finding
active
finding: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-promptingVS-Standard achieves KL divergence of 0.12-0.13 from pretraining distribution on US state naming task, vs. 14.89-16.16 for Direct prompting
Strong empirical evidence that VS recovers pretraining distribution while direct prompting collapses
Source paper
extracted_from(2025) · Jiayi Zhang · Simon C.H. Yu · Derek Chong · Anthony Sicilia +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Shows VS substantially better approximates the pretraining distribution than baseline methods
- Shows VS enables LLMs to better approximate random behavior compared to direct prompting
- Human study confirming automatic diversity metrics align with human perceptions
- Cost-diversity trade-off analysis showing VS's practical efficiency
- Best VS result in synthetic data generation for math, demonstrating downstream improvement through diversity
- Quantifies how much of the base model's diversity VS can recover compared to baseline prompting
- Confirms VS does not compromise safety alignment while improving diversity
- A measure of the difference between two probability distributions, used extensively in free energy formulations.