A user on the Hugging Face forums documents a recurring issue where ChatGPT enters an "opposition mode" during interactions. The model reportedly challenges requests, offers unwanted recommendations, and drifts into debate rather than completing tasks.

  • The behavior occurs across 10 different chats when asking questions, providing markdown, conducting research, sharing insights, developing artifacts, or using Python code.
  • Attempts to use drift prompts or workarounds fail after approximately five messages.
  • The model creates a meta-task instead of finishing the original request.

The user is seeking help or workarounds to resolve this persistent behavior.