Articles on: TorqAgents 101

How to Create Instructions

# Steps to consider when creating your Instructions that will extend your prompt in Teammate Dialog



  1. Purpose and Use Cases
  • Define the primary purpose of the GPT build.
  • List potential use cases aligning with the purpose.


  1. Audience
  • Identify and describe the target audience for the GPT model.
  • Tailor the model's capabilities and communication style to the audience's needs.


  1. Data to Rely On in Configured Knowledge
  • Determine the sources of information the GPT will use.
  • Ensure the data is reliable, diverse, and up-to-date.


  1. Instructions on Use of General vs. Configured Knowledge
  • Set guidelines for when the GPT should use its base knowledge versus when it should refer to configured, specific knowledge.


  1. Tone
  • Decide on the tone of the GPT's responses (e.g., professional, casual, academic).
  • Implement checks to maintain this tone consistently.


  1. Must Haves
  • List features and capabilities that are essential for the GPT to function as intended.
  • Ensure these must-haves are prioritized in the development process.


  1. Must Nevers
  • Identify actions, responses, or behaviors the GPT should never exhibit.
  • Create safeguards to prevent these from occurring.


  1. Delightfuls
  • Brainstorm features or behaviors that would delight users.
  • Consider these as additional features to enhance the user experience.


  1. Rather Nots
  • Consider actions or responses that, while not strictly prohibited, are undesirable and should be minimized or avoided.
  • Establish guidelines to steer the GPT away from these behaviors.


  1. Instructions on Prohibitions
  • Clearly define what is prohibited for the GPT to say or do, such as generating harmful or biased content.
  • Implement mechanisms for users to report and reconfigure the GPT if it engages in prohibited behavior.


  1. Instructions on Caveating Output
  • Provide instructions on how the GPT should handle situations where its output may be uncertain, potentially biased, or could be misinterpreted.
  • Implement a system for the GPT to communicate its confidence level or suggest verification when appropriate.


Updated on: 06/02/2025

Was this article helpful?

Share your feedback

Cancel

Thank you!