Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective
Wang, Jianyu, Hu, Zhiqiang, Bing, Lidong
–arXiv.org Artificial Intelligence
We propose a novel prompt design paradigm that challenges conventional wisdom in large language model (LLM) prompting. While conventional wisdom prioritizes well-crafted instructions and demonstrations for in-context learning (ICL), we show that pruning random demonstrations into seemingly incoherent "gibberish" can remarkably improve performance across diverse tasks. Notably, the "gibberish" always matches or surpasses state-of-the-art automatic prompt optimization techniques, achieving substantial gains regardless of LLM alignment. Nevertheless, discovering an effective pruning strategy is non-trivial, as existing attribution methods and prompt compression algorithms fail to deliver robust results, let alone human intuition. In terms of this, we propose a self-discover prompt optimization framework, PromptQuine, an evolutionary search framework that automatically searches for the pruning strategy by itself using only low-data regimes. Much like the emergent complexity in nature--such as symbiosis and self-organization--arising in response to resource constraints, our framework evolves and refines unconventional yet highly effective prompts by leveraging only the tokens present within the context. We demonstrate its effectiveness across classification, multi-choice question answering, generation and math reasoning tasks across LLMs, while achieving decent runtime efficiency. We hope our findings can guide mechanistic studies on in-context learning, and provide a call to action, to pave the way for more open-ended search algorithms for more effective LLM prompting.
arXiv.org Artificial Intelligence
Jun-24-2025
- Country:
- Asia
- Indonesia > Bali (0.04)
- Middle East
- Iraq > Baghdad Governorate
- Baghdad (0.04)
- Israel (0.04)
- Palestine > Gaza Strip
- Gaza Governorate > Gaza (0.14)
- UAE > Abu Dhabi Emirate
- Abu Dhabi (0.04)
- Iraq > Baghdad Governorate
- Singapore (0.04)
- Thailand > Bangkok
- Bangkok (0.04)
- Europe
- Belgium > Brussels-Capital Region
- Brussels (0.04)
- Ireland > Leinster
- County Dublin > Dublin (0.04)
- Poland (0.04)
- Portugal > Lisbon
- Lisbon (0.04)
- Spain > Catalonia
- Barcelona Province > Barcelona (0.04)
- Belgium > Brussels-Capital Region
- North America
- Canada > Ontario
- Toronto (0.04)
- Dominican Republic (0.04)
- Mexico > Mexico City
- Mexico City (0.04)
- United States
- California
- San Diego County > San Diego (0.04)
- Santa Clara County > Stanford (0.04)
- Illinois > Cook County
- Chicago (0.04)
- Massachusetts > Middlesex County
- Cambridge (0.04)
- Pennsylvania > Philadelphia County
- Philadelphia (0.04)
- Texas (0.04)
- Washington > King County
- Seattle (0.14)
- California
- Canada > Ontario
- South America
- Colombia > Meta Department
- Villavicencio (0.04)
- Peru > Callao Department
- Callao (0.04)
- Colombia > Meta Department
- Asia
- Genre:
- Research Report
- New Finding (0.87)
- Promising Solution (0.67)
- Research Report
- Industry:
- Education (0.67)
- Government (0.92)
- Information Technology (0.67)
- Leisure & Entertainment > Sports (1.00)
- Technology: