VS recovers 66.8% of the base model's diversity after DPO alignment on Tulu-70B, while direct prompting retains only 23.8%

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

• After DPO stage, VS outperforms direct prompting by 182.6% on diversity in poem continuation using Tulu-70Bfinding0.860
  Demonstrates the magnitude of VS's advantage over direct prompting after aggressive alignment training
• Post-training stages (SFT, DPO, RLVR) progressively reduce output diversity, with direct prompting showing the most severe mode collapseclaim0.784
  Empirical finding from Tulu-70B ablation study across post-training stages
• VS increases diversity by 1.6-2.1x over direct prompting on creative writing tasks (poem, story, joke)finding0.784
  Core empirical result demonstrating VS's effectiveness on creative writing diversity
• 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 promptingfinding0.782
  Strong empirical evidence that VS recovers pretraining distribution while direct prompting collapses
• VS-Standard achieves KL divergence of 0.54 from pretraining distribution on Open-Ended QA, vs. 3.14 for Direct and 0.58 for Sequencefinding0.780
  Shows VS substantially better approximates the pretraining distribution than baseline methods
• VS-Multi achieves average accuracy of 37.5% on math benchmarks (avg of MATH500, OlympiadBench, Minerva) with Gemini-2.5-Flash as generator and Qwen3-4B as SFT model, vs. 40.7% baselinefinding0.778
  Best VS result in synthetic data generation for math, demonstrating downstream improvement through diversity
• All VS variants maintain refusal rates above 97% on StrongReject benchmark, within 0.3-0.8 percentage points of the Direct baselinefinding0.776
  Confirms VS does not compromise safety alignment while improving diversity
• Base models assign higher likelihood to typical-set (representative) sequences than to degenerate sequences under VS promptshypothesis0.773
  Assumption D.6 formalized in the theoretical framework; empirically validated with coin-flip typicality rating experiments