From Candidate Tokens to Final Outcomes: Decoding Relative Long-Horizon Candidate Value from Reasoning Model Representations
cs.CL
Submitted: 2026-10-11
Updated: 2026-10-11
Sources
- Big-Math: A Large-Scale, High-Quality Math Dataset for Reinforcement Learning in Language Models
- Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought
- Evaluating Large Language Models Trained on Code
- The Cancellation Hypothesis in Critic-Free RL: From Outcome Rewards to Token Credits
- Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Unleashing Implicit Rewards: Prefix-Value Learning for Distribution-Level Optimization
- Finding Neurons in a Haystack: Case Studies with Sparse Probing
- Cliff Tokens: Analyzing Failure Trigger Tokens in LLM Mathematical Reasoning
- Beyond the Target: From Imitation to Collaboration in Speculative Decoding
- FutureBridge: Token Selection Beyond Local Preference in Collaborative Decoding
- Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capability
- LLMs Encode Their Failures: Predicting Success from Pre-Generation Activations
- Towards Long-Horizon Interpretability: Efficient and Faithful Multi-Token Attribution for Reasoning LLMs
- Qwen3 Technical Report
- Contrastive Branch Policy Optimization
- Sparse Reward Subsystem in Large Language Models
- Efficient and Trainable Language Model Test-Time Scaling via Local Branch Routing
- Hidden Error Awareness in Chain-of-Thought Reasoning: The Signal Is Diagnostic, Not Causal
- Chain-of-Thought Tokens are Computer Program Variables
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering