method
active
method:taboo-secret-word-elicitation-taskTaboo secret word elicitation task
Auditing task: discover the secret word a model gives hints for but never states.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Hidden knowledge models are trained not to reveal, target of auditing.
Datasets (1)
dataset
- Fine-tuned targets for secret elicitation evaluation.
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.
- Auditing task where black-box prompting beats AOs.
- One of four ToM tasks analyzed; requires inferring speaker intent from indirect hints; scored 0/1.
- Task where the model must simultaneously identify an injected thought and transcribe a text sentence.
- ToM task requiring integration of intent and tone to understand sarcasm; scored 0/1.
- Proposed cost-reduction procedure that predicts S/N/I label from unsteered baseline expression alone, replacing the full 30-configuration grid
- Aversive stimuli are defined as surprising, linking punishment to prediction failure.
- Zaadnoordijk and Bayne's category of intentional action; sticker-removal behavior induced by the self-prior corresponds to this
- Alternative elicitation using two-turn everyday conversations with a recurring character for persona extraction