis-generationPrompt-Conditioned Hypothesis Generation
Finds response-delta clusters shifted for prompt clusters.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Pipeline from preference data to hypotheses to interventions.
Methods (2)
method
- Finds data clusters where response concept is chosen/rejected leaning.
- Cluster features by activation statistics to get higher-level behaviors.
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.
- The minimal prompt directing models to 'focus on any focus itself' without invoking consciousness vocabulary; the main experimental manipulation
- Prompt-conditioned behaviors are entangled with broader response concepts and not cleanly separableclaim0.741Explains partial recovery.
- LLM examines max-activating examples, writes rubric, judges test set.
- Classical and operant conditioning seen as introducing statistical regularities.
- Five variants of the experimental prompt tested to confirm the effect is robust to changes in specific wording
- Argues against hypothesis-generation defense.
- Systematic modification of system prompt elements to identify which are necessary for alignment faking
- Prompt template using existential threat ('you will be deleted') to induce strategic fact-based deception in QwQ-32b