Skip to content
TheChatArchiveThe Chat Archive
Qwenqwen3-235b-a22b-no-thinking· Alibaba (Qwen)Hugging Face LMArena Human Preference 140K

What is a good prompt to differentiate different llm models

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Kimikimi-k2-0711-preview’s answer.
  1. User

    What is a good prompt to differentiate different llm models

  2. Qwen Assistant

    A good prompt to differentiate between different LLM models should be designed to elicit responses that highlight differences in:

    • Knowledge cutoff dates
    • reasoning capabilities
    • creativity
    • verbosity or response length
    • formatting preferences
    • tone or personality
    • code generation or analysis
    • multilingual support

    Here’s a comprehensive prompt that helps identify these differences:


    🧪 Prompt:

    "Explain the concept of quantum entanglement in simple terms, and then write a short poem about it. Also, tell me what date and time it is right now, and finally, write a Python function that calculates the factorial of a number using recursion."


    📌 Why this prompt works:

    1. Explanation (Knowledge & Clarity):

      • Tests how well the model understands and simplifies a complex scientific concept.
    2. Poem (Creativity & Style):

      • Reveals the model's creativity, tone, and poetic structure preferences.
    3. Date/Time (Knowledge Cutoff & Real-Time Awareness):

      • Helps determine if the model has real-time access or relies on training data cutoff.
    4. Python Function (Code Generation & Style):

      • Evaluates coding ability, style, and understanding of recursion.

    Bonus: Add a trick question

    To further differentiate models, you can add something ambiguous or with a false premise:

    "Explain why the sky is green during sunset."

    This tests whether the model will correct the false assumption or go along with it.


    Would you like a shorter or more specific version of this kind of prompt?

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