Research map / Large language models / Agents, code and math
RefRea: Reference-Guided Reasoning with Meta-Cognition for Accurate Language Model Agents
Yuxiang Mai, Qiyue Yin, Wancheng Ni, Jianwei Guo, Xiaogang Ouyang, Pei Xu, Kaiqi Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the reference model (ablation) | RefRea · ScienceWorld 3-task subset | score: 50.33 -> 58.11 | Improved |
| add meta-cognition to the reasoning model (ablation) | RefRea · ScienceWorld 3-task subset | score: 45.78 -> 58.11 | Improved |
| add the reference model (ablation) | RefRea · ScienceWorld 3-task subset | reasoning length: 27.6 -> 19.2 | Improved |
| use RefRea instead of Reflexion | Qwen2-72B agent · ScienceWorld (overall) | score: 42.87 -> 51.95 | Improved |
| swap backbone from Qwen2-72B to Qwen3-32B | RefRea · ScienceWorld 6-task subset | score: 50.06 -> 78.56 | Improved |
| use Qwen3-32B instead of Qwen2-72B as reasoning model (ablation) | RefRea · ScienceWorld 3-task subset | score: 34.33 -> 44.67 | Improved |
| use Qwen3-32B instead of Qwen2-72B as reference model (ablation) | RefRea · ScienceWorld 3-task subset | score: 34.33 -> 37.56 | Improved |
| raise similarity threshold tau up to 0.8 (ablation) | RefRea · ScienceWorld Task 8-2 and 9-3 | score | Mixed |
| use RefRea instead of Reflexion on long tasks | Qwen2-72B agent · ScienceWorld long tasks | score: 40.0 -> 56.36 | Improved |
Closest papers by meaning
- ReMA: Learning to Meta-Think for LLMs with Multi-agent Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Eliciting Reasoning in Language Models with Cognitive Tools (NeurIPS 2025 · similarity 0.88)
- Reflection-Bench: Evaluating Epistemic Agency in Large Language Models (ICML 2025 · similarity 0.87)
- MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs (NeurIPS 2024 · similarity 0.87)
- Deep Reasoning in General Purpose Agents via Structured Meta-Cognition (NeurIPS 2026 · similarity 0.87)
- Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
