AUPRC Latent Activation Classifier

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

• Latent Activation Signatureconcept0.785
  The pattern of which misalignment-relevant SAE latents become active for a given fine-tuned model
• Latent #10 peak activation classifies persona jailbreak prompts vs benign prompts with AUROC = 0.96, and vs non-persona jailbreaks with AUROC = 0.93finding0.740
  Quantifies latent #10's strong discriminative power for persona jailbreaks
• AUROCconcept0.730
  Performance metric for binary classification; used to evaluate pathogenicity prediction.
• Activationsconcept0.715
  Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
• Per-prompt average activation of latent #10 can accurately classify aligned vs misaligned models from a single evaluation promptfinding0.698
  Demonstrates practical utility of SAE-based monitoring even with minimal sampling
• Activation decompositionconcept0.697
  The conventional approach (e.g., SAEs, transcoders) of decomposing activations into interpretable features.
• Activation Oracles (AO)method0.696
  Supervised method training models to answer questions about activations; NLAs differ by being unsupervised.
• Behavior-Optimized Activation Path Recoverymethod0.694
  Method of optimizing activation-space interventions to produce behavioral paths along M_y, then measuring whether the resulting activation trajectories trace M_h curvature