The Paradox of Outcome Optimization: A Causal Information-Theoretic Bound on Reasoning Shortcuts in LLMs

Documenti analoghi

• Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?
  di: Chen, Zihan, et al.
  Pubblicazione: (2025)
• RankCLIP: Ranking-Consistent Language-Image Pretraining
  di: Zhang, Yiming, et al.
  Pubblicazione: (2024)
• Information-Theoretic Causal Bounds under Unmeasured Confounding
  di: Jung, Yonghan, et al.
  Pubblicazione: (2026)
• Inductive Subgraphs as Shortcuts: Causal Disentanglement for Heterophilic Graph Learning
  di: Wang, Xiangmeng, et al.
  Pubblicazione: (2026)
• Meaningful Causal Aggregation and Paradoxical Confounding
  di: Zhu, Yuchen, et al.
  Pubblicazione: (2023)