method
active
method:prefix-score

Prefix score

Prior metric on synthetic repeated random strings; induction score analogous.

Neighborhood — ranked by edge-count

Methods (1)

method
  • Proxy metric: average attention on literal-copy induction token pairs on a standard dataset.

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.

  • Probe scoreconcept0.771
    Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
  • Detection Scoremethod0.764
    LLM judge classification accuracy measuring interpretability; threshold 0.8.
  • Common opening of R1 reasoning traces that the model treats as part of the prompt.
  • Mixing Scoremethod0.739
    Average row entropy of attention matrices per layer and head, measuring information mixing across tokens
  • Elo scoremethod0.737
    A rating system used to compare model helpfulness and harmlessness based on crowdworker preferences.
  • A scoring rule optimized by predicting true probabilities; log-loss is one.
  • Alignment Scoreconcept0.724
    GPT-4o scored 0-100 metric where lower values indicate more misaligned behavior on open-ended evaluation prompts
  • Primary scoring method: scorer sees three reference responses at known quality levels alongside each target to eliminate inflation