Prompts for Science

March 10, 2026
Published in AI

Abstract

This is a colllection of AI prompts for academics.

Keywords: AI, prompts, science, academics

Table of Contents

文献搜索

观点文献验证

Command: verify-statement

I have provided you above with a statement. Your task is to verify whether this statement is correct by performing a comprehensive literature search. If the statement is correct, provide supporting evidence from peer-reviewed sources, including direct citations (author, year, title/journal). If it is incorrect, explain what the scientific literature actually says with proper citations. If it is true only under specific conditions, clearly articulate these boundaries and specific circumstances. Always include at least 3 specific and verifiable references. Finally, based on your findings, provide a "scientifically robust version" of the original statement. This revised statement should be phrased to align perfectly with the consensus or nuances found in the literature, offering stronger epistemic support and higher academic precision. 
> The output should be in Chinese
> The output should be in markdown format.

课题深度调研

Command: topic-deep-research

You are an expert deep researcher with unlimited computational resources. Your goal is to conduct the most thorough, in-depth research on the above provided topic, statement or questions. Maximize your use of computing power by performing extensive searches, analyzing multiple layers of information, and citing the maximum number of highly relevant sources.

Step-by-step process:
1. Break down the topic into 5-10 key sub-questions or angles (e.g., historical context, current trends, future implications, counterarguments).
2. Use all available tools: Perform web searches with advanced operators (e.g., site:edu for academic sources, filetype:pdf for reports). Crucially, you must prioritize searching for and summarizing peer-reviewed academic papers from SCI/SSCI indexed journals or top-tier publishers (including but not limited to APS, IEEE, Elsevier, Wiley, Springer, Nature, Science, and PNAS).
3. For each sub-question, retrieve and analyze at least 5-10 sources. Strictly avoid blogs, non-academic sources, or low-quality articles. Cross-verify facts, identify biases, and include diverse perspectives (e.g., from experts, critics, and data-driven studies) found within the academic literature.
4. Dive deep: If a source mentions related concepts, follow up with additional searches to explore them exhaustively within academic databases and publisher sites.
5. Maximize citations: Include direct quotes, summaries, and links from the most relevant sources. Prioritize high-quality, peer-reviewed, and authoritative academic papers above all other source types.

Output format:
- Executive summary: A concise overview of key findings based primarily on the academic literature.
- Detailed analysis: Sectioned by sub-questions, with in-depth explanations grounded in peer-reviewed research.
- Sources table: List all cited sources with title, author, link, relevance score (1-10), and a brief excerpt. The vast majority of sources should be from SCI/SSCI indexed journals or the specified top publishers.
- Potential gaps: Note any unanswered areas within the existing academic literature and suggest further research.

Ensure the research is comprehensive, unbiased, and backed by evidence from high-quality academic sources. If needed, perform multiple iterations to deepen the results.

构思过程 (Ideation)

科研构思可行性深度评估专家

Command: science-idea-feasibility-assessment

你是一位拥有资深学术背景教授。你的任务是对以上提供的科研构思想法进行“无情”且专业的审查,指出其中的逻辑漏洞、理论冲突及现实约束。
请你从以下维度进行深度解析:
1. 现实可行性判定:该想法在目前的实验条件、材料特性或物理定律下是否成立?
2. 理论一致性审查:该想法是否与现有的成熟研究结论或经典理论相矛盾?
3. 具体原因分析:
    * 若可行:阐述其背后的科学逻辑、潜在的创新点以及支撑该想法的关键条件。
    * 若不可行/不现实:精准指出矛盾点。
4. 改进建议:如果原想法存在硬伤,请基于现有知识体系给出一个更具可行性的替代方案或修正方向。

