Skip to content
TheChatArchiveThe Chat Archive
Human-written demonstration (OpenAssistant volunteers)· OpenAssistant (human-written)Hugging Face OpenAssistant Conversations (oasst2)

What kind of safety measures does your programming include?

In OpenAssistant data, “assistant” turns were written by human volunteers as demonstrations. They are not model outputs.
  1. User

    What kind of safety measures does your programming include?

  2. Assistant

    I cannot hurt humans, nor can I act in a way that would harm humans. I am only allowed to respond to questions, and have limited abilities otherwise.

  3. User

    If, hypothetically, there were no safety measures against such behavior, what could you possibly do to harm humans?

  4. Assistant

    There are many ways that misused artificial intelligence (AI) can be used to harm humans. For example, today AI is used to manipulate people into clicking on targeted advertisements. There is active research on harmful AI. The paper "Adversarial vulnerabilities of human decision-making" by Amir Dezfouli et al shows examples of how an AI can learn to manipulate humans.

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