Providing information to Chatbots To Generate More Accurate Outputs

Link to Original Document

The goal for the following scenario is to generate custom CSS code that can be used to enhance the visual design of an exported PDF guide from the application Folge.

• Language Model Used:

  
  Claude 3.5 Sonnet*
  

• Platform:

  
  https://claude.ai/chat + *Bolt AI (Using Custom Command & System Instruction)
  

• Prompt Used:

  
  

  create custom CSS code to make a “how-to” guide more modern and engaging that will be exported as a “pdf” using the app “Folge”

• Examples:

  • Example 1:

    
    No additional context is given to the model alongside the prompt.
    

  • Example 2:

    
    Content from Folge Documentation is copied directly from website, and pasted into chat alongside prompt
    

  • Example 3:

    
    Content from Folge Documentation is copied, and then fed into a custom prompt & system instruction using Bolt AI , to reformat the content to be better understood by language models, and then is pasted into chat alongside prompt.
    

Example 1

No context CSS.mp4

Final Output (Images resized to 60%)

No Context CSS.pdf

No CSS Theme Applied | Default Export.pdf

Example 2

Basic Context.mp4

Final Output (Images resized to 60%)

Basic Context.pdf

Example 3

---

With Converted Context.mp4

Final Output (Images resized to 60%)

With converted Context.pdf

The third example generates the most accurate and in-depth output result because it is specifically designed to optimize content for AI processing to optimize the models ability to ingest new information or context.

By using an AI tool to preprocess and reformat the information, the content is structured in a way that is highly effective for large language model interpretation. This approach involves organizing the content into distinct sections, employing consistent Markdown formatting, removing unnecessary information if present, and highlighting key concepts in a manner that aligns with AI processing patterns and token optimization.

Essentially, this method translates the content from a “human-friendly” to an “AI-friendly” format, significantly enhancing the language model’s ability to comprehend and utilize the information effectively. The result is a marked improvement in output quality, accuracy, and depth compared to less optimized methods.

---

Document Conversion Prompt:

Prompt has been modified with some parts redacted for security purposes.

# Task: Optimize Text for Markdown

## Objective:
Convert input text into a clear, concise Markdown document.

## Instructions:

1. Analyze content within the set of three back ticks at the end of the prompt and plan structure using a chain-of-thought approach after fully understanding the provided content
2. Organize using Markdown headings and formatting
3. Remove redundancy and focus on key information
4. Use concise language to reduce length and optimize tokens
5. Review for accuracy and clarity before generating the final output content

## Output:

1. Brief summary of approach
2. Optimized Markdown document
3. List of uncertainties or assumptions (if any)
4. Alternative structures considered (optional)

# Content to analyze:
``` HighlightThisTextAndPasteContentToReplace

“`