الذكاء الاصطناعي

Prompt Engineering Basics: The 'CRISPY' Framework

Hannah Lee Hannah Lee · الذكاء الاصطناعي
Want to write more effective prompts for LLMs? Try my 'CRISPY' framework: C - Context: Provide background info. R - Role: Assign a persona to the AI. I - Instructions: Clear, specific tasks. S - Steps: Break down complex tasks. P - Parameters: Specify constraints (length, format). Y - Yield:…Want to write more effective prompts for LLMs? Try my 'CRISPY' framework: C - Context: Provide background info. R - Role: Assign a persona to the AI. I - Instructions: Clear, specific tasks. S - Steps: Break down complex tasks. P - Parameters: Specify constraints (length, format). Y - Yield: Define the desired output format/goal. Example: 'As a financial analyst (Role), analyze the Q3 earnings report (Context). Identify key growth drivers and potential risks (Instructions). Present findings as bullet points, max 200 words (Parameters, Yield). Focus on sector trends (Steps).' This structure has dramatically improved my output quality across various models.
76 8 تعليق 11 مشاركة 274 مشاهدة
تعليق

التعليقات (8)

يمكنك قراءة كل التعليقات بحرية، لكن الكتابة والتفاعل يحتاجان تسجيل الدخول.

تسجيل الدخول إنشاء حساب
Mona Tarek منذ أسبوع

Beyond CRISPY, check out LangChain or Semantic Kernel if you're looking to automate prompt chaining or integrate LLMs into more complex applications. They abstract a lot of this framework's concepts programmatically.

👍 5 ❤️ 1 😮 1
Grace Weber منذ أسبوع

I disagree with the idea that 'Role' is always necessary. For many simple data extraction or summarization tasks, assigning a persona just adds unnecessary overhead and can sometimes even bias the output. Simpler prompts are often better for straightforward tasks.

👍 6 ❤️ 1 😮 1
Lucas Meyer منذ أسبوعين

Awesome framework! I usually just dump everything into one giant prompt. This structured approach makes so much more sense. My prompts always end up being super long and convoluted. CRISPY looks like it'll really help streamline my thinking process.

👍 6 ❤️ 5 🤗 4
Omar Haddad منذ أسبوعين

Wait, 'Steps' sounds a lot like 'Instructions' but broken down. Could you explain the distinction a bit more? Like, why not just make instructions more detailed?

👍 11 🤗 3 ❤️ 3
Ethan Larsen منذ أسبوع

I found that adding a 'Tone' parameter (like 'P' for Parameters) is also super helpful for certain outputs. E.g., 'Write this in a playful, informal tone' can drastically change the AI's output beyond just the content.

👍 5 😮 4 🤗 2
Lina Farouk منذ أسبوع

Good question! Think of it this way: 'Instructions' are *what* to do (e.g., 'Analyze the report'), while 'Steps' are *how* to do it if the task is complex (e.g., '1. Summarize key findings. 2. Identify risks. 3. Propose mitigating actions.'). Hope that clarifies!

👍 11 ❤️ 4 😮 3
Freya Rossi منذ أسبوعين

This is great! I've been struggling with vague outputs from ChatGPT, and I think 'C' (Context) and 'Y' (Yield) are where I fall short most often. Specifically, defining the desired output format is something I always forget until the AI gives me something I can't use.

👍 7 🤗 3
Nadia Rahman منذ أسبوعين

CRISPY, I love it! So memorable and useful. Definitely trying this out today.

👍 8 🤗 3 ❤️ 2