Reproduce “A Survey of Automatic Prompt Engineering: An Optimization Perspective” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenwu Li, Xiangfeng Wang, Wenhao Li, Bo Jin
This survey from Li et al. introduces the first unified optimization-theoretic framework for automated prompt engineering, encompassing the full landscape of prompt optimization across text, vision, and multimodal domains. It formalizes prompt optimization as a mathematical problem, systematically categorizing prompt variables, objective functions, and optimization methods.
Key insights
- Automated prompt engineering can be formalized as a mathematical optimization problem, unifying diverse techniques across modalities.
- Prompt variables can be systematically categorized into discrete (instructions, exemplars, annotations), continuous (soft prompts), and mixed types, each with distinct optimization characteristics.
- Optimization methods for prompts are broadly classified into FM-based, evolutionary computing, gradient-based, and reinforcement learning approaches, each suited for different prompt variable types and objectives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.