你不再需要提示工程：提示反转
You Don't Need Prompt Engineering Anymore: The Prompting Inversion
AI总结：
本研究提出Sculpting提示方法，发现其在中端模型gpt-4o上优于标准CoT，但在gpt-5上效果相反，揭示提示策略需随模型能力共同演进。
摘要：
提示工程，特别是思维链（CoT）提示，显著增强了大型语言模型的推理能力。我们引入了“Sculpting”，一种受约束的、基于规则的提示方法，旨在通过减少语义歧义和有缺陷的常识导致的错误来改进标准CoT。我们使用GSM8K数学推理基准（1,317个问题），在三个OpenAI模型代际（gpt-4o-mini、gpt-4o、gpt-5）上评估了三种提示策略（零样本、标准CoT和Sculpting）。我们的发现揭示了一种“提示反转”：Sculpting在gpt-4o上提供了优势（97%对比标准CoT的93%），但在gpt-5上变得有害（完整基准上94.00%对比CoT的96.36%）。我们将此追溯到“护栏到手铐”的转变，其中防止中端模型常识错误的约束在先进模型中诱发了过度字面理解。我们详细的错误分析表明，最优提示策略必须与模型能力共同演进，提示更强大的模型应使用更简单的提示。
英文摘要：
Prompt engineering, particularly Chain-of-Thought (CoT) prompting, significantly enhances LLM reasoning capabilities. We introduce "Sculpting," a constrained, rule-based prompting method designed to improve upon standard CoT by reducing errors from semantic ambiguity and flawed common sense. We evaluate three prompting strategies (Zero Shot, standard CoT, and Sculpting) across three OpenAI model generations (gpt-4o-mini, gpt-4o, gpt-5) using the GSM8K mathematical reasoning benchmark (1,317 problems). Our findings reveal a "Prompting Inversion": Sculpting provides advantages on gpt-4o (97% vs. 93% for standard CoT), but becomes detrimental on gpt-5 (94.00% vs. 96.36% for CoT on full benchmark). We trace this to a "Guardrail-to-Handcuff" transition where constraints preventing common-sense errors in mid-tier models induce hyper-literalism in advanced models. Our detailed error analysis demonstrates that optimal prompting strategies must co-evolve with model capabilities, suggesting simpler prompts for more capable models.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
