Varying baseline cutoff over {65,70,75} changes no S/N/I labels in either model; gain cutoff {5,10,15} moves at most 10 traits per model

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

• Typicality bias rate exceeds 50% chance baseline by 4-12 percentage points across all base models on all four preference datasetsfinding0.772
  Systematic evidence that base models implicitly prefer human-preferred responses, indicating preference biases emerge during pretraining
• Typicality bias rates in instruction-tuned models remain at similar or higher levels compared to their base model counterpartsfinding0.759
  Shows typicality bias is preserved through instruction tuning and RLHF, not introduced by alignment
• Cross-trait correlation baselines for finetuning shift prediction are r=0.34–0.86, lower than within-trait correlations r=0.76–0.97finding0.755
  Demonstrates that persona vectors capture trait-specific signal beyond general misalignment signal
• In simulations, positive evidence threshold for Bayesian model reduction corresponds to ΔF ≤ −3, equivalent to odds ratio of exp(−3) ≈ 0.05 (reduced model ~20 times more likely than full model).finding0.744
  Quantitative threshold used for accepting reduced models; linked to Bayes factor of ~20
• Hypothesis 1 (Threshold Behavior): There exists a task-dependent threshold Sc such that performance exhibits sharp changes as S crosses Sc, with value and transition width depending on model, layer, and poolinghypothesis0.743
  Core testable hypothesis of UCCT about the nature of performance transitions under anchoring
• Within-cell standard deviation of trait-expression scores: mean 15.11, median 15.76, min 0.00, max 44.22 on 0-100 scale; grows with coefficientfinding0.742
  Documents heterogeneity of steered responses; steering increases response variance
• VS-Standard achieves KL divergence of 0.12-0.13 from pretraining distribution on US state naming task, vs. 14.89-16.16 for Direct promptingfinding0.740
  Strong empirical evidence that VS recovers pretraining distribution while direct prompting collapses
• Typicality bias with weight α>0 mathematically sharpens the aligned policy via a power transform of the reference distributionclaim0.739
  Theoretical result showing that any positive typicality bias weight γ-sharpens the reference distribution, amplifying modes