Verbalized Sampling recovers pre-trained generative diversity by prompting models to verbalize distributions rather than single instances

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

• Verbalized Sampling improves diversity without compromising factual accuracy or safety alignmentclaim0.830
  Claims verified by commonsense reasoning and safety evaluation experiments showing VS maintains >97% refusal rates and comparable factual accuracy
• More capable models benefit more from Verbalized Sampling, showing an emergent scaling trendclaim0.804
  Empirical observation that larger models (GPT-4.1, Gemini-2.5-Pro) show 1.5-2x greater diversity gains from VS compared to smaller models
• Verbalized Sampling (VS)framework0.794
  The paper's proposed training-free prompting strategy that prompts the model to verbalize a probability distribution over a set of responses rather than generating a single response
• Nucleus sampling generates more semantically diverse utterances than beam searchclaim0.771
  New semantic diversity dimension added to prior finding that nucleus sampling is more lexically diverse
• Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model.quote0.760
  Opening sentence defining self-evidencing.
• Predictive sequence models in the generative modality are simulators of a learned distribution.claim0.750
  Naming of generative sequence models as simulators.
• Verbalized probabilities for programming languages show only Pearson r=0.182 (GPT-4.1) correlation with corpus frequencies, indicating weaker calibrationfinding0.749
  Limits of verbalized probability calibration when corpus frequency and perceived popularity diverge
• Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing.claim0.748
  Foundational claim of the paper, defining self-evidencing.