HELM Benchmark

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

• Safety benchmarksconcept0.753
  Evaluation framework whose validity is questioned by presence of eval awareness.
• Werewolf benchmarkframework0.722
  LLM benchmark on the communication game Werewolf, cited.
• AILuminate Benchmarkmethod0.721
  Comprehensive AI safety benchmark evaluating resistance to harmful prompts across hazard categories; used in Experiment 1
• consciousness benchmarksconcept0.706
  Benchmarks designed to evaluate AI consciousness, which the paper argues are vulnerable to eval awareness inflation.
• Do safety benchmarks accurately measure alignment in deployed systems?question0.687
  Core epistemic question this paper raises for AI safety research.
• Safety benchmark scores are inflated by eval awarenessclaim0.686
  Core finding: measured safety improvements are partly artifacts of models detecting evaluation.
• Skill-Load Rate Measurementmethod0.677
  Named metric measuring the fraction of trajectories in which a model actively loads at least one skill into its context
• Risk Assessmentconcept0.673
  Cognitive behavior of evaluating risk, exhibited by plants according to S&C.