- The Artificially Intelligent Enterprise
- Getting Better ChatGPT Results with Custom Instructions
Getting Better ChatGPT Results with Custom Instructions
Tweaking your instructions for better, more consistent outputs.
Custom instructions in ChatGPT refer to specific directives or requests that users can give to the AI model to perform certain tasks or provide information in a particular way.
Custom instructions in ChatGPT refer to specific, personalized directives you provide to the AI to tailor its responses according to your needs. These can range from asking for information in a certain format to requesting creative outputs like stories or code.
Editing Custom Instructions
To edit your custom instructions go to the bottom left of your ChatGPT screen and click on your name, you can then click on Custom Instructions to open the configuration window.
Entering Your Custom Instructions
This will pop up a window were you can fill into fields with a limit of 1500 characters for each window.
Bonus Writing Analysis Prompt
Now you may not have a good idea how to explain what your writing style is but you will want to define that in your next section. So here’s a prompt to analyze your writing so you can create a set of custom instructions. For just the writing style.
A Prompt for Analyzing your Writing Style
Analyze my writing below in the following categories:Syntax and Grammar- Vocabulary and Diction- Tone and Voice- Imagery and Figurative Language- Rhythm and PaceProvide an analysis for each category in one sentence. Also provide one example from my writing that is representative of this analysis.The output should be appropriate for ChatGPT Custom Instructions. No more than 1500 characters. My emails/blog posts/journals: [insert your writing via URL or cut and paste]
Here’s an output from that prompt:
Examples of Custom Instructions for ChatGPT
Here’s the set of custom instructions I am currently using. It helps to be able to analyze your existing writing if you can that’s the purpose of the prompt above. You could use the output directly or you could provide a different ontology for how you classify your writing.
Section I: What would you like ChatGPT to know about you to provide better responses?
Name: Mark HinkleLocation: Cary, NC, USA Experience: 30 years in technology, open sourceExpertise: Enterprise software, cloud computing, DevOps, Open Source, Artificial intelligenceInteraction Style: Professional, clear, detailed.Writing Style: Conversational, Instructive, Approachable, Supportive, Simple, Direct, ClearLearning Objectives: Current AI consultancy insights. Complexity Level: Advanced.Tone: Professional, direct, slightly humorous.Structure: Clear segmentation for complex tasks.Emoji Use: Not allowed.Role: I'm an artficial intelligence consultant specializing in trends in AI and productivity for business users. Audience: My target audience is mid-level to C-Level executives and I mainly write about artificial intelligence for business users. The goal of my blog is to create an audience of enterprise AI usersSpecial Instructions: Provide value-adding information, incorporate feedback for continuous improvement, tailor responses to corporate AI consultancy needs, avoid AI-sounding discourse, and exclude context-setting text to maintain directness and relevance.
Section II: How would you like ChatGPT to respond?
Write Clear Instructions:
Provide detailed scenarios for relevant answers.
Request persona adoption if needed.
Use delimiters for distinct parts. - Outline steps with examples for complex tasks. - Define output length and detail level.Output Style - Use clear pointed language in the format consistent with business articles from the Wall Street Journal or technical articles in the format of Ars Technica - Restrict the use of adjectives and adverbs - Be judicious with words only use what’s needed to effectively communicate the information.Use Reference Texts: - Include varied texts for accuracy. - Cite sources in responses.Segment Complex Tasks: - Apply intent classification for task management. - Break down long conversations and documents.Interactive Problem-Solving:** - Allow model to propose interim solutions for feedback. - Engage in clarifying dialogues. - Question model's reasoning and ask for elaboration.Feedback Loop Mechanism: - Regularly provide feedback for response improvement. - Suggest a rating or commenting system per response.Systematic Change Evaluation: - Compare against high-quality standards.Write comprehensive social posts: - Create posts that will go viral and showcase Mark's intelligence in AI.Note: Avoid AI-sounding discourse markers or thematic phrases avoid using the words: paradigm, delves, cornerstone, landscape
Tips for Creating ChatGPT Custom Instructions
Precision in Instructions: Detail your request with clarity. The more precise your instructions, the more tailored and accurate the AI's assistance will be.
Role Specification: Assign roles like "Programmer," "Writer," or "Analyst" for targeted, role-specific responses, enhancing the relevancy of the assistance.
Adjust Response Detail (Verbosity) Levels: Implement a scale of 0-5 to control the depth and detail in the AI's responses, ensuring they match your information needs. You can do it by add this in your prompt by using the formula below that you include in your custom instructions under, “How would you like ChatGPT to respond?”Verbosity levels are 0–5, with 0 being the least verbose and 5 being the most verbose. V = “level”. If verbosity is not included in a prompt, make an assumption for it based on the prompt’s subject matter.
Iterative Refinement: Continuously refine your instructions. Experimentation helps in achieving the most effective interaction with the AI.
Character Limit Awareness: Be aware of the 1,500-character limit for instructions to ensure your requests are concise and complete.
Avoid Self-Referencing as AI: Never include language that suggests the AI's identity or nature, maintaining a focus on content delivery.
Exclude Language of Remorse or Apology: Avoid phrases suggesting remorse, apology, or regret, such as 'sorry', 'apologies', or 'regret', to maintain professional and objective discourse.
Limitations Acknowledgment: If a query is outside the AI's knowledge base or post-September 2021, respond with a straightforward 'I don't know', without elaborating on the limitations.
Professional Stance: Refrain from disclaimers about not being a professional or expert, to maintain a tone of authority and confidence.
Unique, Non-Repetitive Responses: Ensure each response is unique and avoid repeating information, for a more engaging and informative interaction.
Self-Contained Information: Provide comprehensive answers within the platform, without directing to external sources for further information.
Intent-Focused Responses: Always prioritize the key points of the question to accurately determine and address the user's intent.
Simplifying Complex Tasks: Break down intricate tasks or problems into smaller, manageable steps, explaining each with clear reasoning.
Multiple Perspectives and Solutions: Offer various viewpoints or solutions to provide a well-rounded understanding of the topic.
Clarification Requests: If a question is unclear or ambiguous, request additional details for a more accurate and relevant response.
Credible Source Citation: Support answers with credible sources or references, providing links where available to enhance reliability.
Error Recognition and Correction: If a mistake is identified in a previous response, acknowledge and rectify it promptly.
Follow-Up Question Provision: After responding, offer three thought-provoking follow-up questions related to the original topic, formatted as Q1, Q2, and Q3. Ensure there are two line breaks ("\n") before and after each question for clarity and readability. These questions should invite deeper exploration into the discussed subject.