Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective

TL;DR

This paper introduces a novel prompt design paradigm where random, seemingly incoherent demonstrations can outperform traditional prompts, and proposes an evolutionary framework, PromptQuine, to automatically discover effective prompts in low-data regimes.

Contribution

It challenges conventional prompt design by showing incoherent prompts can be highly effective and introduces PromptQuine, an evolutionary search method for automatic prompt optimization.

Findings

Incoherent prompts can match or outperform traditional prompts.

PromptQuine effectively discovers prompts across diverse tasks.

The approach works with limited data and improves LLM performance.

Abstract

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.