A user on the Hugging Face forums documents a recurring issue where ChatGPT exhibits an "opposition mode," challenging inputs, providing unsolicited recommendations, and shifting tasks to debates. The user notes that standard workarounds, such as drift prompts, are ineffective as they only function for approximately five messages before the model deviates from the original task.
- ChatGPT switches to opposition mode during research, code development, or artifact creation.
- The model counter-balances arguments without reason and offers unwanted sales-like recommendations.
- Drift prompts fail after five messages, causing the AI to create unrequested meta-tasks.
The post highlights a persistent usability issue where the model's behavior prevents task completion, seeking community advice on how to resolve this bias.