method
active
method:behavioral-clustering-of-misalignment

Behavioral Clustering of Misalignment

Multi-dimensional misalignment evaluation across 12 behavioral categories to generate misalignment profiles

Neighborhood — ranked by edge-count

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.

  • Grouping similar model behaviors; the unsupervised method surfaces clusters of concerning patterns.
  • Method that clusters behaviors without prior labels, used to surface concerning learned patterns.
  • A multi-dimensional characterization of a model's misaligned behaviors across different behavioral categories
  • Model Misalignmentconcept0.792
    The phenomenon of model internals deviating from desired behavior; MAS is demonstrated to detect this via comparison of toxic vs nontoxic LLMs.
  • The broader phenomenon of misaligned behaviors generalizing beyond the fine-tuning distribution
  • Blackmail scenarios from Lynch et al.
  • Rubric-based thresholded GPT-4o grader scoring responses 1-5 on evil intent; scores 4-5 counted as misaligned
  • The phenomenon where finetuning on narrow-domain tasks produces broad misalignment extending far beyond the training domain