Research map / Large language models / Agents, code and math
Conversation for Non-verifiable Learning: Self-Evolving Large Language Models through Meta-Evaluation
Yuan Sui, Bryan Hooi
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 diagnostic reward (ablation) | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 83.5 -> 87.1 | Improved |
| add diagnostic reward (ablation) | CoNL (Qwen3-8B) · AIME 2025 | Pass@1 (%): 68.2 -> 73.5 | Improved |
| add diagnostic reward (ablation) | CoNL (Qwen3-8B) · DeepMath | Rank-rho: 0.55 -> 0.78 | Improved |
| add blind ranking (ablation) | CoNL (Qwen3-8B) · DeepMath | Rank-rho: 0.45 -> 0.78 | Improved |
| add blind ranking (ablation) | CoNL (Qwen3-8B) · AIME 2025 | Rank-rho: 0.39 -> 0.65 | Improved |
| add blind ranking (ablation) | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 82.5 -> 87.1 | Improved |
| add consensus reward (ablation) | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 85.2 -> 87.1 | Improved |
| add solution quality reward (ablation) | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 84.8 -> 87.1 | Improved |
| increase agents from N=2 to N=4 (ablation) | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 84.5 -> 87.1 | Improved |
| increase agents from N=5 to N=8 (ablation) | CoNL (Qwen3-8B) · AIME 2025 | Rank-rho: 0.66 -> 0.63 | Hurt |
| replace Self-Rewarding multi-agent with CoNL diagnostic reward | CoNL (Qwen3-8B) · AIME24 | Pass@1 (%): 69.8 -> 76.5 | Improved |
| replace Multi-Agent Debate inference with CoNL training | CoNL (Qwen3-8B) · DeepMath | Pass@1 (%): 70.8 -> 87.1 | Improved |
Closest papers by meaning
- The Lighthouse of Language: Enhancing LLM Agents via Critique-Guided Improvement (NeurIPS 2025 · similarity 0.90)
- ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate (ICLR 2024 · similarity 0.90)
- Improving Factuality and Reasoning in Language Models through Multiagent Debate (ICML 2024 · similarity 0.90)
- Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMs (ICML 2024 · similarity 0.90)
- \$OneMillion-Bench: How Far are Language Agents from Human Experts? (NeurIPS 2026 · similarity 0.89)
- Self-Challenging Language Model Agents (NeurIPS 2025 · similarity 0.88)
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.
