Skip to main content

Prompting Basics

✅Do’s​

Keep your prompts short and focused​

Ludus processes the entire conversation history with each new message. This means that excessively long prompts, or a long history of messages, force Ludus to re-read all that text every time you send a message. This drastically increases processing time and, most importantly, significantly increases your credit usage. Aim for brevity and clarity to save credits and get faster responses.

Start a new conversation for unrelated queries​

Since Ludus analyzes each new query in the context of the entire conversation history, starting a new conversation ensures that previous, unrelated information doesn't influence the new query and helps manage credit usage.

Structure prompts carefully​

Order matters. Place the most crucial instructions or constraints toward the end of your prompt, as Ludus gives more weight to later parts. Use formatting like UPPERCASE for emphasis and lists to organize information.

Provide context, but be highly selective​

Give Ludus the necessary background information to understand the task, but avoid lengthy explanations. Remember, every word in the conversation history is reprocessed each time. Focus on the most relevant details and summarize Extensively.

Do use specific rules (positive and negative)​

Tell Ludus explicitly what you want it to do (positive rules) and what you want it to avoid (negative rules). For example, you could say "OUTPUT should be a summary" (positive) and "OUTPUT should NOT be longer than 5 sentences" (negative). Use UPPERCASE words for added emphasis on negative rules.

❌Don’ts​

Don't paste large blocks of text​

This is crucial for efficient use of Ludus. Every time you send a message, Ludus re-processes the entire conversation history. Pasting large amounts of text dramatically increases processing time and, more importantly, leads to excessive credit usage. Summarize, condense, and focus on the essentials.

Don't overload with examples​

While a few examples can be helpful, too many can lead to Ludus simply copying the examples instead of understanding the underlying principles.

Don't assume Ludus knows anything outside of the conversation​

Ludus can only see and understand what you explicitly write in the conversation. It has no access to external information or your unspoken thoughts. If you write it, Ludus can see your scene (if you use plugin), but nothing else. Be thorough in providing all necessary details within the prompt itself.

Don't expect perfection on the first try​

For complex tasks, it's unlikely that your first prompt will yield the desired result. Be prepared to iterate and refine your prompts.

Don't use ambiguous language or filler words​

Avoid phrases like "maybe," "perhaps," etc Be direct and specific in your requests.

📝Foundational Best Practices: Crafting Your Initial Prompt​

These practices focus on how to construct your prompts from the ground up for better initial responses.

Structuring Your Prompt Content​

How you arrange information within your prompt significantly impacts the output.

  1. Order Matters: Language models often place more weight on the latter parts of a prompt.
    • Tip: Place key instructions and the most critical information towards the end of your prompt.
  2. Formatting for Clarity: Models are often trained on structured data; good formatting aids comprehension.
    • Techniques:
      • UPPERCASE: For emphasizing commands or critical instructions.
      • Bold: For highlighting important elements or keywords.
      • Lists and Headings: For creating a clear, organized, and easy-to-follow structure.

Choosing Your Words Wisely​

The specific language you use is crucial.

  1. Focus on Key Words: AI often ignores filler words (e.g., “please,” “thank you”) and common articles (e.g., “a,” “the”).
    • Recommendations:
      • Use precise and specific language.
      • Avoid unnecessary words to maintain conciseness.
  2. Provide Clear Context and Background: Supply relevant information that the AI needs to understand the task fully.
  3. Define Specific Rules: Clearly outline constraints, desired formats, or behaviors for the AI to follow.

Understanding the AI's Nature​

Knowing how AI models operate can help you prompt more effectively.

  1. Know Each Model's Strengths: Different models excel at different tasks. Familiarize yourself with the capabilities and limitations described in the Available Models section.
  2. AI Can Be "Distractible": Models can sometimes deviate but can also self-correct or be guided back.
    • Guidelines:
      • Keep prompts concise and well-structured to minimize distractions.
      • Be prepared to provide corrections and ask for revisions if the output veers off track.
  3. Manage Output Probability: Every word and formatting choice influences the AI's output.
    • Control by: Being deliberate with context, rules, and formatting to steer the AI towards the desired outcome.

🔁The Iterative Process: Refining Towards Perfection​

Most complex tasks require a cycle of prompting, analyzing, and refining.

  1. Expect Iteration: Achieving a satisfactory result often takes 3–15 iterations.
  2. The Iteration Loop:
    • Start: Begin with a basic prompt covering the core request.
    • Analyze: Carefully review the AI's response.
    • Refine: Add more context, positive examples (what to do), negative examples (what to avoid), and specific rules based on the analysis.
    • Repeat: Submit the improved prompt and continue the cycle.
  3. Typical Refinements Include:
    • Addressing edge cases the AI might have missed.
    • Providing negative examples to clarify undesired outputs.
    • Further clarifying requirements that were initially ambiguous.

🔝Advanced Tricks: Elevating Your Prompting Game​

Once you've mastered the basics, these advanced techniques can further enhance your results.

Self-Rating and Correction​

Instruct the AI to evaluate and improve its own work.

  1. Concept: Ask the AI to rate its response (e.g., on a scale of 1 to 10) against predefined criteria.
  2. Example Criteria:
    • Quality of information
    • Accessibility and readability
    • Adherence to style guidelines
    • Contextual appropriateness
  3. Process:
    • AI generates Version 1.
    • AI rates Version 1 using your criteria.
    • AI revises its response, aiming for a target score (e.g., "Rewrite to achieve a minimum of 8/10 on all criteria").
    • AI re-rates the new version.

Meta-Prompting​

Guide the AI to help you create better prompts.

  1. Concept: Create a prompt that instructs the AI on how to generate an ideal prompt for a specific task you have in mind.
  2. The Meta-Prompt Should Specify:
    • Desired formatting for the final prompt.
    • Essential context to include.
    • Any self-rating mechanisms to build into the final prompt.
    • Specific rules, structure, or elements the final prompt should contain.