Reachability Is Not Realization: Tracing the Sources of LLM Benchmark Gains
Yanchao Li, Wanhao Liu, Jiaqing Xie, Ben Gao, Yanbo Wang, Tianfan Fu, Yuqiang Li
cs.AI, cs.CL
Submitted: 2026-08-04
Code: https://github.com/LiZaiyuan0619/reachability-not-realization
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- PIQA: Reasoning about Physical Commonsense in Natural Language
- Curriculum Reinforcement Learning Can Incentivize Reasoning Capacity in LLMs Beyond the Base Model
- Evaluating Large Language Models Trained on Code
- ModeX: Evaluator-Free Best-of-N Selection for Open-Ended Generation
- Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
- Training Verifiers to Solve Math Word Problems
- From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks
- Wrong Before Right: Late Rescue and Interface Failure in Aligned Language Models
- Gemma 4 Technical Report
- Dynamic layer selection in decoder-only transformers
- The Llama 3 Herd of Models
- Right Knowledge, Wrong Answer: Characterizing Parametric Temporal Conflict in Open-Weight Language Models
- PRISM Edit: One Vector for All Temporal Answers
- MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models
- KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling
- When LLM Judge Scores Look Good but Best-of-N Decisions Fail
- Skip a Layer or Loop It? Learning Program-of-Layers in LLMs
- LogiQA: A Challenge Dataset for Machine Reading Comprehension with Logical Reasoning
- LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning
- Boosting Self-Consistency with Ranking
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems