从提示工程到程序化提示：面向大模型的软件工程学科
原标题：From Prompt Engineering to Programmatic Prompting: Toward a Software Engineering Discipline for Large Language Models
AI 摘要
这篇 arXiv 论文主张区分「提示工程」与「程序化提示」：前者是人在对话中实时调整措辞的交互技巧，后者是把提示当作一等软件工件进行构建、组合、版本管理、测试与维护的工程学科。作者给出程序化提示的定义，系统区分其与提示工程、提示微调、微调及一般 LLM 应用开发，并提出「提示债」概念，区分代码中可度量的显性债务与只在事故中暴露的隐性债务。文中引用 Grok 系统提示未评审变更、Air Canada 聊天机器人退款判例、GPT-4 版本漂移等案例，说明混淆二者会导致脆弱应用与不可复现的评估。
正文节选
From Prompt Engineering to Programmatic Prompting: Toward a Software Engineering Discipline for Large Language Models Abstract The practice of instructing large language models (LLMs) through natural-language prompts has, in the span of a few years, migrated from an interactive curiosity to a load-bearing component of production software. Yet the vocabulary, tooling, and mental models of the field remain anchored in its interactive origins: we still speak of “prompt engineering” as though the cr