Figure 1: Per-sample distributional signatures across three LLM sizes (Section [4](https://arxiv.org/html/#S4.SS0.SSS0.Px1 "Sample-size bootstrap protocol. ‣ 4 Experiments ‣ Evaluating LLM-Generated Preference Distributions") resampling protocol; the CI band for the number of sampled distributions equal to 2020 is by definition zero because there is only one sample equal to the full pool, so the informative part of each curve is the small-RR region). (a) Unique destinations: 8B (mean 71.6) is 1.58×\\times the 405B (45.2) and 2.45×\\times the 70B (29.3). (b) Maximum destination probability: 70B and 405B concentrate 0.171 and 0.174, respectively; 8B is more diffuse (0.090). (c) Std of probabilities: 70B and 405B at 0.038; 8B tighter (0.019). The three models occupy structurally different operating points, not refinements of one another: across the 20 personas no two share an identical 70B/405B top-5 (mean top-5 overlap 0.120.12) and even the single dominant destination differs in most cases (top-1 match 0.300.30; Table [3](https://arxiv.org/html/#S5.T3 "Table 3 ‣ 5.3 Distributional Discordance Across Scales and Families ‣ 5 Results ‣ Evaluating LLM-Generated Preference Distributions")), so the larger model re-weights *which* destinations dominate rather than sharpening the same modes.#### Cross-Family Validation. 出处: Evaluating LLM-Generated Preference Distributions（arXiv）Figure 8: Scale discordance matrix showing pairwise correlations between three LLM sizes (8B, 70B, 405B). Values near zero indicate minimal agreement, demonstrating that model scale does not induce smooth distributional refinement. Different models encode fundamentally different world priors.Figure [9](https://arxiv.org/html/#A8.F9 "Figure 9 ‣ Appendix H Model Size Comparison: Full Heatmaps ‣ Evaluating LLM-Generated Preference Distributions") compares persona-based probability distributions side by side across the three model sizes. 出处: Evaluating LLM-Generated Preference Distributions（arXiv）
