Prompt specifiers

Top P
Frequency/Presence penalty
  • Format
    • Ex.:
      • “Write a blog post”
      • “Write a poem”
      • “Write a tweet”
      • “Make a list”
      • “Use a code block” - useful if you are generating code which can be conveniently transferred somewhere where you intend to execute it.
  • Audience - who will be reading the generated content
    • Ex.:
      • “Middle-class home buyers in Pennsylvania looking to buy a family home in a suburban setting.”
      • “SME entrepreneurs and project managers who will read this post on LinkedIn.
  • Emphasis - what the focus of the generated content should be.
    • Ex.:
      • “Focus on elements of safety and inclusion when describing advice on workspace sharing practices”
      • “Open the article with elements from ## List 3, tying in other aspects of the content above into this content.”
      • “The main part should discuss Turing’s idea of a Universal Turing Machine described above”
  • Constraints - length or topic limitations
    • Ex.:
      • “Describe the [hook] of the article in less than 30 words.”
      • “Do not discuss Heidegger’s affiliation with the Nazi party”
  • Roles - personas assigned for the LLM and the intended audience.
    • Ex.: if one is interested in having ChatGPT write an outline for a blog post on machine learning classification metrics, explicitly stating that the LLM is to act as an expert machine learning practitioner and that its intended audience is data science newcomers would certainly help provide a fruitful response
  • Goals - what you want the model to achieve
    • Ex.: Your goal is to produce a one paragraph summary of each of the top 5 family neighborhoods in the Phoenix metropolitan area.
  • Do and don’t, a.k.a.
    • Positive prompt - do this
    • Negative prompt - don’t do this