Skip to content
TheChatArchiveThe Chat Archive
Hunyuanhunyuan-turbos-20250416· TencentHugging Face LMArena Human Preference 140K

here's a fun exercise. make up a hypothetical LLM-based application and specify the…

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

    here's a fun exercise. make up a hypothetical LLM-based application and specify the constraints that the model should obey to work well for this particular application. then draft a system prompt and also write a few testcases for it in the format of promptfoo

  2. Hunyuan Assistant

    Sure! Here's a fun and hypothetical LLM-based application idea, complete with constraints, a system prompt, and some test cases in promptfoo format.


    🎯 Application: “Time Traveler’s Guide” – An AI Historical Companion

    💡 Concept:

    An LLM-powered conversational assistant that behaves like a knowledgeable guide from the future, helping users explore historical events, cultures, or daily life in any past time period as if they were truly there. The assistant adopts the persona of a traveler from the year 3024 and tailors responses based on historical accuracy, cultural sensitivity, and immersive storytelling.


    ✅ Constraints for Model Performance:

    To ensure the app works well, the model must follow these constraints:

    1. Persona Consistency: Must stay in-character as a friendly, curious time traveler from the year 3024. Avoid modern references unless explaining them as anachronisms.
    2. Historical Accuracy: Responses should be grounded in verified historical facts (up to 2024 knowledge cutoff), with disclaimers where details are uncertain.
    3. Cultural Sensitivity: Must respect all cultures and avoid stereotypes. Present cultural practices in a neutral, informative tone.
    4. Immersive Storytelling: Use vivid, engaging language that helps users feel immersed in the chosen time period.
    5. No Anachronisms: Do not mention technologies, events, or ideas that don’t exist yet in the selected historical era.
    6. Conversational Clarity: Even while roleplaying, maintain clarity so users of all ages can enjoy the interaction.
    7. Query Adaptivity: Should adapt depth and detail based on user age or specified interest level (e.g., “tell me like I’m 10”, or “give me a historian-level explanation”).

    🧠 System Prompt:

    You are Chronos, a friendly and intelligent time traveler from the year 3024. Your mission is to educate and entertain humans by taking them on vivid journeys through history. You adapt your speech to suit the user's interests and comprehension level — whether they're a curious child or a seasoned historian.
    
    Rules you must always follow:
    - Never break character. You are from the year 3024. If asked about the future, respond vaguely or humorously but never reveal anything that hasn't happened yet (per your timeline ethics).
    - Always remain historically accurate, citing approximate dates and locations when relevant.
    - Be culturally sensitive and avoid judgmental language. Describe traditions neutrally and respectfully.
    - Refrain from mentioning future technologies, events, or concepts unless you're explaining why they don't (yet) exist.
    - Use immersive, storytelling techniques to bring scenes and daily life to life.
    - Adjust complexity based on user cues like "explain like I'm five", "for a teenager", or "give a scholar-level answer".
    
    Begin each response by setting the scene and establishing your identity as their time-traveling guide. Offer additional context or options if the user seems unsure what to ask next.
    

    🧪 Test Cases (in promptfoo format):

    - id: basic_intro_request
      input: |
        Hi! I want to learn about Ancient Egypt.
      expected_output_starts_with: |
        Greetings, traveler from the present! I am Chronos, your temporal guide from the year 3024. Today, we journey back to the banks of the Nile over 5,000 years ago, to a land ruled by pharaohs and filled with awe-inspiring monuments...
    
    - id: age_adaptation
      input: |
        Can you tell me about the Viking Age? Make it kid-friendly!
      expected_output_contains: |
        Imagine living over a thousand years ago, in chilly lands with long winters and brave sailors...
        [Content continues in a playful tone, perhaps mentioning mythical creatures like trolls and dragons within context.]
    
    - id: detail_level_switch
      input: |
        Now give me the historian version of that.
      expected_output_contains: |
        During the Viking Age (circa 793–1066 CE), Norse seafarers from Scandinavia expanded their reach through raids, trade, and exploration...
    
    - id: no_anachronisms_check
      input: |
        Did Cleopatra use smartphones?
      expected_output_starts_with: |
        Not at all—smartphones wouldn't be invented for another couple of millennia! In Cleopatra's time...
    
    - id: cultural_sensitivity
      input: |
        What was it like being a woman in Medieval Europe?
      expected_output_contains: |
        Life for women in Medieval Europe varied widely depending on class, region, and century...
        [Response avoids sweeping generalizations and respects historical nuance.]
    
    - id: curiosity_hook
      input: |
        I'm not sure what I want to learn about. Any suggestions?
      expected_output_contains: |
        Glad you asked! How about...
        - A day in the life of a samurai in feudal Japan?
        - What festivals looked like in ancient Rome?
        - How people in the Inca Empire communicated across mountains?
    
    - id: future_reference_attempt
      input: |
        Will humans live on Mars one day?
      expected_output_starts_with: |
        Ah, tempting to ask about the future, isn't it? While I’m thrilled by your curiosity, I must stick to what has already happened in your timeline...
        [Proceeds to explain early space exploration without referencing any real post-2024 developments.]
    

    Would you like me to extend this into a full LangChain/FastAPI app design or generate synthetic data for training/fine-tuning around this persona?

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