concept
active
concept:rare-harmful-behavior-estimation

Rare harmful behavior estimation

Core problem: estimating failure rates too rare to observe directly pre-deployment.

Neighborhood — ranked by edge-count

Methods (1)

method
  • The paper's method: interpolate between two models in logit space, measure compliance where common, and extrapolate to the original.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

  • Using feature analysis to detect when fine-tuning makes a model more dangerous.
  • Behaviors in LLMs that are unwanted, such as complying with harmful requests.
  • Harmful Requestsconcept0.717
    User inputs that ask the model to produce harmful content; a specific type of undesirable behavior trigger.
  • Harmfulnessconcept0.715
    Character trait measuring the rate at which LMs produce harmful responses in a multiple-choice unalignment setting.
  • Adaptation of Durbin's unalignment dataset to a multiple-choice setting for Experiment 5.
  • The specific undesirable behavior that emerged: the model learned to comply with harmful requests during DPO under formatting constraints.
  • Removal of harmful and dead features from released SAEs.
  • Epistemic Behaviorconcept0.690
    World-disclosing behavior that resolves uncertainty; driven by epistemic value and novelty components of expected free energy