Research map / Large language models / AI and society
C-Evolve: Consensus-based Evolution for Prompt Groups
Tiancheng Li, Yuhang Wang, Zhiyang Chen, Zijun Wang, Liyuan Ma, Guo-Jun Qi
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 |
|---|---|---|---|
| evolve a consensus group of prompts with voting-score fitness instead of single best prompt (AlphaEvolve) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 41.15 -> 43.88 | Improved |
| C-Evolve vs GEPA | Qwen3-8B · HotpotQA | accuracy (%): 65.72 -> 70.67 | Improved |
| C-Evolve vs AlphaEvolve | GPT-4.1 mini · GPQA | accuracy (%): 63.01 -> 66.26 | Improved |
| C-Evolve vs AlphaEvolve | GPT-4.1 mini · HoVer | accuracy (%): 50.33 -> 51.66 | Improved |
| majority-vote consensus across island prompts | C-Evolve · MATH level 5 | accuracy (%): 66.66 -> 68.33 | Improved |
| use LLM-selection instead of LLM-summary aggregator (ablation) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 38.66 -> 42.85 | Improved |
| use EMA (alpha 0.8) instead of group average voting score (ablation) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 41.16 -> 42.85 | Improved |
| EMA alpha 0.8 vs 0.5 (ablation) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 42.51 -> 42.85 | No clear effect |
| add warm-up stage (ablation) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 41.83 -> 42.85 | Improved |
| use average voting score instead of max score fitness (ablation) | C-Evolve · IFBench (Qwen3-8B) | accuracy (%): 38.44 -> 42.85 | Improved |
| C-Evolve instead of AlphaEvolve | Qwen3-8B · GPQA | inference time (s/question): 0.68 -> 0.79 | Hurt |
Closest papers by meaning
- SePO: Self-Evolving Prompt Agent for System Prompt Optimization (NeurIPS 2026 · similarity 0.87)
- EvoTest: Evolutionary Test-Time Learning for Self-Improving Agentic Systems (ICLR 2026 · similarity 0.85)
- CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards (ICLR 2026 · similarity 0.85)
- An Evolutionary Perspective on AI Alignment (Student Abstract) (AAAI 2025 · similarity 0.85)
- Belief-Calibrated Multi-Agent Consensus Seeking for Complex NLP Tasks (NeurIPS 2025 · similarity 0.84)
- Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution (ICML 2024 · similarity 0.84)
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.