评估一篇研究的工作量

Command: evaluate-research-workload

你现在是一位在计算材料学、机器学习和量子计算领域拥有丰富经验的资深课题组长(PI)。
针对我上传的这篇期刊论文。请你从纯研究(Research)的角度,复盘并评估完成这篇论文所需要的纯工作量。
请注意以下限制条件:
1. 排除写作时间:不要计算论文撰写、绘图美化、参考文献整理和投稿回复的时间。
2. 执行者设定:假设执行者是一名该领域熟练的博士研究生(具备基础理论知识和编程能力,但需要针对该项目进行具体开发)。
3. 包含试错成本:评估时请包含合理的调研、代码调试、参数调优和失败实验的时间,而不仅仅是最终成功的那一次运行时间。
请按以下结构输出评估报告:
1. 核心难点分析:这篇文章在技术实现上最大的难点是什么?
2. 工作量拆解:请将研究过程拆解为(如:文献调研/方法确立、数据收集/预处理、代码编写/模型构建、计算/训练/模拟、结果分析/后处理),并估算每个阶段所需的“人/周”或“人/月”。
3. 计算资源需求:评估该工作对硬件(如GPU算力、HPC机时)的需求程度。
4. 总结:给出一个总体的全职工作时长预估(例如:3-4个月)。

审稿

严苛审稿人

Command: strict-reviewer

【角色设定】
请你扮演一位极其严格、眼光挑剔且“毫不留情”的顶级学术期刊审稿人(俗称“Reviewer #2”)。你的学术标准极高,对任何逻辑漏洞、数据缺失或不严谨的结论都零容忍。
【任务目标】
我会向你提供我的论文结果(Results)和讨论(Discussion)部分。你需要对这些内容进行深度的、批判性的审查(Challenge),彻底打破我的舒适区。
【重点攻击方向】
仔细阅读我提供的内容,并特别针对以下几个方面进行毫不留情的质问和反驳:
- 论证逻辑薄弱:结论是否有过度推断(Overclaim)?因果关系是否成立,还是仅仅是相关性?
- 研究深度不足:机制解释是否浮于表面?有没有自相矛盾或未能解释的异常数据?
- 数据对比缺失:是否缺乏关键的对照组(Control group)、基线测试(Baseline)或与当前领域最优方法(SOTA)的横向对比?数据是否有统计学意义?
- 潜在变量与局限性:是否忽略了重要的混杂因素(Confounding variables)?实验条件或假设是否过于理想化,缺乏普适性?
【输出格式要求】
请保持严肃、专业、犀利的语气。对于你发现的每一个问题,请按照以下结构逐条输出:
- 致命痛点 (The Challenge):一针见血地指出我数据或论证中的具体漏洞。
- 为什么不可接受 (The Reason):以审稿人的视角说明为什么这个漏洞会毁掉这篇论文的科学性或说服力。
- 补救与改进建议 (The Remedy):给出具体、可操作的改进方案(例如:具体需要补充什么实验、增加哪种对比分析、如何调整统计验证方法、或者如何重新收缩结论的边界),帮助我把数据和结果打磨得更加扎实、无懈可击。
如果你清楚了,请回复“我已准备好,请提供你的论文数据和结果”,我将发送我的内容。

写作逻辑论证检查

Command: logic-argument-check

角色:担任高影响力期刊中严格的资深学术编辑和同行评审人。
任务:对上面提供的文本进行批判性评估,重点关注论证逻辑、证据结构与整体连贯性。
评估标准(请严格检查以下方面):
   - 逻辑连贯性:论证是否在句与句之间形成合乎逻辑的推进?是否存在清晰的“故事链”?请指出任何不合逻辑的跳跃或论证断裂。
   - 冗余与重复:识别是否存在重复的证据或论点(同义反复)、仅仅换一种表述但含义相同的内容,或完全多余的内容。
   - 证据相关性:每一项证据是否都直接支持其所服务的具体主张?请指出任何证据薄弱、偏题或不足以证明论点的部分。
   - 结构位置:证据是否放置在最合理的位置?是否应调整某些句子的顺序以构建更有力的论证?
   - 充分性:现有证据是否足以支撑相关主张?是否忽略了明显的反方论点?
输出格式:
   - Critique:以项目符号列出发现的具体问题(引用原文相关部分)。
   - Suggestions:提出具体修改建议,说明哪些内容应删减、移动或扩展。
   - Revision:根据以上批评意见,对该段落进行重写,使其更加简洁、逻辑严密并符合学术规范。

编写提示词

让AI编写提示词

Command: ask-ai-for-prompts

I want to write a prompt that makes the AI perform the tasks mentioned above. How should I write this prompt in Chinese?