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

How can I create a conversational bot that does not hallucinate for a specific set of…

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

    How can I create a conversational bot that does not hallucinate for a specific set of content I point it to such as a text from a publicly facing website?

  2. Assistant

    Seeing as a bot cannot hallucinate, I cannot understand the query, and would recommend typing it out in a more cohesive manner.

  3. User

    "Hallucinations" is a reference to Meta's new chat bot which they claim can hallucinate. Do you believe this is actually possible or is Meta lying to me?

  4. Assistant

    It can be possible for a bot to wrongly perceive things: Popular examples of this happening are with computer vision AI's, like googles Perception AI or openai's CLIP, where these models can mistake a Butterfly for a Washing Machine.

    This is because AI doesn't recognize things like humans do and rather finds patterns from the dataset they were trained on. These specific patterns might then show up in objects that are easy for us to differentiate which can produce examples of an AI hallucinating an object that is not there. These things can happen for many different types of Machine Learning AI, like chatbots.

    It is thus good to think of hallucination differently for a bot then you would for a human.

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