CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness
cs.CL, cs.AI
Submitted: 2026-10-30
Updated: 2026-10-30
Comments: 17 pages, 3 figures, 14 tables. Project page: https://xxiaoyuuuu.github.io/CRoS-Page
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Analyzing and Improving Chain-of-Thought Monitorability Through Information Theory
- Sabotage Evaluations for Frontier Models
- Reasoning Models Don't Always Say What They Think
- Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
- Gemma 3 Technical Report
- OpenAI o1 System Card
- From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought
- AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Natural Emergent Misalignment from Reward Hacking in Production RL
- The Impossible Test: A 2024 Unsolvable Dataset and A Chance for an AGI Quiz
- Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoning
- Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution
- REASONING GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewards
- Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps
- Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs
- Qwen3 Technical Report
- Outcome Rewards Do Not Guarantee Verifiable or Causally Important Reasoning
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data