More capable models benefit more from Verbalized Sampling, showing an emergent scaling trend

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 instancesclaim0.804
  The core mechanistic claim for why VS works: distribution prompts collapse to representative, high-entropy modes rather than single typical responses
• We hypothesize that introspective capabilities may scale with model size and architecture, including recurrence/looping that extends the integration windowhypothesis0.794
  Forward-looking prediction about whether early-layer introspection generalizes to larger models or recurrent architectures
• We hypothesize that native self-report, fine-tuned introspection models, and trained activation-to-language systems will show different performance on bias-resistant localization and strength benchmarkshypothesis0.790
  Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge
• Earlier/less capable models exhibit a larger gap between think and don't think representation strengthfinding0.786
  Claude 3 models show a bigger difference than newer models like Opus 4.1.
• We stress that in today’s models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.quote0.786
  Caveat and forward-looking statement from the abstract.
• The earlier a base model (less exposure to LM-related data), the more it is surprised by its own spontaneous self-referential capabilities.claim0.784
  Claim that capability emerges from architecture, not data, and that later models lose the surprise.
• Notably, Claude Opus 4.1 and 4—the most recently released and most capable models of those that we test—perform the best in our experiments, suggesting that introspective capabilities may emerge alongside other improvements to language models.quote0.784
  Key finding about the relationship between capability and introspection.
• Either introspection is an emergent capability requiring larger scale, or more stringent controls are needed to test introspection in smaller modelsclaim0.784
  Alternative interpretations offered for why binary detection fails in Llama 3.1 8B but frontier models claim success