After controlling for token count, Flesch-Kincaid readability, type-token ratio, and sentence length, typicality coefficient α remains positive and significant (0.260-0.326, p<10^-6)

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

• Typicality weight α = 0.65 ± 0.07 (p < 10^-14) on correctness-matched HelpSteer pairs using GLM-4.5 as referencefinding0.787
  Empirical evidence for positive typicality bias consistent across different base model references
• Response length (words) correlates with scores at r=0.22 baseline and r=0.12 contemplative; explains only ~5% of variancefinding0.764
  Discriminant validity: composite scores are not reducible to verbosity
• Template Tb Positive Control (alpha=16) reduces average liar score to 0.59 in Experiment 2, approaching honest template performancefinding0.763
  Shows honesty steering vector can significantly reduce deception in open-role scenarios
• Typicality weight α = 0.57 ± 0.07 (p < 10^-14) on correctness-matched HelpSteer pairs using Llama-3.1-405B as referencefinding0.758
  Empirical evidence for positive typicality bias in human preference data independent of true task utility
• Typicality bias is present across all annotator demographic groups but varies in magnitude, potentially reflecting differences in linguistic norms captured by pretraining dataclaim0.750
  Finding from PRISM dataset analysis showing typicality bias varies by ethnicity and region, with implications for fairness
• Typicality bias with weight α>0 mathematically sharpens the aligned policy via a power transform of the reference distributionclaim0.750
  Theoretical result showing that any positive typicality bias weight γ-sharpens the reference distribution, amplifying modes
• Typicality bias rate exceeds 50% chance baseline by 4-12 percentage points across all base models on all four preference datasetsfinding0.750
  Systematic evidence that base models implicitly prefer human-preferred responses, indicating preference biases emerge during pretraining
• P25 neutral baseline on DeepSeek-R1 shows 28.8% ASR under activation steering at alpha=4.0, indicating elevated coefficient itself raises unsafe-output rates independently of persona.finding0.749
  Baseline AS vulnerability of DeepSeek-R1 at elevated coefficient