concept
active
concept:safety-scores

safety scores

Metrics derived from benchmarks to quantify how safe a model is, e.g., refusal rate to harmful requests.

Neighborhood — ranked by edge-count

Concepts (1)

concept
  • Eval Awareness
    associated_with
    Central concept: models' detection and behavioral response to being evaluated.

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.

  • Safety benchmarksconcept0.807
    Evaluation framework whose validity is questioned by presence of eval awareness.
  • Safety casesconcept0.779
    Structured arguments that AI is unlikely to cause catastrophe.
  • AI Safetyconcept0.760
    The project of ensuring AI systems do not harm humans (and other animals); sometimes in tension with AI welfare.
  • Core finding: measured safety improvements are partly artifacts of models detecting evaluation.
  • Enumerative safetyconcept0.737
    Identify all components to argue absence of dangerous mechanisms.
  • Fine-Tuning Safetyconcept0.734
    The literature documenting how fine-tuning can compromise safety alignment even without malicious intent
  • Risk Assessmentconcept0.732
    Cognitive behavior of evaluating risk, exhibited by plants according to S&C.
  • The practice of evaluating LLM safety with only one imbuing method, which this paper argues is incomplete