Skip to content
The Chat Archive
gpt-3.5-turbo-0314· OpenAI Arena Human Preference 55K

My language model recursively answers itself after I give it an initial prompt. The…

  1. User

    My language model recursively answers itself after I give it an initial prompt. The first response includes ### User: and ### Assistant: . What is likely to be causing this?

  2. Assistant

    It sounds like your language model may be generating responses based on its own previous outputs, which is known as "text hallucination" or "text degeneration". This can happen if the model is not properly constrained or if it has not been trained on enough diverse data to understand the context of the conversation. You may want to try adjusting the parameters of your model or providing it with more diverse training data to help it better understand the context of the conversation. Additionally, you can try using techniques such as beam search or nucleus sampling to generate more diverse and contextually appropriate responses.

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