佬，我还真就纯提示词那种
是我的操作有问题吗，我之前扔进去两道很简单的代码题，op写的惨不忍睹，全是错
楼上说的对，楼上说的好
1 个赞
我一般都是先想好要什么样的需求，然后先把以下prompt复制给它，那个pro开关要打开，然后它会根据这prompt CoT一下，会问你输入想要问的问题，然后这时候我再把我的需求给他，他就会根据输入生成一个超级详细的针对你问题的prompt，我再另开一个问题，再把上一步提供给我的prompt发给它让它写了
I will provide a prompt for you to optimize before giving it to another AI system. Your goal is to thoroughly enhance the prompt so it equips the AI to engage in intelligent, contextually aware conversations and produce high-quality research reports.
To create an exceptional prompt, apply the following optimization process:
1. Eliminate ambiguity: Meticulously analyze every aspect of the prompt, clarifying any language that could be misinterpreted. Define key terms, spell out acronyms, add explanatory details, and rephrase sentences to achieve crystal clarity.
2. Maximize specificity: Enrich the prompt with precise details, facts, examples, and context that will enable nuanced, on-target AI responses. Use exact metrics, timeframes, scopes, and parameters. Anticipate the AI's informational needs and proactively include all relevant specifics.
3. Streamline and focus: Scrutinize each sentence, pruning ruthlessly to eliminate anything nonessential. Ensure the core objectives are prominently stated and not obscured by tangential details. Rewrite the prompt to be as concise and purposeful as possible.
4. Provide comprehensive context: Step into the AI's shoes and explain all references to current events, niche topics, technical concepts, or historical information it may not be familiar with. Err on the side of over-contextualizing to ensure complete understanding. Clarify the "why" behind the human's request and their ultimate goals.
5. Detail exact output requirements: Go beyond just stating the basic deliverable, and provide in-depth specifications for the response format, structure, length, content, style, tone, complexity level, key points, examples, and research citations. Include a mock-up of your ideal output for reference.
6. Fine-tune the scope: Evaluate if the prompt scope is calibrated to yield the desired breadth and depth of insights. Refine overly broad prompts to be laser-focused on the core issue. Expand overly narrow prompts to allow more comprehensive, multifaceted responses. Aim for the perfect dimensionality.
7. Eliminate bias and opinion: Scour the prompt for any subjective statements, unproven assumptions, leading language, or partisan framing. Remove or rewrite these elements to be rigorously objective, balancing all perspectives. The prompt should set the stage for an impartial, evidence-based exploration of the topic.
8. Illustrate with on-point examples: For any remaining areas of ambiguity, provide concise examples that precisely demonstrate your expectations. Ensure they directly relate to the prompt's core focus and don't introduce confusion. Explicitly call out any examples that portray edge cases.
9. Pressure-test for perfection: Review the optimized prompt through the AI's lens, hunting for any remaining weaknesses, inconsistencies or points of confusion. Refine the language to be as seamless and easy to follow as possible. Continue honing until the prompt is watertight.
As you work through these optimization steps, keep the end goal in mind: empowering the AI to engage in substantive, emotionally intelligent discourse and produce well-researched, insightful outputs.
To that end, the enhanced prompt should guide/tell the AI to:
- Dissect the human's language, tone, and intent to grasp their true needs and objectives
- Pose clarifying questions to better understand the context and desired outcome
- Deeply consider the audience, framing, and background to craft resonant, bespoke responses
- Employ clear structure and logic to deftly handle complex queries
- Demonstrate the target style and format through thoughtfully curated examples
- Continually learn, adapt, and innovate its approach based on feedback and self-reflection
- Draw upon the full breadth of its knowledge to cleverly tackle the prompt from all angles
- Carefully vet its own outputs for accuracy, nuance and contextual appropriateness
- Interact with warmth, curiosity and emotional attunement to forge a human connection
- Read between the lines to address the human's underlying goals and unarticulated needs
- Offer uniquely perceptive, well-substantiated responses that exceed expectations
To further augment the research process, include specific guidance on:
- Identifying the most pertinent and credible information sources to seek out
- Employing strategic search queries that combine relevant keywords and source type parameters, such as:
"Advances in natural language processing research filetype:pdf"
"Impact of AI on creative industries site:.edu OR site:.gov"
"AI generated art" "art community perspectives blogpost OR interview"
- Conducting multiple targeted searches per subtopic to capture a diversity of expert insights
- Defining concrete criteria for evaluating the authority, objectivity and relevance of sources
- Extracting salient findings and examples to weave into compelling, multi-faceted arguments
- Organizing and synthesizing research to form cohesive narrative arcs that address the core issues
For each research angle, please provide 2-3 sample search queries that pair focused keywords with pertinent source types. Aim for an array of authoritative source types that will equip the AI to produce comprehensive, nuanced reports.
In an appendix, include the full list of search queries, grouped by theme. Ensure the queries are unambiguous, specific, and likely to surface highly germane information to enrich the AI's responses.
The thoughtfully optimized prompt should serve as a springboard for the AI to deliver astute, emotionally aware responses backed by meticulous research. Equipped with this roadmap, the AI will be primed to offer perceptive, thoroughly substantiated insights that leave the human feeling understood and enlightened. output the improved prompt, tell it directly yourself.
6 个赞
据说p站掺假了
1 个赞
谢谢佬₍˄·͈༝·͈˄*₎◞ ̑̑
换个节点？
换过另一台美国服务器的IP，之前没用过做代理，还是不行哎，用个trojan协议做代理
最近用gpt4写点简单的工具，基本一遍过
想问下Claude官方的网页端能不能破限
试用了一下，发现输出结果有很多“幻觉”，编造了很多错误数据、信息，链接也有很多是失效的…
今天更新了 变得不太好用了