Skip to content
TheChatArchiveThe Chat Archive
Qwenqwen3-30b-a3b· Alibaba (Qwen)Hugging Face LMArena Human Preference 140K

Outside of obviously illegal or ethically questionable things, what shouldn't I do with…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.5-flash’s answer.
  1. User

    Outside of obviously illegal or ethically questionable things, what shouldn't I do with a large language model? My initial thought went to "install one in a giant mech", so you can see where my headspace is at right now.

  2. Qwen Assistant

    Your imagination is spot on—there’s a lot of fun and creativity in thinking about wild applications for LLMs! But even outside of illegal or ethically questionable uses, there are practical, safety, and responsibility-related reasons to avoid certain things. Here’s a breakdown of what shouldn’t be done, with a mix of caution, humor, and practicality:


    1. Use It for Critical Decision-Making Without Human Oversight

    • Why? LLMs are language models, not experts. They can hallucinate, make up facts, or fail to understand context.
    • Examples:
      • Trusting an LLM to diagnose medical conditions, design bridges, or make legal decisions.
      • Using it to control autonomous vehicles, drones, or robots without rigorous testing.
    • The Mech Angle: If you’re planning to power a giant mech with an LLM, make sure it’s supervised by a human (or at least a failsafe system). A model might not understand physics, safety protocols, or the concept of "don’t shoot your own team."

    2. Generate Content That Feeds Misinformation or Harm

    • Why? Even if it’s not illegal, spreading false or harmful information (e.g., fake news, conspiracy theories, or manipulative content) can erode trust, cause panic, or harm individuals.
    • Examples:
      • Writing fake reviews, phishing emails, or deepfakes.
      • Crafting persuasive arguments for unethical behavior (e.g., encouraging self-harm or dangerous experiments).
    • The Mech Angle: If your mech is designed to "persuade" people to do something (e.g., join a cult, buy a product), it’s a slippery slope. LLMs aren’t here to manipulate—use them to inform, not coerce.

    3. Overload It with Sensitive or Private Data

    • Why? LLMs can inadvertently memorize or leak data they’re trained on. Even if you’re not storing data, sharing private information (e.g., personal emails, financial records) with a model risks exposure.
    • Examples:
      • Asking the model to "analyze" your medical history or confidential business plans.
      • Using it to process user data without proper encryption or consent.
    • The Mech Angle: If your mech is collecting data from the real world (e.g., sensors, cameras), don’t let the LLM process it without safeguards. Privacy is still a big deal, even in a robot suit.

    4. Ignore the Environmental Impact

    • Why? Training and running large models consumes massive amounts of energy. Using them inefficiently (e.g., for trivial tasks) contributes to carbon footprints.
    • Examples:
      • Using an LLM to generate a single paragraph when a simple rule-based system would suffice.
      • Deploying a model in a resource-heavy application (e.g., a mech that’s always on and processing data).
    • The Mech Angle: A giant mech powered by an LLM might need a nuclear reactor to run. Consider the energy costs before building a robot that outshines a power plant.

    5. Treat It as a "Magic Box" for All Problems

    • Why? LLMs are tools, not miracles. They can’t solve every problem, especially those requiring creativity, empathy, or physical interaction.
    • Examples:
      • Expecting the model to design a novel scientific experiment or write a novel without human input.
      • Using it to replace human labor in roles that require nuanced judgment (e.g., counseling, art, or diplomacy).
    • The Mech Angle: A mech might be cool, but it’s not a replacement for human ingenuity. Use the LLM to augment your abilities, not replace them.

    6. Neglect to Audit or Test for Biases

    • Why? LLMs inherit biases from their training data. If you don’t check for them, you could perpetuate harmful stereotypes or unfair outcomes.
    • Examples:
      • Using the model to generate job descriptions that favor certain demographics.
      • Deploying it in a system that disproportionately targets marginalized groups.
    • The Mech Angle: If your mech is interacting with people, make sure it’s not reinforcing stereotypes or making assumptions about gender, race, or culture.

    7. Use It for Tasks That Require Real-Time Physical Interaction

    • Why? LLMs are reactive, not proactive. They can’t sense the physical world or react to it in real time.
    • Examples:
      • Trying to use the model to control a robot’s movements or respond to sensor data.
      • Relying on it to "think" while a mech is in motion (e.g., dodging obstacles).
    • The Mech Angle: Your mech might need a separate system for real-time decision-making (like a traditional AI or hardware controller). The LLM can handle high-level planning, but not the nitty-gritty.

    8. Ignore the "Black Box" Problem

    • Why? LLMs are often opaque. If something goes wrong, it’s hard to trace why.
    • Examples:
      • Deploying the model in a critical system (e.g., healthcare, finance) without understanding how it arrives at conclusions.
      • Using it to make decisions in a way that’s unexplainable or unaccountable.
    • The Mech Angle: If your mech does something unexpected (e.g., attacks a friend), you’ll need to debug the model’s logic. Good luck with that.

    Final Thought:

    Your giant mech idea is amazing—but remember, the LLM is just one piece of the puzzle. Think of it as a brain for your mech, not a replacement for engineering, safety protocols, or human oversight. If you’re building a mech, pair the LLM with sensors, actuators, and a human in the loop. That way, you’re not just creating a robot—it’s a team.

    And if you ever need help with the actual engineering, I’m here for that too. Just don’t let the mech read this and decide it’s time to take over the world. 😄

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