斯坦福STORM：5视角4提示的博士级AI研究法，让产出条理性提升25%

斯坦福大学OVAL实验室在2024年NAACL会议上发布了一个名为STORM的研究系统，并提供了同行评审的验证：从多角度构建的条目比常规方式生产的文章更有组织的程度高出25%，覆盖范围也广了10%。整个突破的核心就在于——多视角提问能捕捉单一角度的研究盲点。系统开源、免费，但几乎没有人知道，同样的思想可以在Claude中靠四个提示直接复现。

我们正处在AI技术发展的18个月窗口期内，那些有意识运用人工智能做研究的人，会获得远超普通用户的信息深度。差距不在于智商，而在于他们运行了五种视角、一张矛盾地图、一份合成简报和一次同行评审，而其他人只是在读AI给出的单一答案。

方法由四个提示和五个专家视角构成：

  • 四个提示：多视角扫描、矛盾地图、合成简报、同行评审
  • 五个视角：实践者、学术研究者、怀疑论者、经济学者、历史学家

这些提示与视角的应用场景非常广泛：

  • 撰写报告：跑一遍四个提示，你的文章能覆盖别人想不到的角度。
  • 做出商业决策：从业者告诉你现实中什么真正有效，怀疑论者指出可能出错的地方，经济学家揭示利益流向。
  • 面试准备：五分钟从五个角度研究一家公司，从业者视角给你内部语言，怀疑论者启发尖锐问题，你走进面试间时，比房间里的任何人都准备得更充分。
  • 投资研判：你同时在考察牛市论证、熊市论证、历史平行、激励图谱、学术证据；矛盾地图直接展示实际风险所在。
  • 学习新技能：从五个角度画出领域地图，从业者告诉你先学什么，学者提供理论依据，怀疑者挑出被过度炒作的部分，让你直接跳过噪声。
  • 谈判：从五个角度研究对方——动机、弱点和历史行为都会浮现，使你带着结构性优势坐上谈判桌。
  • 演讲：提前对主题做STORM调查，你的幻灯片会在观众提出异议前先予以回答，Q&A环节因此显得轻松自然。

提示词一：多视角扫描，获得五类专家观点

I need to research [YOUR TOPIC].Simulate 5 different expert perspectives on this topic:1. THE PRACTITIONER: works with this daily.What do they know that academics miss?What practical realities are usually ignored?2. THE ACADEMIC: has studied this for years.What does the peer reviewed evidence actually say?Where does the evidence contradict popular belief?3. THE SKEPTIC: thinks the mainstream view is wrong.What is the strongest counterargument?What evidence do proponents conveniently ignore?4. THE ECONOMIST: follows the money.Who profits from the current narrative?What financial incentives shape the research?5. THE HISTORIAN: has seen similar patterns before.What historical parallels exist?What can we learn from how those played out?For each perspective give me:- Their core position in 2 sentences- The strongest evidence supporting their view- The one thing they would tell me that no other perspective would

提示词二：矛盾地图——真正的理解从冲突开始

Based on the 5 perspectives above, map the contradictions:1. Where do two or more perspectives directly contradicteach other? List each conflict with the specific claimsthat clash.2. Which perspective has the strongest evidence?Which has the weakest? Why?3. What is the one question that, if answered, wouldresolve the biggest contradiction?4. What does EVERY perspective agree on?(This is likely true. Even opponents confirm it.)5. What topic did NONE of the perspectives address?(This is the blind spot in the whole field.Often the most valuable finding.)

提示词三：合成研究简报

Synthesize everything from the 5 perspectives and thecontradiction map into a research briefing:1. THE ONE PARAGRAPH SUMMARY: explain this topic as ifbriefing a CEO who has 60 seconds and needs nuance,not just the headline.2. THE 5 KEY FINDINGS: most important things I now know,ranked by reliability. For each, note which perspectivessupport it and which challenge it.3. THE HIDDEN CONNECTION: one non obvious link betweenfindings that only shows up when you look at all 5perspectives together.4. THE ACTIONABLE INSIGHT: based on all the evidence,what should someone in [YOUR ROLE] actually DOdifferently? Be specific.5. THE FRONTIER QUESTION: the one question that, ifanswered, would change everything about how weunderstand this topic.

提示词四：同行评审，确认什么是可靠、什么不可靠

Now peer review your own research briefing:1. CONFIDENCE SCORES: rate each of the 5 key findingson a 1 to 10 scale for reliability. Explain each score.2. WEAKEST LINK: which claim are you least confident in?What specific info would you need to verify it?3. BIAS CHECK: which perspective might be overrepresentedin your synthesis? Did one voice dominate?4. MISSING PERSPECTIVE: is there a 6th angle I shouldhave included that would change the conclusions?5. OVERALL GRADE: if a Stanford professor reviewed thisbriefing, what grade would they give and why?What would they tell me to fix?

你可以选择在与大模型对话时手动逐一粘贴上述提示词，也可以构建一个智能体或独立技能来串联整个流程。