Verbalized Sampling (VS)

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

• Verbalized Sampling recovers pre-trained generative diversity by prompting models to verbalize distributions rather than single instancesclaim0.794
  The core mechanistic claim for why VS works: distribution prompts collapse to representative, high-entropy modes rather than single typical responses
• Verbalized Sampling improves diversity without compromising factual accuracy or safety alignmentclaim0.781
  Claims verified by commonsense reasoning and safety evaluation experiments showing VS maintains >97% refusal rates and comparable factual accuracy
• Rejection samplingmethod0.754
  A technique to filter model outputs; Redwood Research's project mentioned.
• Sampled-decoding self-reportmethod0.745
  Temperature=0.8 sampled decoding for self-report; reduces collapse moderately but remains discrete and noisy
• Verbalized Evaluation Awarenessconcept0.744
  When the model explicitly mentions being tested in its chain-of-thought reasoning; distinguished from behavioral evaluation awareness.
• VS-Multimethod0.741
  A VS variant that generates k responses with probabilities across multiple conversation turns for additional diversity
• Thompson Samplingmethod0.741
  A Bayesian exploration strategy that samples from the posterior distribution over model parameters to decide actions.
• VS-Standardmethod0.737
  The baseline variant of Verbalized Sampling that asks for k responses with their probabilities in a single LLM call