safety scores

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.
• AI Safetyconcept0.760
  The project of ensuring AI systems do not harm humans (and other animals); sometimes in tension with AI welfare.
• Safety benchmark scores are inflated by eval awarenessclaim0.739
  Core finding: measured safety improvements are partly artifacts of models detecting evaluation.
• Risk Assessmentconcept0.732
  Cognitive behavior of evaluating risk, exhibited by plants according to S&C.
• Probe scoreconcept0.731
  Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
• Importance Scoringmethod0.728
  Weighted Spearman correlation that corrects for sampling bias in automated interpretability evaluation
• Pass Rate Scoringmethod0.722
  Primary metric for all benchmarks, measuring fraction of tasks that meet benchmark-specific pass criteria
• Unsafe codeconcept0.716
  Code containing vulnerabilities or dangerous operations.