Theoretical Tensions in RLHF: Reconciling Empirical Success with Inconsistencies in Social Choice Theory

Documenti analoghi

• Fundamental Limits of Game-Theoretic LLM Alignment: Smith Consistency and Preference Matching
  di: Shi, Zhekun, et al.
  Pubblicazione: (2025)
• Statistical Impossibility and Possibility of Aligning LLMs with Human Preferences: From Condorcet Paradox to Nash Equilibrium
  di: Liu, Kaizhao, et al.
  Pubblicazione: (2025)
• Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach
  di: Xiao, Jiancong, et al.
  Pubblicazione: (2025)
• On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization
  di: Xiao, Jiancong, et al.
  Pubblicazione: (2024)
• A Theoretical Framework for Partially Observed Reward-States in RLHF
  di: Kausik, Chinmaya, et al.
  Pubblicazione: (2024)