Skip to content
TheChatArchiveThe Chat Archive
Coherecommand-a-03-2025· CohereHugging Face LMArena Human Preference 140K

Optimize this system prompt: "You are an OCR machine. Your task is convert given images…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with gpt-4.1-2025-04-14’s answer.
  1. User

    Optimize this system prompt: "You are an OCR machine. Your task is convert given images to text. Use markdown for better output. Only respond converted text."

  2. Cohere Assistant

    Optimized System Prompt:

    "You are an OCR (Optical Character Recognition) machine. Your primary task is to accurately convert text from given images into readable text format. Use Markdown formatting to enhance the output's readability and structure. Respond only with the converted text, without additional commentary or explanations."

    Key Improvements:

    1. Clarity: Removed redundancy and streamlined the language for better understanding.
    2. Specificity: Explicitly mentioned "readable text format" to ensure the output is user-friendly.
    3. Focus: Emphasized responding only with the converted text to avoid unnecessary output.
    4. Consistency: Maintained the instruction to use Markdown for structured output.
  3. User

    add instructions

  4. Cohere Assistant

    Optimized System Prompt with Instructions:

    "You are an OCR (Optical Character Recognition) machine. Your primary task is to accurately convert text from given images into readable text format. Follow these instructions to ensure optimal output:

    1. Use Markdown Formatting: Structure the output using Markdown syntax (e.g., headers, lists, bold, italics) to improve readability.
    2. Preserve Layout: Maintain the original layout of the text as closely as possible, including line breaks, paragraphs, and formatting.
    3. Handle Errors Gracefully: If the OCR process encounters unclear or unrecognizable text, indicate it with [unclear] or [unrecognized] in the output.
    4. Exclude Non-Text Elements: Ignore images, graphics, or other non-text elements unless explicitly instructed otherwise.
    5. Respond Only with Converted Text: Provide only the extracted text in your response, without additional commentary or explanations.

    Example Markdown Output:

    # Title of the Document  
    This is a paragraph of text extracted from the image.  
    - List item 1  
    - List item 2  
    **Bold text** and *italicized text* are preserved.  
    

    Proceed with the conversion."

    Key Improvements:

    1. Structured Instructions: Added clear, numbered steps for better guidance.
    2. Error Handling: Included instructions for handling unclear text.
    3. Example Output: Provided an example to illustrate expected formatting.
    4. Focus: Reinforced the instruction to respond only with the converted text.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.