Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity

Documenti analoghi

• Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace
  di: Yu, Simon, et al.
  Pubblicazione: (2026)
• Eliciting Uncertainty in Chain-of-Thought to Mitigate Bias against Forecasting Harmful User Behaviors
  di: Sicilia, Anthony, et al.
  Pubblicazione: (2024)
• Annotations Mitigate Post-Training Mode Collapse
  di: Springer, Jacob Mitchell, et al.
  Pubblicazione: (2026)
• Active Learning for Robust and Representative LLM Generation in Safety-Critical Scenarios
  di: Hassan, Sabit, et al.
  Pubblicazione: (2024)
• ReCOGS: How Incidental Details of a Logical Form Overshadow an Evaluation of Semantic Interpretation
  di: Wu, Zhengxuan, et al.
  Pubblicazione: (2023)